Description:DESCRIPTION
BACKGROUND
[0001] Unless otherwise indicated herein, the materials described in this section are not prior art to the claims in this application and are not admitted to being prior art by inclusion in this section.
Field of the invention:
[0002] The subject matter described herein generally relates to industrial machine monitoring systems and, more particularly, to systems and methods for determining machine operational states based on electrical current analysis.
Description of Related Art
[0003] In industrial and manufacturing environments, continuous and reliable monitoring of machine operation is significant for improving productivity, minimizing unplanned downtime, and enhancing overall equipment effectiveness (OEE). Conventional machine monitoring systems typically rely on direct integration with machine controllers, programmable logic controllers (PLCs), or dedicated hardware interfaces. Such approaches often involve substantial installation complexity, increased cost, and may require modifications to existing machine infrastructure, thereby making them difficult to implement, particularly in brownfield or legacy industrial setups.
[0004] In many practical scenarios, legacy machines lack standardized communication protocols or digital interfaces, which limits the ability to acquire accurate and real-time operational data. As a result, existing monitoring solutions may depend on invasive sensor installations, extensive wiring, or manual data collection methods. These approaches not only increase deployment effort and maintenance overhead but also reduce scalability across large manufacturing facilities.
[0005] Certain monitoring techniques employ electrical signal analysis to infer machine behaviour; however, such techniques often rely on fixed threshold values, predefined rules, or require extensive calibration based on prior knowledge of machine characteristics. These approaches may be sensitive to variations in operating conditions, machine types, and load profiles, thereby limiting their adaptability. Consequently, such systems may produce inaccurate, inconsistent, or non-generalizable results when deployed across diverse industrial environments.
[0006] Additionally, conventional systems may not effectively employ edge computing capabilities for real-time data processing, nor do they seamlessly integrate with cloud-based platforms for advanced analytics, large-scale data storage, model training, and remote monitoring. This lack of integration restricts the ability to achieve scalable, flexible, and intelligent monitoring solutions.
[0007] Accordingly, there exists a need for an improved system that enables accurate determination of machine operational states using non-invasive sensing techniques, while supporting scalable deployment, real-time processing, and adaptive learning across varying industrial conditions.
SUMMARY
[0008] In one aspect, the present disclosure discloses a system for monitoring and determining an operational state of a machine is disclosed. The system includes an edge device having at least one processor and a memory storing executable instructions, and a sensor configured to detect a current signal associated with the machine. The processor is configured to sample the current signal to generate current samples, filter the current samples to remove transient spikes, and determine current magnitude values over predefined time intervals. The processor applies a clustering technique to group recurring current patterns into behaviour clusters corresponding to operational states of the machine. The processor further classifies, using a trained machine learning model, an operating condition of the machine into one of a plurality of operating states. The processor enables monitoring of machine operation and identification of operational conditions based on the current signal.
BRIEF DESCRIPTION OF THE DRAWINGS
[0009] Embodiments are illustrated by way of example and not limitation in the figures of the accompanying drawings, in which like references indicate similar elements and in which:
[0010] FIG. 1 illustrates a system 100 for monitoring operational states of a machine 102 based on electrical current analysis, in accordance with an exemplary embodiment;
[0011] FIG. 2 illustrates a block diagram 200 of an edge device 108 and a cloud server 122, which interface with a user interface 116 within the system 100, in accordance with the exemplary embodiment;
[0012] FIG. 3 illustrates system 300 for machine learning-based training and updating of a machine operating state determination model at an edge device 320, in accordance with the exemplary embodiment;
[0013] FIG. 4A illustrates a time-series current waveform mapped with date and time registers over predefined time windows for cycle analysis, in accordance with the exemplary embodiment;
[0014] FIG. 4B illustrates predefined operating state ranges of the machine 122 based on current magnitude thresholds, in accordance with the exemplary embodiment;
[0015] FIG. 4C illustrates mapping of a current waveform to operating state ranges for determining machine states over time, in accordance with the exemplary embodiment;
[0016] FIG. 4D illustrates the identification of idle and running states of the machine 102 based on current variation over time, in accordance with the exemplary embodiment;
[0017] FIG. 5 illustrates clustering of current magnitude values into clusters with corresponding centroids for determining machine operating conditions using K-means clustering, in accordance with the exemplary embodiment;
[0018] FIG. 6 illustrates a method 600 for monitoring operational states of the machine 102 based on electrical current analysis, in accordance with an exemplary embodiment;
[0019] FIG. 7 illustrates a computing environment 700 for implementing systems and methods for monitoring and determining operational states of machines based on electrical current analysis, in accordance with the exemplary embodiment; and
[0020] FIG. 8 illustrates the block diagram 800 of the edge device 108 on which the system 100 is executed, in accordance with the exemplary embodiment.
DETAILED DESCRIPTION
[0021] The following detailed description includes references to the accompanying drawings, which form a part of the detailed description. The drawings show illustrations in accordance with example embodiments. These example embodiments, which may be herein also referred to as “examples” are described in enough detail to enable those skilled in the art to practice the present subject matter. However, it may be apparent to one with ordinary skill in the art, that the present invention may be practiced without these specific details. In other instances, well-known methods, procedures, and components have not been described in detail so as not to unnecessarily obscure aspects of the embodiments. The embodiments can be combined, other embodiments can be utilized, or structural, logical, and design changes can be made without departing from the scope of the claims. The following detailed description is, therefore, not to be taken in a limiting sense, and the scope is defined by the appended claims and their equivalents.
[0022] In this document, the terms “a” or “an” are used, as is common in patent documents, to include one or more than one. In this document, the term “or” is used to refer to a nonexclusive “or” such that “A or B” includes “A but not B,” “B but not A,” and “A and B,” unless otherwise indicated.
[0023] Reference in the specification to “one embodiment” or to “an embodiment” means that a particular feature, structure, or characteristic described in connection with the embodiments is included in at least one embodiment of the invention, and multiple references to “one embodiment” or “an embodiment” should not be understood as necessarily all referring to the same embodiment.
[0024] Referring now to FIG. 1, a system 100 for monitoring and analysing operation of a machine 102 based on electrical current signals is disclosed. The system 100 may be deployed in an industrial environment to enable real-time monitoring, operational state determination, and performance analysis of the machine 102, as known to a person skilled in the art.
[0025] In an embodiment, the machine 102 may represent any electrically operated industrial equipment that draws current during operation. One or more sensors 104 may be operably associated with the machine 102 and may be configured to acquire electrical signals corresponding to the operation of the machine 102 (refer FIG. 4D). The sensors 104 may include, but are not limited to, current transformers, Hall-effect sensors, or clamp-based sensing devices, as known to a person skilled in the art. The sensors 104 may generate a stream of current data indicative of machine behaviour. In an embodiment, the sensor 104 itself may function as an edge device 108, wherein the sensing and processing functionalities may be integrated within a single unit. In such an embodiment, the sensor 104 may comprise one or more processors 204, a current acquisition module 206, an evaluation module 208, a clustering module 210, and a cycle-magnitude engine 212, each of which may perform specific functions associated with the acquisition and processing of current signals, without requiring a separate and distinct processing unit. Accordingly, references to the edge device 108 throughout the present disclosure may, in applicable embodiments, refer to such an integrated sensor-edge device unit.
[0026] In an embodiment, the edge device 108 may be directly connected to the machine 102, for example, by being physically mounted on or attached to the machine 102 or by being coupled through a direct wired connection. In an alternative embodiment, the edge device 108 may be communicatively connected to the machine 102 via a local network 106, which may include, but is not limited to, wired or wireless communication infrastructure such as Ethernet, Wi-Fi, Bluetooth, or industrial communication protocols operating within the vicinity of the machine 102. The local network 106 may enable data transfer between the sensor 104 and the edge device 108 and may further interconnect one or more additional components within the system 100, including but not limited to a mobile device 112, a computing system 114 comprising an interface 116, and a database 110. The manner of connection between the edge device 108 and the machine 102 may be determined based on installation constraints, environmental conditions, or operational requirements, as known to a person skilled in the art.
[0027] In an embodiment, the machine 102 may comprise a multi-phase electrical machine, such as a three-phase electrical machine commonly used in industrial applications. In such configurations, electrical power may be supplied through multiple phase lines, each carrying a portion of the load. The sensor 104 may be operably associated with at least one of the phase lines, for example, by being disposed on a single-phase conductor of the multi-phase supply. The sensor 104 may acquire the stream of current samples corresponding to the electrical current flowing through the selected phase, and the acquired samples may be processed by the edge device 108 using techniques including, but not limited to, filtering, magnitude determination, and state evaluation, as described herein.
[0028] In an embodiment, the edge device 108 may be positioned in proximity to the machine 102 and may function as a local processing unit for real-time analysis of the acquired signals. The edge device 108 may include one or more processors 204 configured to execute instructions for data acquisition, processing, and analysis. The edge device 108 may further include a current acquisition module 206, an evaluation module 208, a clustering module 210, and a cycle-magnitude engine 212, each of which may perform specific functions associated with the processing of current signals. The current acquisition module 206 may be configured to receive and organise incoming current samples from the sensors 104. The cycle-magnitude engine 212 may be primarily configured to identify cycle start boundaries and cycle end boundaries from the current signal, increment a production cycle counter upon detection of each complete production cycle, and record per-cycle parameters including cycle start time, cycle end time, and cycle duration. The evaluation module 208 may analyse the acquired data to validate detected cycles and derive information indicative of machine operation. The clustering module 210 may group processed data into clusters corresponding to different operating behaviours, wherein such grouping may assist in distinguishing valid production cycles from idle periods or anomalous current excursions. These modules may operate in conjunction under the control of the processors 204 to derive the number of completed production cycles and associated operational parameters of the machine 102.
[0029] In an embodiment, one or more user devices, including a mobile device 112 and a computing system 114 comprising an interface 116, may be communicatively coupled to the local network 106. These devices may provide visualisation of machine status, operational metrics, alerts, and historical trends, thereby enabling user interaction and monitoring. The processed data and derived operational information may further be stored in a database 110, which may be locally accessible within the system 100. The database 110 may store electrical current-related data, processed values, operational states, and other relevant information for further analysis, retrieval, or reporting, as known to the person skilled in the art.
[0030] In an embodiment, the system 100 may further include a remote network 120 configured to enable communication between the edge device 108 and a remote server 122. The remote network 120 may include, but is not limited to, the Internet, a wide-area network, or a cellular communication network, as known to the person skilled in the art. Through the remote network 120, selected data from the edge device 108 or the database 110 may be transmitted to the remote server 122 for long-term storage, advanced analytics, or remote access.
[0031] In an alternative embodiment, the remote server 122 and the remote network 120 may be absent from the system 100. In such an embodiment, all processing, analysis, and storage operations may be performed entirely by the edge device 108, without transmitting data to any external or cloud-based platform. The edge device 108 may be configured to locally execute all functions including, but not limited to, current signal acquisition, operational state determination, clustering, cycle-magnitude analysis, data storage in the database 110, and presentation of results through the interface 116 of the computing system 114 or the mobile device 112. This embodiment may be particularly suitable for deployments in environments with limited or no network connectivity, or where data privacy and localised processing are required. Accordingly, the system 100 may be operable in a fully self-contained manner without dependency on any remote or cloud-based infrastructure.
[0032] In operation, the sensors 104 may continuously acquire electrical current signals from the machine 102 (refer to FIG. 4D). In embodiments where the edge device 108 is directly connected to the machine 102, the acquired signals may be transmitted directly to the edge device 108 for processing.
[0033] In embodiments where the edge device 108 is connected via the local network 106, the acquired signals may be communicated through the local network 106 to the edge device 108. The edge device 108 may process the incoming data using the processors 204 to derive machine operational characteristics and determine operating states. The derived information may be stored locally in the database 110, displayed to users through the mobile device 112 and/or the computing system 114 via the interface 116, and optionally transmitted to the remote server 122 through the remote network 120 for further analysis and storage, where such remote infrastructure is present.
[0034] Referring to FIG. 2, in an embodiment, the system 100 depicting three primary architectural layers: an edge intelligence layer 202, a user interface layer 230, and a cloud layer 240 is disclosed. This layered architecture may enable a modular and scalable deployment of the system 100 across industrial environments of varying complexity and scale, as would be known to the person skilled in the art of distributed computing and industrial IoT system design.
[0035] In an embodiment, the processor 204 may be further configured to receive machine-related input data corresponding to the machine 102, including but not limited to machine type, installation location, machine model, rated current, or configuration parameters. A machine identifier may be defined as a unique identifier associated with the machine 102, which may include, but may not be limited to, an alphanumeric code, a universally unique identifier (UUID), a serial number, or any other unique reference known to the person skilled in the art, that may distinguish the machine 102 from other machines within the system 100. Based on the received machine-related input data, the processor 204 may generate at least one machine identifier corresponding to the machine 102. A gateway identifier may be defined as a unique identifier associated with the edge device 108, which may include, but may not be limited to, a network address, a media access control (MAC) address, a device serial number, or any other unique reference that may distinguish the edge device 108 from other edge devices within the system 100. The processor 204 may further generate at least one gateway identifier corresponding to the edge device 108. The processor 204 may map the machine identifier to the gateway identifier to establish a persistent association between the machine 102 and the edge device 108, wherein such mapping may enable unambiguous identification of the machine 102 during processing of the stream of current samples. The mapping between the machine identifier and the gateway identifier may be stored in the database 110 and may be retrieved during processing to associate acquired current samples with the corresponding machine 102. When data is transmitted to the remote server 122 through the remote network 120, at least one of the machine identifier or the gateway identifier may be transmitted together with the data, thereby enabling the remote server 122 to correlate the transmitted data with the specific machine 102 from which the data was acquired. For example, in an industrial facility operating multiple machines, the identifier pairing may allow the remote server 122 to distinguish and independently process data streams from different machines, thereby enabling machine-specific monitoring, analysis, and reporting across the facility.
[0036] In an embodiment, the edge intelligence layer 202 may be the most proximate layer of the system 100, being physically and logically closest to the machine 102 and the sensors 104. The edge intelligence layer 202 may include the edge device 108 comprising processors 204. The processors 204 may include, but are not limited to, one or more central processing units (CPUs), microcontrollers, digital signal processors (DSPs), field-programmable gate arrays (FPGAs), or application-specific integrated circuits (ASICs), or any combination thereof, as would be known to the person skilled in the art of embedded systems design. The processors 204 may serve as the central computational support of the edge device 108, orchestrating the operation of all other modules within the edge intelligence layer 202.
[0037] The current acquisition module 206 may be operably coupled to the processors 204 and may be configured to receive electrical signals from the sensors 104 associated with the machine 102. In an embodiment, the sensors 104 may generate analog signals corresponding to measured current, and optionally voltage or power. The current acquisition module 206 may sample such signals at a predefined sampling interval, for example, in the range of 1 kHz to 10 kHz. Where analog signals are received, the current acquisition module 206 may perform analog-to-digital conversion to generate a sequence of digital samples. For example, when a motor-driven machine starts operating, a rise in current may be captured as a sequence of increasing sample values.
[0038] In an embodiment, the stream of current samples acquired by the current acquisition module 206 may exhibit characteristic variations over time that correspond to the operating behaviour of the machine 102. Such characteristic variations may be referred to as current patterns, wherein a current pattern may be defined as a recurring sequence of current magnitude values or a recurring profile of current signal behaviour observed over one or more predefined time intervals during operation of the machine 102. A current pattern may be characterised by one or more attributes including, but not limited to, a magnitude range within which the current values fall, a duration for which the current remains within the magnitude range, a rate of change of current at the commencement and conclusion of the pattern, and a degree of variation exhibited by the current within the pattern. For example, during an idle condition of the machine 102, the current pattern may be characterised by a substantially constant low magnitude with minimal variation, whereas during an active operational cycle of the machine 102, the current pattern may be characterised by a sustained elevated magnitude with observable fluctuations corresponding to mechanical load variations. The current pattern that recurs consistently across multiple operational cycles of the machine 102 may be referred to as a recurring current signature pattern, wherein the recurring nature of the pattern may indicate a repeatable and identifiable operating behaviour of the machine 102. The current acquisition module 206 may provide the stream of current samples to the signal filtering module 216 for removal of transient components, following which the filtered current samples may be analysed to identify and characterise such recurring current signature patterns. The identified recurring current signature patterns may subsequently be grouped into behaviour clusters by the clustering module 210, wherein each behaviour cluster may represent a distinct and recurring operating behaviour of the machine 102, thereby enabling classification of the operating condition and identification of operational cycle boundaries based on transitions between current signature patterns.
[0039] The signal filtering module 216 may receive the sampled data from the current acquisition module 206 and may be configured to remove noise and unwanted disturbances from the acquired current signal using a transient-exclusion filter. The transient-exclusion filter may be configured to identify current spikes having a duration less than a predefined minimum spike duration and may remove such identified spikes from the stream of current samples, thereby producing filtered current samples that more accurately reflect the actual operating behaviour of the machine 102. A current spike may be defined as a short-duration excursion in the current signal wherein the current magnitude rises above or falls below a reference level for a duration that is less than the predefined minimum spike duration, such excursions being attributable to electrical interference, switching transients, or mechanical noise rather than genuine changes in machine operating behaviour. The predefined minimum spike duration may be a configurable parameter, which may be set based on the characteristics of the machine 102 and the nature of the operating environment.
[0040] In an alternating embodiment, the transient-exclusion filter may be implemented as a duration-based gating filter, wherein a detected current excursion may be retained in the filtered current samples only if the duration of the excursion meets or exceeds the predefined minimum spike duration, and may be removed from the filtered current samples if the duration of the excursion is less than the predefined minimum spike duration. In another embodiment, the transient-exclusion filter may additionally incorporate a low-pass filtering stage configured to attenuate high-frequency noise components, or a moving average filtering stage configured to smooth short-duration fluctuations in the current signal, prior to or following the duration-based gating operation. For example, if a current spike of 2 milliseconds duration appears in the signal due to electrical switching interference, and the predefined minimum spike duration is configured as 10 milliseconds, the transient-exclusion filter may identify and remove such a spike from the stream of current samples, thereby ensuring that the filtered current samples reflect only sustained current variations corresponding to genuine machine operating conditions.
[0041] A variation analysis module 214 may be operably coupled to the processors 204. The variation analysis module 214 may be configured to analyse changes in the filtered signal over time within a predefined time window. In one embodiment, the predefined time window may be a fixed duration, such as 100 milliseconds, 250 milliseconds, or 500 milliseconds, as may be selected based on the application. The variation analysis module 214 may compute statistical measures such as variance or rate of change of current within each time window. For example, when the machine 102 is idle, the current within the predefined time window may remain nearly constant, resulting in low variation values. When the machine 102 is actively performing an operation, such as cutting or pressing, the current within the predefined time window may exhibit continuous fluctuations, resulting in higher variation values. Based on such time-window-based analysis, the variation analysis module 214 may distinguish between stable and active operating conditions.
[0042] In one embodiment, the operating condition of the electrical machine 102 may be determined based solely on the measure of variation computed by the variation analysis module 214, independent of predefined threshold levels, model training, and historical data storage. In such an embodiment, the operating state module 220 may set the operating condition to a running state when the measure of variation exceeds a predefined variation value over the predefined time window, indicating continuous fluctuation in the current signal, and may set the operating condition to an idle state when the measure of variation remains below the predefined variation value over the predefined time window, indicating stability in the current signal. The measure of variation may represent a magnitude of change in the current signal within the predefined time window, and the determination of the operating condition may be made independent of absolute current magnitude values.
[0043] The cycle-magnitude engine 212 may be primarily configured to identify boundaries of machine operation cycles and to maintain a count of completed production cycles over a monitoring period. A cycle start boundary may be identified when the current signal exceeds a predefined threshold value and remains above the threshold for at least a minimum duration or a predefined number of consecutive samples. A cycle end boundary may be identified when the current signal falls below a corresponding threshold and remains below the threshold for at least the minimum duration. Upon detection of a complete cycle transition from a cycle start boundary to a cycle end boundary, the cycle-magnitude engine 212 may increment a production cycle counter, thereby enabling determination of the total number of production cycles completed within any given monitoring period, shift, or operational interval. In addition to cycle counting, the cycle-magnitude engine 212 may compute representative magnitude-related parameters for each identified cycle, including root-mean-square (RMS) current, mean current, and peak current values, wherein such parameters may serve as per-cycle features for quality assessment and cluster-based classification.
[0044] In one embodiment, a cycle start boundary may be defined as a point in time at which the signal exceeds a predefined threshold value and remains above the threshold for at least a minimum duration or a predefined number of consecutive samples, wherein the threshold corresponds to a reference signal level, such as a current magnitude, used to distinguish between idle and active operating conditions of the machine 102. The threshold may be predetermined, configured, or adaptively derived based on observed signal characteristics. Similarly, a cycle end boundary may be defined as a point in time at which the signal falls below the threshold and remains below the threshold for at least a minimum duration. In certain implementations, hysteresis may be applied by defining separate entry and exit thresholds to prevent false detection due to signal fluctuations. The entry threshold may correspond to a higher value for detecting a transition from an idle condition to an active condition, while the exit threshold may correspond to a lower value for detecting a transition from an active condition to an idle condition. For example, when a conveyor motor starts, a sustained rise in current above the entry threshold may indicate the beginning of a cycle, and when the motor stops, a sustained drop in current below the exit threshold may indicate the end of the cycle.
[0045] In an embodiment, the processor 204 may be configured to determine the operating condition of the electrical machine 102 based on a plurality of predefined threshold levels applied to the current signal. The plurality of predefined threshold levels may comprise an upper threshold, a lower threshold, and one or more intermediate threshold levels, thereby enabling classification of the operating condition into more than two states. The operating condition may be determined according to the range within which the current signal falls relative to the plurality of predefined threshold levels. For example, a current signal below the lower threshold may indicate an idle state, a current signal between the lower threshold and an intermediate threshold may indicate a light-load running condition, and a current signal above the upper threshold may indicate a heavy-load or overload condition. Hysteresis may be applied by defining, for each operating condition, a dedicated entry threshold and a dedicated exit threshold. The operating condition may be switched only when the current signal crosses the applicable entry threshold and remains beyond the corresponding exit threshold for at least a predefined duration, thereby preventing spurious state changes due to transient signal fluctuations near threshold boundaries.
[0046] Further, in an embodiment, the processor 204 may be configured to receive one or more digital input signals from the electrical machine 102, wherein each digital input signal carries one of two discrete states, such as a logic high or logic low value, representative of a binary operational condition of the machine 102 such as a run command, stop command, or mode selection indicator. The digital input mapping engine 224 may map combinations of the discrete states of the digital input signals to corresponding operating conditions based on predefined mapping conditions. For example, a digital input signal in a high state indicating "machine running" may be mapped to a running operating condition, while a digital input signal in a low state indicating "machine stopped" may be mapped to an idle operating condition. Where multiple digital input signals are simultaneously received, the digital input mapping engine 224 may evaluate the combination of discrete states to resolve the appropriate operating condition.
[0047] The evaluation module 208 may be configured to interpret the processed signal and one or more derived parameters to determine intermediate operating conditions of the machine 102, and to validate whether a detected cycle boundary transition constitutes a valid and complete production cycle suitable for inclusion in the production cycle count. The derived parameters may include magnitude-related parameters such as root-mean-square (RMS) current, mean current, peak current, and minimum current, as well as variation-related parameters including variance, standard deviation, and rate of change of current over the predefined time window. The evaluation module 208 may further consider temporal parameters such as cycle duration and persistence of a condition. For example, if a detected current excursion does not persist for a minimum predefined duration, the evaluation module 208 may flag the corresponding event as an incomplete or anomalous cycle, thereby excluding it from the production cycle count. Similarly, if the current magnitude and variation remain below corresponding threshold values for a defined duration, the evaluation module 208 may determine that the machine 102 is in an idle condition and that no productive cycle has occurred during the corresponding interval.
[0048] The clustering module 210 may be configured to group processed data into clusters corresponding to different operating behaviours, wherein the primary purpose of such grouping may be to distinguish valid productive cycle intervals from idle periods, dry-run conditions, or anomalous current excursions, thereby supporting accurate determination of the production cycle count. The clustering module 210 may implement a K-means clustering technique. Under such an approach, data points representing current magnitude values or feature vectors may be grouped based on similarity. For example, a first cluster may include low current values corresponding to idle operation, a second cluster may include moderate values corresponding to normal productive running associated with a valid production cycle, and a third cluster may include higher values corresponding to heavy load or abnormal cycle conditions.
[0049] Each cluster may be defined by a centroid, wherein the centroid may represent a computed central reference value corresponding to the mean of all current magnitude values assigned to that cluster. For example, if a set of current magnitude values observed during normal productive operation of the machine 102 ranges between 8 amperes and 12 amperes across multiple production cycles, the centroid of the corresponding cluster may be computed as the arithmetic mean of such values, for example 10 amperes, thereby serving as a representative reference point for that cluster in the feature space. The centroid may be initially assigned at the commencement of the clustering process and may be iteratively updated as new data points are assigned to the cluster, wherein the updated centroid may be recomputed as the mean of all data points currently assigned to that cluster. This iterative assignment and re-computation may continue until the centroid positions stabilize or a predefined convergence condition is satisfied. Each cluster may further correspond to a bounded interval of current magnitude values determined based on distances of the current magnitude values from the centroid, wherein a data point may be considered to belong to a cluster if the distance between the data point and the centroid of that cluster is less than the distance between the data point and the centroids of all other clusters.
[0050] Each new data point may be assigned to a cluster based on proximity to a cluster centroid, wherein proximity may be determined using a distance metric such as Euclidean distance computed between the new data point and each of the available cluster centroids. The clustering module 210 may assign each current magnitude value to the cluster associated with the minimum distance between the current magnitude value and the corresponding cluster centroids. For example, a newly acquired current magnitude value of 9.5 amperes may be compared against the centroid of the idle cluster at 3 amperes, the centroid of the normal productive running cluster at 10 amperes, and the centroid of the heavy load cluster at 16 amperes, and may accordingly be assigned to the normal productive running cluster owing to the minimum distance between the data point and that cluster centroid. The cluster identifier assigned to each current magnitude value may thereby indicate whether the corresponding interval contributes to a valid production cycle or represents an idle, anomalous, or non-productive period, and the determined cluster identifier may be used to support real-time cycle validation and production cycle counting.
[0051] In an embodiment, the clustering module 210 may be operably coupled to a trained machine learning classifier configured to determine the operating condition of the electrical machine 102 based on relationships between the current magnitude values derived from the filtered current samples and the corresponding operating conditions. The trained machine learning classifier may be implemented as a clustering-based classification model, wherein the cluster centroids established by the K-means clustering technique may serve as the learned parameters of the classifier. Specifically, the trained machine learning classifier may receive, as input, one or more current magnitude values or feature vectors derived from the filtered current samples over a predefined time interval, and may produce, as output, a cluster identifier corresponding to the cluster whose centroid is nearest to the input value in the feature space, wherein the cluster identifier may be mapped to a corresponding operating condition of the electrical machine 102. The trained machine learning classifier may be trained at the edge device 108 using historical data comprising current magnitude values and corresponding validated operating conditions accumulated over a plurality of production cycles, as stored in the database 110. Upon training, the classifier may be deployed at the edge device 108 via the trained classifier inference module 310 to perform real-time classification of the operating condition of the electrical machine 102 for each subsequently acquired current magnitude value or feature vector. The trained machine learning classifier may be periodically retrained using newly accumulated data to reflect changes in machine behaviour arising from factors including, but not limited to, tool wear, load variation, or process modifications, thereby enabling adaptive and accurate classification of operating conditions over time.
[0052] In addition to clustering based on current magnitude values, the clustering module 210 may be configured to identify and learn recurring patterns in the electrical behaviour of the machine over time. Such patterns may correspond to temporal and statistical characteristics of machine operation, including steady-state operation, cyclic variations, transient events, and gradual drifts in current consumption. For example, a repetitive sequence of current magnitude values observed across multiple machine cycles may represent a normal production pattern, whereas deviations from such learned patterns may indicate anomalies, faults, or changing load conditions. The clustering module 210 may analyse distributions, transitions, and persistence of data points within clusters to capture such patterns, including frequency of occurrence, duration of cluster membership, and transitions between clusters. In one embodiment, patterns may be represented as sequences or trajectories of cluster identifiers over time, thereby enabling characterization of machine behaviour beyond individual data points. These learned patterns may be utilized by the system to improve classification accuracy, detect abnormal conditions, and adapt to variations in machine operation through continuous machine learning-based updates
[0053] The digital input mapping engine 224 may be configured to receive discrete input signals from the machine 102, when such signals are available. These signals may include machine run status or mode indicators. The digital input mapping engine 224 may map combinations of these inputs to corresponding operating states. For example, a signal indicating “machine running” may be mapped to a running state, while a signal indicating “machine stopped” may be mapped to an idle state.
[0054] The time and event linking module 218 may be configured to associate identified cycle boundaries and operating conditions with corresponding time information, thereby enabling time-referenced tracking of production cycle counts across monitoring periods. A cycle may correspond to a complete production operation of the machine 102, such as performing a machining operation, completing a pressing action, or executing a movement sequence. The time and event linking module 218 may record a timestamp when a cycle start boundary or a cycle end boundary is identified, as determined by the cycle-magnitude engine 212, thereby enabling computation of individual cycle durations and cumulative production cycle counts. The time and event linking module 218 may further aggregate cycle count records across predefined intervals such as shifts, days, or production orders, thereby providing a chronological and quantitative record of machine productivity. The time and event linking module 218 may also record timestamps corresponding to changes in operating conditions determined by the operating state module 220, wherein such state transitions may serve as indicators of cycle boundary events. For example, a transition from an idle condition to a running condition may be recorded as a cycle start timestamp, and a subsequent transition from running to idle may be recorded as a cycle end timestamp, with the corresponding completed cycle incrementing the production cycle counter..
[0055] In an embodiment, the processor 204 may be configured to record a timestamp corresponding to the stream of current samples, wherein the timestamp comprises a date value and a time value associated with the acquisition of the stream of current samples. The processor 204 may link the timestamp with at least one of the current magnitude values, the determined operating condition, or the transmitted data. The linked timestamp may enable identification of the occurrence of any given operating condition with respect to a specific date and time, thereby providing a time-referenced record of machine behaviour suitable for production tracking, audit purposes, or post-hoc performance analysis.
[0056] The operating state module 220 may be configured to determine a final operating condition of the machine 102 based on processed data generated by the system, wherein the determination of operating conditions may primarily serve to identify cycle boundary transitions and support accurate counting of production cycles. In one embodiment, the operating state module 220 may consider results obtained from the evaluation module 208, the clustering module 210, the variation analysis module 214, and the digital input mapping engine 224. A transition from the idle condition to the running condition, as determined by the operating state module 220, may indicate the cycle start boundary, and a subsequent transition from the running condition to the idle condition may indicate the cycle end boundary, whereupon the production cycle counter may be incremented. For example, if both the variation analysis module 214 and the clustering module 210 indicate a transition to active operation, the operating state module 220 may determine that a new production cycle has commenced. If a sudden high current is detected beyond a predefined limit defined by the user or by the system 100, the operating state module 220 may determine a surge or alarm condition, wherein such a condition may be flagged as an anomalous cycle event to be recorded separately from valid production cycles in the database 110.
[0057] The operating state learning module 222 may be configured to update system parameters over time based on observed data. In one embodiment, the operating state learning module 222 may analyse historical data stored in the database 110 and adjust thresholds or cluster definitions. For example, if the normal operating current of the machine changes due to tool wear or load variation, the operating state learning module 222 may update the reference values to maintain accurate classification.
[0058] In an embodiment, the processor 204 may be configured to train a classifier at the edge device 108, wherein the trained classifier is configured to determine the operating condition of the electrical machine 102 based on relationships between current magnitude values derived from the filtered current samples and corresponding operating conditions. The training may comprise the following operations. First, the processor 204 may identify one or more production cycles from the stream of current samples based on variation in the current signal corresponding to a start and an end of operation of the electrical machine 102, using the cycle boundary detection techniques described with reference to the cycle-magnitude engine 212. Second, for each identified production cycle, the processor 204 may determine one or more current magnitude values from the filtered current samples, including RMS current, mean current, peak current, and minimum current. Third, the determined current magnitude values may be linked with a corresponding operating condition of the electrical machine 102 to form stored training data. Fourth, the training data corresponding to a plurality of production cycles may be stored in the database 110 as historical data. Fifth, the classifier may be trained using the stored historical data to establish a relationship between the current magnitude values and the operating condition, for example using a K-means clustering approach as described with reference to the clustering module 210 and the classifier training module 316. Upon completion of training, the edge device 108 may deploy the trained classifier via the trained classifier inference module 310 to determine the operating condition for a subsequent production cycle based on current magnitude values derived from that cycle.
[0059] Further, the processed data, feature values, and operating states determined by the edge device 108 may be communicated to the user interface layer 230 for visualization and interaction. The user interface layer 230 may include the user interface 116, which may be communicatively coupled to the edge intelligence layer 202 through the database 110 or through direct communication links, as known to the person skilled in the art. The user interface 116 may include a dashboard module 232 configured to present real-time operational information of the machine 102.
[0060] In one embodiment, the dashboard module 232 may display the production cycle count as the primary metric, including total cycles completed within a current shift, a current day, or a selected production order. Additional parameters such as average cycle duration, minimum and maximum cycle durations, and current operating state may be displayed as supporting information. For example, when the cycle-magnitude engine 212 detects completion of a production cycle, the dashboard module 232 may update the cycle count display in real time, thereby enabling operators and supervisors to monitor production output against targets. Alert conditions such as surge events or anomalous cycles may also be indicated on the dashboard module 232, along with their corresponding timestamps, to enable prompt corrective action.
[0061] A visualization module 234 within the user interface 116 may be configured to render graphical representations of the processed data. In one embodiment, the visualization module 234 may display time-series plots of current signals, trend charts of RMS values, or graphical representations of cluster distributions. For example, a user may view a waveform plot showing current variations during a machining cycle, thereby enabling identification of abnormal patterns.
[0062] A dashboard interface module 236 may be configured to provide extended interface functionality, including historical data analysis, report generation, and configuration management. For example, a user may access historical cycle data stored in the database 110 to analyse performance over a selected time period or may configure threshold values or system parameters through the user interface 116.
[0063] The user interface layer 230 may thereby provide a mechanism for presenting processed data generated at the edge device 108 in a human-readable form, enabling operators, engineers, or supervisors to monitor machine behaviour and respond to detected conditions.
[0064] The system 100 may further include the cloud layer 240 comprising the cloud server 122, which may be communicatively coupled to the edge device 108 through the remote network 120, as described with reference to FIG. 1. The cloud layer 240 may receive selected data from the edge intelligence layer 202 for further processing, storage, and analysis.
[0065] A data ingestion module 242 within the cloud server 122 may be configured to receive data transmitted from one or more edge devices 108. In one embodiment, the data ingestion module 242 may validate incoming data streams, perform formatting or normalization, and route the data to appropriate modules. For example, the data ingestion module 242 may receive time-stamped current magnitude values and operating state information and organize such data for storage.
[0066] A cloud storage module 244 may be configured to store the received data in a scalable storage system. The cloud storage module 244 may maintain historical records including raw signal segments, processed feature vectors, operating states, and configuration parameters. For example, cycle-level data collected over several days or months may be stored for long-term analysis and reporting.
[0067] An analytics module 246 may be configured to perform advanced analysis on the stored data. In one embodiment, the analytics module 246 may execute machine learning algorithms, statistical analysis, or trend analysis to derive insights regarding machine performance. For example, the analytics module 246 may identify gradual increases in current consumption over time, which may indicate tool wear or mechanical degradation.
[0068] A digital twin module 248 may be configured to generate and maintain a virtual representation of the machine 102 based on data received from the edge device 108. In one embodiment, the digital twin module 248 may simulate machine behaviour under varying conditions or predict future operating states. For example, based on historical current patterns, the digital twin module 248 may predict the likelihood of an upcoming fault condition.
[0069] The cloud layer 240 may further communicate processed results, updated parameters, or trained models back to the edge device 108. In one embodiment, updated cluster centroids, threshold values, or classifier parameters generated by the analytics module 246 may be transmitted to the operating state learning module 222 for deployment. This feedback mechanism may enable continuous improvement of machine state determination.
[0070] In this manner, the edge intelligence layer 202, the user interface layer 230, and the cloud layer 240 may operate in a coordinated manner. The edge device 108 may perform real-time data acquisition and processing, the user interface layer 230 may provide visualization and interaction, and the cloud layer 240 may support large-scale storage, advanced analytics, and adaptive learning. Such a distributed architecture may enable efficient, scalable, and responsive monitoring of machine operation, as known by the person skilled in the art.
[0071] Referring to FIG. 3, a system 300 for machine learning-based monitoring and determination of machine operating states is disclosed. The system is implemented at an edge device 320 and operates as a closed-loop learning architecture in which data acquisition, feature extraction, inference, validation, and continuous model training are performed locally.
[0072] The system 300 includes an analog-to-digital conversion stage 302 configured to receive electrical signals from a machine and convert the analog current signals into digital samples at a predefined sampling rate suitable for downstream processing.
[0073] The digital samples are provided to a cycle identification module 304, which segments the continuous stream of samples into discrete machine cycles. Each cycle corresponds to a complete production operation of the machine, identified using characteristics such as current thresholds, inrush detection, or deviation from idle conditions. The cycle identification module 304 may maintain a running tally of completed production cycles, wherein the cycle count is incremented upon detection of each complete cycle boundary transition from an active state to an idle state. The accumulated production cycle count may be the primary output of the cycle identification module 304 and may be communicated to downstream modules for validation, storage, and transmission.
[0074] The segmented cycle data is processed by a current magnitude determination module 306, which extracts a set of features for each cycle. The features include statistical and temporal parameters such as root-mean-square current, mean current, peak current, minimum current, variance, and rate of change of current. These features are combined to form a feature vector representing the operational behaviour of each cycle.
[0075] The feature vector is provided to an operating state module 220, which may assign an initial operating state using predefined rules or thresholds. This initial classification may serve as baseline labelling input for subsequent machine learning processes.
[0076] The feature vector is further input to a trained classifier inference module 310, which applies a machine learning model to predict the operating state of the machine. In one embodiment, the model comprises a clustering-based algorithm in which stored cluster centroids represent learned machine behaviours. The feature vector is compared with the centroids, and the nearest centroid determines the predicted operating state.
[0077] The predicted operating state is provided to a decision control module 312, which performs validation of the prediction using temporal consistency, logical rules, and comparison with prior states. The decision control module 312 generates outputs including the validated production cycle count, per-cycle timestamps, cycle durations, alerts, control signals, and production metrics, and identifies validated cycle events. The production cycle count may represent the primary validated output of the decision control module 312, wherein each validated cycle boundary transition may result in an increment of the cycle counter. The validated outputs, including the production cycle count and associated per-cycle data, may be stored in the database 110 and also provided as feedback for model improvement.
[0078] The system further includes an operating state learning module 222, which is responsible for machine learning-based training and updating of the model. The operating state learning module 222 includes a historical data retrieving module 314 configured to retrieve stored feature vectors and corresponding validated operating states from database 110.
[0079] The retrieved data is processed by a training data formation module 318, which prepares the dataset for training by performing data cleaning, filtering of invalid cycles, removal of outliers, and normalization of feature values to ensure consistency across the dataset.
[0080] The prepared dataset is then provided to a classifier training module 316, which performs machine learning training. In one embodiment, the classifier training module 316 executes a clustering algorithm such as K-means, wherein feature vectors are iteratively assigned to clusters based on similarity, and cluster centroids are updated as the mean of assigned feature vectors. Through this iterative optimization process, the model learns distinct representations of machine operating states, including idle, normal operation, and abnormal conditions.
[0081] The updated model parameters, including refined cluster centroids, are transmitted to the trained classifier inference module 310 as an updated trained model. This enables the inference module to perform more accurate predictions based on newly learned patterns.
[0082] Training is performed continuously in a closed-loop manner. The decision control module 312 provides validated feedback and labelled data, which is stored in database 110 and periodically retrieved by the operating state learning module 222. Retraining may be triggered upon accumulation of a predefined volume of new data, detection of drift in feature distributions, or user-defined conditions. During retraining, recent data reflecting current machine behaviour is incorporated, allowing the model to adapt dynamically to changes such as tool wear, load variation, or process modifications.
[0083] Accordingly, the system implements a continuous machine learning pipeline at the edge device 320, wherein inference, validation, feedback generation, and model retraining are iteratively performed to improve accuracy and robustness over time without reliance on remote computing resources.
[0084] Referring to FIG. 4A, in an embodiment, a time-series current waveform of the machine is disclosed. The time-series current waveform includes a current magnitude represented along the vertical axis I and time is represented along the horizontal axis T. The time-series current waveform may correspond to a stream of current samples acquired from the machine over a monitoring duration and grouped into the plurality of predefined time windows. For each predefined time window, the system may maintain a register of date 402 and a register of time 404. The register of date 402 may store the calendar date associated with the corresponding acquired window, while the register of time 404 may store the time stamp, time interval, or time instant corresponding to such window. In one embodiment, the predefined time window may be a fixed interval such as a few milliseconds, seconds, or any configurable interval suitable for the machine under observation. The date registers 402 and the time register 404 together may enable chronological indexing of each windowed current segment so that each observed current pattern may be correlated with a particular production instant, event occurrence, or machine cycle.
[0085] As shown in FIG. 4A, the acquired current waveform may exhibit a plurality of current excursions, plateaus, rising segments, falling segments, and transient peaks over successive windows. Such waveform behaviour may correspond to different machine cycles, sub-cycles, or operational stages. The cycles shown illustratively as Cycle-1 to Cycle-5 may represent counted production operations, repeated machine actions, or monitored current events occurring over successive predefined windows, wherein each such cycle may contribute to the total production cycle count maintained by the system 100. Certain cycles may be considered valid and complete production cycles and may be included in the production cycle count, whereas certain other cycles may indicate incomplete, anomalous, or non-productive current excursions and may be excluded from or separately classified within the production cycle count. Since the current waveform is associated with the date register 402 and the time register 404, any detected cycle may be traced back to the exact date and time window at which the cycle commenced and concluded, thereby enabling shift-wise, day-wise, or order-wise aggregation of production cycle counts. This may be useful for production tracking, output analysis, audit trail generation, fault investigation, and model learning, as would be understood by a person skilled in the art.
[0086] In one exemplary implementation, the current values within each predefined time window may be sampled, digitized, and stored with corresponding entries in the date register 402 and the time register 404. Thereafter, one or more representative magnitude values, such as RMS current, mean current, peak current, or equivalent derived magnitude, may be computed for each window. These magnitude values may then be compared with one or more operating state ranges of the machine in order to determine the instantaneous or window-based operating state.
[0087] Referring to Fig. 4B, in an embodiment, an operating state range map of the machine is disclosed. The operating state range map includes a plurality of current ranges that are defined for classifying different operating states. In an embodiment, the operating state ranges defined by the thresholds 406 to 414 may be predetermined, configured, or adaptively generated based on machine behaviour. In one embodiment, such thresholds may be pre-set by a user during commissioning of the system. For example, a user may define an idle threshold 414 based on observed no-load current, a running threshold range between 408 and 410 based on typical production current, and an alarm threshold above 406 based on rated current limits of the machine 102.
[0088] In another embodiment, the system may automatically determine such threshold ranges by analysing historical current data. For example, the system may compute statistical distributions of current magnitude over a plurality of cycles and may define threshold ranges based on percentile values or clustering of the data. For example, the lowest 10% of observed current values may correspond to the idle range, mid-range values may correspond to setup or running conditions, and the highest values may correspond to warning or alarm states. In this way, the system may adaptively generate threshold ranges specific to the operating characteristics of the machine 102 without requiring manual configuration.
[0089] Referring to FIG. 4C, an exemplary mapping of a time-varying current waveform to the operating state ranges 406 to 414 is illustrated. In one embodiment, the current waveform may be divided into predefined time windows, and a representative current magnitude, such as an RMS value, may be computed for each time window. Each computed value may then be compared with the threshold ranges 406 to 414 to assign an operating state to that time window.
[0090] For example, during an initial time interval, the computed current magnitude may lie within the range corresponding to threshold 414, thereby indicating an idle condition. In a subsequent interval, the magnitude may increase and fall within the range between thresholds 412 and 410, indicating a setup or dry-run condition. Thereafter, the magnitude may rise further into the range between thresholds 410 and 408, indicating a normal running condition. If the magnitude subsequently exceeds threshold 406, the corresponding time window may be classified as an alarm condition.
[0091] In an alternative embodiment, the current magnitude may oscillate near a boundary between two ranges. In such cases, additional validation techniques such as hysteresis or minimum dwell time may be applied to ensure stable classification. For example, a transition from running to warning state may be confirmed only if the current remains within the warning range for a predefined duration or number of consecutive time windows.
[0092] Referring to FIG. 4D, in an embodiment, a time-series representation of machine current for distinguishing between idle and running states is disclosed, wherein an idle state region 420 and a running state region 422 are illustrated. The vertical axis represents current magnitude in amperes and the horizontal axis represents time progression. The waveform may correspond to continuously acquired current samples from the machine 102, optionally processed using filtering and smoothing techniques to generate one or more representative traces. The idle state region 420 corresponds to a condition in which the machine 102 is powered but not actively performing an operation, such as a CNC machine that is powered ON but executing no machining cycle and is characterised by a substantially constant low current magnitude with minimal variation. The running state region 422 corresponds to active machine operation, such as motor actuation or tool engagement, and exhibits an elevated current magnitude with observable variations including transient peaks, periodic fluctuations, or step changes. Each region may be identified across one or more predefined time windows associated with corresponding date and time information, thereby enabling tracking of idle and running durations over a monitoring period.
[0093] Transitions between the idle state region 420 and the running state region 422 may be identified based on changes in current magnitude, variation characteristics, or both. In one embodiment, such a transition may be detected when the current crosses a predefined threshold and remains beyond the threshold for at least a minimum duration or a predefined number of consecutive samples. In another embodiment, variation-based criteria alone may be applied, independent of absolute current magnitude. The illustrated waveform may include segments of the running state region 422 interspersed with segments of the idle state region 420, reflecting intermittent operation such as loading, tool repositioning, or transitions between production cycles. The waveform may accordingly be segmented into discrete time windows, and each window may be assigned a state label, wherein the resulting sequence of labelled windows may be used to compute operational metrics including total idle duration, total running duration, number of cycles, and frequency of state transitions, and may further serve as input for clustering based classification, anomaly detection, or training of machine learning models. In an alternate embodiment, the idle state region 420 may itself serve as a reference baseline for adaptive threshold determination, wherein an average or median current value computed over identified idle windows defines the baseline and thresholds for detecting the running state region 422 are defined relative thereto. The representation of FIG. 4D is illustrative, and waveform characteristics may vary depending on machine type, operational conditions, and environmental factors.
[0094] Referring to FIG. 5, an exemplary representation of machine status determination using clustering is disclosed. FIG. 5 illustrates use of a K-means clustering technique for deriving operating conditions of the machine 102 based on grouping of current magnitude values or feature vectors into a plurality of clusters having corresponding centroids. In an embodiment, a plurality of data points is represented in a feature space, wherein each data point may correspond to a current magnitude value or to a feature vector derived from filtered current samples over predefined time intervals or production cycles. The feature vector may include one or more parameters such as root mean square (RMS) current, mean current, peak current, minimum current, variance, or rate of change of current. For example, a time window corresponding to an idle machine condition may generate a feature vector having low RMS current and low variance, whereas a time window corresponding to an active machining operation may generate a feature vector having higher RMS current and larger fluctuations.
[0095] The clustering process may group the data points according to similarity in the feature space to form distinct clusters, including a first cluster 502, a second cluster 504, and a third cluster 506, using a K-means clustering algorithm, wherein each cluster includes data points associated with similar machine behaviour and is characterised by a corresponding centroid representing a mean location of the data points assigned to that cluster. In one embodiment, the K-means clustering process may begin by selecting a number of clusters, which may be predefined by a user, selected during commissioning, or determined based on the nature of the dataset. Initial centroid positions may be assigned in the feature space, each data point may be compared with the centroids using a distance metric such as Euclidean distance and assigned to the nearest centroid, and a revised centroid may thereafter be computed for each cluster as the mean position of all data points currently assigned thereto. This process of assignment and centroid updating may be repeated until the centroid positions stabilise or a predefined convergence condition is satisfied. The third cluster 506 may correspond to an idle or low load operating condition, wherein the associated data points are concentrated around a centroid representing relatively low current magnitude and low fluctuation, for example when the machine 102 is powered but not actively machining, pressing, or conveying. The second cluster 504 may correspond to a normal operating condition, wherein data points are distributed around a centroid representing moderate current magnitude values observed during productive operation such as a normal machining cycle in which a cutting tool engages a workpiece. The first cluster 502 may correspond to a running with warning or high load condition, wherein data points are grouped around a centroid representing current magnitude values higher than those of the second cluster 504, indicating that the machine 102 is drawing elevated current for a sustained duration, as may arise from overload, excessive tool engagement, abnormal process resistance, process drift, or progressive tool wear.
[0096] Once the first cluster 502, the second cluster 504, and the third cluster 506 are established, the clustering module 210 may classify newly acquired data in real time by computing a distance from each new data point to each centroid and assigning the data point to the nearest cluster, whereupon the operating condition of the machine 102 is determined according to the assigned cluster. In alternate embodiments, clustering based classification may be used together with additional decision logic; for example, if a data point is assigned to the first cluster 502 and the current signal remains elevated above a predefined threshold for at least a predefined duration, the system 100 may further classify the condition as a surge, overload, or alarm state, and where repeated assignments occur near a boundary between the first cluster 502 and the second cluster 504, temporal validation, hysteresis, or dwell time logic may be applied to reduce false state transitions. In an industrial IoT deployment, the clustering approach of FIG. 5 may be executed at the edge device 108, with filtered values, feature vectors, and cluster identifiers computed in near real time, the resulting cluster assignments, feature values, and state labels stored in the database 110 and communicated to a user interface for monitoring, and selected data transmitted through a network to a remote platform for long term storage, fleet level analytics, or retraining. The centroids associated with the first cluster 502, the second cluster 504, and the third cluster 506 may be updated over time, wherein newly acquired data may be incorporated into a retraining process such that shifts in normal machine behaviour due to tooling changes, wear, material variation, or environmental conditions are reflected in updated centroid positions, and the number of clusters may also be revised to represent additional operating conditions where appropriate.
[0097] Referring to FIG. 6, in an embodiment, a computer-implemented method 600 for monitoring an electrical machine based on current analysis is disclosed. The method 600 may be performed by an edge device, such as the edge device 108, in communication with one or more sensors associated with the machine.
[0098] At step 602, a current signal representative of electrical current drawn by the machine 102 is sampled at a predefined sampling interval to generate a stream of current samples. The current signal is generated by the sensor 104 disposed in a power supply line of the machine 102, and the sampling is performed by the edge device 108 to obtain a sequence of digital current samples representing instantaneous current values corresponding to different operating conditions of the machine 102.
[0099] At step 604, the stream of current samples is filtered using a transient-exclusion filter configured to identify and remove current spikes having a duration less than a predefined minimum spike duration, thereby producing filtered current samples that more accurately reflect the actual operating behaviour of the machine. The transient-exclusion filter may identify a current spike as a short-duration excursion in the current signal wherein the duration of the excursion is less than the predefined minimum spike duration, such excursions being attributable to electrical interference, switching transients, or mechanical noise rather than genuine changes in machine operating behaviour. The predefined minimum spike duration may be a configurable parameter set based on the characteristics of the machine and the operating environment. Only current excursions having a duration meeting or exceeding the predefined minimum spike duration may be retained in the filtered current samples for further processing.
[00100] At step 606, one or more current magnitude values are measured from the filtered current samples over a plurality of time intervals of predefined duration. The current magnitude values include at least one of root-mean-square current, mean current, peak current, or other statistical representations, thereby generating compact representations of the current signal over time.
[00101] At step 608, a clustering technique is applied to the one or more current magnitude values to group recurring current signature patterns into a plurality of behaviour clusters. The clustering is performed using a machine learning algorithm, including in one embodiment a K-means algorithm, such that each behaviour cluster represents a distinct operational pattern of the electrical machine 102. Each cluster is associated with a cluster identifier and is characterized by a centroid corresponding to grouped current magnitude values.
[00102] At step 610, an operating condition of the electrical machine 102 is classified using a trained machine learning classifier based on the clustered current magnitude values, wherein the classification of operating conditions primarily serves to identify cycle boundary transitions for the purpose of production cycle counting. The trained classifier maps the behaviour clusters or associated feature representations to one of a plurality of operating states. A transition from an idle operating state to a running operating state may be identified as a cycle start boundary, and a subsequent transition from a running operating state to an idle operating state may be identified as a cycle end boundary, whereupon the production cycle counter is incremented to reflect completion of a production cycle. The plurality of operating states further includes a surge state, wherein the surge state is identified when the current signal exceeds a predetermined threshold continuously for at least a predetermined duration, and wherein such a surge state may be recorded as an anomalous cycle event separately from valid production cycles.
[00103] At step 612, data may be transmitted to the remote server 122. The transmitted data may include the production cycle count representing the total number of completed production cycles within the monitoring period, per-cycle timestamps indicating the start time and end time of each production cycle, per-cycle duration values, at least a portion of the current magnitude values, a cluster identifier assigned to each value, and the determined operating condition of the machine. For example, the edge device may transmit a summary indicating that 47 production cycles were completed during the monitoring period, with an average cycle duration of 38 seconds, along with the corresponding timestamps and magnitude values for each cycle. Such transmitted data may be used for production tracking, remote monitoring, long-term storage, analytics, and reporting.
[00104] In one embodiment, the steps of the method 600 may be repeated continuously or periodically so as to provide ongoing monitoring of machine operation. For example, the edge device may repeatedly acquire current samples, update filtered values, compute magnitude values, perform clustering, determine operating conditions, and communicate results in near real time. In this manner, the method 600 may enable continuous and automated determination of machine states from electrical current behaviour. The sequence shown in FIG. 6 is an exemplary embodiment. One or more steps may be modified, combined, repeated, or supplemented depending on the machine type, deployment environment, or implementation requirements, as known to the person skilled in the art.
[00105] Referring to FIG. 7, in an embodiment, a computing environment 700 for implementing monitoring and determination of operating conditions of one or more machines is disclosed. The computing environment 700 may comprise a layered architecture including a presentation layer 710, a cloud server layer 720, an edge computing layer 730, and a data layer 740. Such an architecture may enable distributed processing, centralized management, and scalable deployment across multiple machines and locations.
[00106] The presentation layer 710 may provide user interaction and visualization capabilities. In one embodiment, the presentation layer 710 may include an operator interface 706 accessible to a system manager 702 and a monitoring dashboard 708 accessible to one or more customers 704. The operator interface 706 may enable configuration, system control, and monitoring of machine states, while the monitoring dashboard 708 may present real-time and historical data, including operating conditions, alerts, and performance metrics. For example, a user may view machine utilization, idle duration, and warning conditions through graphical representations.
[00107] The cloud server layer 720 may provide centralized processing, coordination, and management functions. In one embodiment, the cloud server layer 720 may include an API interface 712 configured to enable communication between the edge computing layer 730 and the presentation layer 710. An access control module 714 may manage authentication, authorization, and user roles. A workflow engine 716 may coordinate processing tasks, event handling, and system operations, while an analytics engine 718 may perform data analysis, trend identification, and model-related computations.
[00108] The cloud server layer 720 may further include a digital twin module 722 configured to maintain a virtual representation of machine behaviour based on received data. A device communication module 724 may manage communication with one or more edge devices, including receiving processed data and transmitting configuration updates or model parameters. For example, updated clustering parameters or threshold values may be communicated to the edge computing layer 730 to improve operating state determination.
[00109] The edge computing layer 730 may be positioned proximate to the machine and may perform real-time data acquisition and processing. In one embodiment, the edge computing layer 730 may include an edge data acquisition module 732 configured to receive current signals from sensors associated with the machine. A configuration manager 734 may manage local settings, thresholds, and operational parameters. A clustering and condition monitoring module 736 may process the acquired data to identify cycle boundaries, increment the production cycle counter upon detection of each complete production cycle, and determine operating conditions of the machine using techniques such as filtering, feature determination, and clustering, as described herein. For example, the edge computing layer 730 may classify machine operation into idle or running states to identify cycle boundary transitions and may maintain a real-time count of completed production cycles, which may be transmitted to the cloud server layer 720 for aggregation and analysis.
[00110] The data layer 740 may provide storage and data management capabilities. In one embodiment, the data layer 740 may include an operational data repository 742 configured to store real-time data and recent operating states, a historical data store 744 configured to maintain long-term records of machine behaviour, and an analytics database 746 configured to support advanced analysis and reporting. For example, historical data may be used to identify trends in machine performance or to update clustering models.
[00111] In operation, data acquired at the edge computing layer 730 may be processed locally to determine machine states and may be transmitted to the cloud server layer 720 for storage, analysis, and visualization. The cloud server layer 720 may, in turn, provide insights, configuration updates, and model parameters back to the edge computing layer 730. The presentation layer 710 may enable users to interact with the system, monitor machine conditions, and manage system behaviour.
[00112] In this manner, the computing environment 700 may provide an integrated and scalable architecture for monitoring and analysing machine operation, wherein real-time processing is performed at the edge, and centralized coordination, analysis, and visualization are performed through network-accessible infrastructure, as would be understood by a person skilled in the art.
[00113] Referring to FIG. 8, in an embodiment, a hardware block diagram of the edge device 108 for implementing one or more aspects of the present disclosure is disclosed. The computing system 800 may correspond to the edge device 108 and may be configured to perform data acquisition, processing, and operating state determination as described herein.
[00114] A processor 806 may be configured to execute instructions stored in the memory 810 to perform functions described herein, including acquisition of current samples, signal processing, feature determination, clustering, and operating state determination. The processors 806 may include, but are not limited to, central processing units (CPUs), microcontrollers, digital signal processors (DSPs), or other programmable logic devices. For example, the processors 806 may execute instructions to compute RMS current values over predefined time windows and assign corresponding operating states.
[00115] The memory 810 may store executable instructions, configuration parameters, and data associated with machine monitoring. In one embodiment, the memory 810 may include volatile memory, such as random-access memory (RAM), and non-volatile memory, such as flash memory or solid-state storage. The memory 810 may store data including current samples, filtered signals, feature vectors, cluster centroids, and determined operating states. For example, the memory 810 may maintain recent current magnitude values for real-time processing and historical data for learning and analysis.
[00116] An I/O device 802 may include devices configured to receive input or provide output to a user or external system. In one embodiment, the I/O devices 802 may include display units, keyboards, touch interfaces, or other user interaction devices. Additionally, the I/O devices 802 may include interfaces to sensors, such as current sensors, through which electrical signals from the machine may be received.
[00117] An I/O controller 804 may manage communication between the processors 806 and the I/O devices 802. The I/O controller 804 may handle data transfer, signal conditioning, and interfacing protocols required for communication with peripheral devices. For example, the I/O controller 804 may facilitate acquisition of current signals from sensors and delivery of processed outputs to display devices.
[00118] A network interface module 808 may be configured to enable communication between the computing system 800 and external systems, including remote servers, databases, or user devices. The network interface module 808 may support wired or wireless communication protocols, including Ethernet, Wi-Fi, cellular communication, or industrial communication protocols. For example, the network interface module 808 may transmit operating condition data to a remote platform and receive configuration updates or model parameters.
[00119] A security module 812 may be configured to provide security functions associated with data processing and communication. In one embodiment, the security module 812 may implement authentication, encryption, and data integrity mechanisms to ensure secure transmission and storage of data. For example, the security module 812 may encrypt data transmitted over a network and may verify access credentials for authorized users or devices.
[00120] The system bus 814 may provide a communication pathway between the components of the computing system 800. The system bus 814 may include one or more data buses, address buses, or control buses that facilitate data exchange among the processors 806, memory 810, I/O controller 804, network interface module 808, and security module 812.
[00121] In operation, the computing system 800 may execute instructions to acquire electrical signals, process the signals to derive feature values, determine operating states of a machine, and communicate results to other components within the system. The computing system 800 may operate independently or as part of a distributed architecture including edge and remote systems, as described herein.
[00122] Many alterations and modifications of the present invention will no doubt become apparent to a person of ordinary skill in the art after having read the foregoing description. It is to be understood that the phraseology or terminology employed herein is for the purpose of description and not of limitation. It is to be understood that the description above contains many specifications; these should not be construed as limiting the scope of the invention but as merely providing illustrations of some of the personally preferred embodiments of this invention. Thus, the scope of the invention should be determined by the appended claims and their legal equivalents rather than by the examples given. , Claims:CLAIMS
We claim:
1. A system (100) for monitoring operational behaviour of machine (102)s, the system (100) comprising:
an edge device (108) comprising a processor (204) and a database (110), the edge device (108) is configured to receive a current signal indicative of current associated with the machine (102), wherein the processor (204) is operatively coupled to a current sensor (104) connected to a power supply line of the machine (102); and
a remote server (122) communicatively coupled to the edge device (108), wherein the processor (204) is configured to:
sample the current signal at a predefined sampling interval to generate a stream of current samples;
filter the stream of current samples using a transient-exclusion filter to remove samples corresponding to spikes having a duration less than a predefined minimum duration, thereby generating filtered current samples;
measure one or more current magnitude values from the filtered current samples for a predefined duration;
apply a clustering technique to the measured current magnitude values to group recurring current patterns into a plurality of behaviour clusters, each behaviour cluster corresponding to an operational state of the machine (102); and
classify, using a trained machine learning classifier, an operating condition of the machine (102) into one of a plurality of operating states.

2. The system (100) as claimed in claim 1, wherein the processor (204) is further configured to:
receive machine (102)-related input data corresponding to the machine (102);
generate at least one machine (102) identifier corresponding to the machine (102) and at least one gateway identifier corresponding to the edge device (108);
map the machine (102) identifier to the gateway identifier to establish identification of the electrical machine (102) for processing of the stream of current samples; and
transmit the data to the remote server (122) together with at least one of the machines (102) identifiers or the gateway identifier to enable correlation of the transmitted data with the electrical machine (102).

3. The system (100) as claimed in claim 1, wherein the processor (204) is configured to determine the operating condition of the machine (102), wherein the processor (204) is further configured to:
receive one or more digital input signals from the machine (102), wherein each of the digital input signals includes one of two discrete states;
map the discrete states of the digital input signals to corresponding operating conditions based on predefined mapping conditions;
compare the current signal with a plurality of predefined threshold levels, wherein the plurality of predefined threshold levels comprises an upper threshold, a lower threshold, and one or more intermediate threshold levels;
determine the operating condition according to a range within which the current signal falls relative to the plurality of predefined threshold levels;
apply a hysteresis condition by defining, for each operating condition, an entry threshold, and an exit threshold; and
switch the operating condition only when the current signal crosses the entry threshold and remains beyond the exit threshold for at least a predefined duration.

4. The system (100) as claimed in claim 1, wherein the processor (204) is configured to determine the operating condition of the machine (102) based on clustering of the current magnitude values, wherein the processor (204) is further configured to:
group the current magnitude values into the plurality of clusters using a K-Means clustering technique, wherein each cluster is defined by a centroid and corresponds to a bounded interval of current magnitude values determined based on distances of the current magnitude values from the centroid;
assign each of the current magnitude values to one of the plurality of clusters based on a distance between the current magnitude values and corresponding cluster centroids; and
determine the operating condition of the electrical machine (102) based on the cluster to which the current magnitude values are assigned.

5. The system (100) as claimed in claim 4, wherein the processor (204) is further configured to:
assign each current magnitude value to one of the plurality of clusters based on a minimum distance between the current magnitude value and the corresponding cluster centroids;
generate a cluster identifier corresponding to the assigned cluster; and
determine the operating condition of the electrical machine (102) in real-time based on the cluster identifier.

6. The system (100) as claimed in claim 1, wherein the machine (102) comprises a multi-phase machine (102), including a three-phase machine (102), wherein the sensor (104) is disposed on any one of the single-phase power supply lines associated with the machine (102), wherein the processor (204) is configured to:
acquire the stream of current samples corresponding to the current signal from the single-phase power supply line;
determine the operating condition of the machine (102) based on the stream of current samples acquired from the single-phase power supply line;
determine that the plurality of operating states derived from the stream of current samples acquired from the single-phase power supply line corresponds to the plurality of operating states of the multi-phase machine (102), wherein the plurality of operating states includes a surge state, wherein the surge state corresponds to the current signal exceeding a predefined surge threshold for the predefined duration.

7. The system (100) as claimed in claim 1, wherein the processor (204) is configured to:
record a timestamp corresponding to the stream of current samples, wherein the timestamp comprises a date and a time value associated with acquisition of the stream of current samples;
link the timestamp with at least one of the current magnitude values, the operating condition, or the transmitted data, wherein the timestamp enables identification of the occurrence of the operating condition with respect to a specific date and time.

8. The system (100) as claimed in claim 1, wherein the processor (204) is configured to:
determine the operating condition of the machine (102) based on variation in the current signal independent of a plurality of predefined threshold levels;
compute a measure of variation in the current signal over a predefined time window;
set the operating condition to a running state when the measure of variation exceeds a predefined variation value over the predefined time window;
set the operating condition to an idle state when the measure of variation remains below the predefined variation value over the predefined time window; and
set the operating condition to a surge state when the current signal exceeds a predetermined surge threshold continuously for at least a predetermined duration, wherein determination of the operating condition is independent of predefined threshold levels, model training, and historical data storage.

9. The system (100) as claimed in claim 8, wherein the processor (204) is configured to:
determine the operating condition of the machine (102) based on a measure of variation in the current signal computed over a predefined time window, wherein the measure of variation represents a magnitude of change in the current signal within the predefined time window;
set the operating condition to a running state when the measure of variation indicates continuous fluctuation in the current signal over the predefined time window;
set the operating condition to an idle state when the measure of variation indicates stability in the current signal over the predefined time window;
set the operating condition to a surge state when the current signal exceeds a predetermined surge threshold for at least a predetermined duration; and
determine the operating condition based on the measure of variation independent of absolute current magnitude values for the running state and the idle state.

10. The system (100) as claimed in claim 1, wherein the processor (204) is configured to train a classifier, wherein the classifier is configured to determine the operating condition of the electrical machine (102) based on relationships between the current magnitude values derived from the filtered current samples and the operating condition of the electrical machine (102), wherein the training includes:
identifying a production cycle from the stream of current samples based on variation in the current signal corresponding to a start and an end of operation of the electrical machine (102);
determining one or more current magnitude values for the production cycle from the filtered current samples;
linking the current magnitude values with a corresponding operating condition of the electrical machine (102) to form stored training data;
storing the training data corresponding to a plurality of production cycles as historical data; and
training the classifier using the historical data to establish a relationship between the current magnitude values and the operating condition, wherein the edge device (108) is further configured to determine the operating condition for a subsequent production cycle using the trained classifier and the current magnitude values derived from the subsequent production cycle.

11. A computer-implemented method for monitoring and determining operational states of machine (102)s based on electrical current analysis, wherein the method comprises:
in an edge computing device comprising a processor (204) and a database (110) module:
sampling the current signal at a predefined sampling interval to generate a stream of current samples;
filtering the stream of current samples using a transient-exclusion filter to remove samples corresponding to spikes having a duration less than a predefined minimum duration, thereby generating filtered current samples;
measuring one or more current magnitude values from the filtered current samples for a predefined duration;
applying a clustering technique to the measured current magnitude values to group recurring current patterns into a plurality of behaviour clusters, each behaviour cluster corresponding to an operational state of the machine (102); and
classifying, using a trained machine learning classifier, an operating condition of the machine (102) into one of a plurality of operating states.

12. The computer-implemented method as claimed in claim 11, further comprising:
receiving machine (102)-related input data corresponding to the machine (102);
generating at least one machine (102) identifier corresponding to the machine (102) and at least one gateway identifier corresponding to the edge device (108);
mapping the machine (102) identifier to the gateway identifier to establish identification of the electrical machine (102) for processing of the stream of current samples; and
transmitting the data to the remote server (122) together with at least one of the machines (102) identifiers or the gateway identifier to enable correlation of the transmitted data with the electrical machine (102).

13. The computer-implemented method as claimed in claim 11, wherein determining the operating condition of the machine (102) further comprises:
receiving one or more digital input signals from the machine (102), wherein each of the digital input signals includes one of two discrete states;
mapping the discrete states of the digital input signals to corresponding operating conditions based on predefined mapping conditions;
comparing the current signal with a plurality of predefined threshold levels, wherein the plurality of predefined threshold levels comprises an upper threshold, a lower threshold, and one or more intermediate threshold levels;
determining the operating condition according to a range within which the current signal falls relative to the plurality of predefined threshold levels;
applying a hysteresis condition by defining, for each operating condition, an entry threshold, and an exit threshold; and
switching the operating condition only when the current signal crosses the entry threshold and remains beyond the exit threshold for at least a predefined duration.

14. The computer-implemented method as claimed in claim 11, wherein determining the operating condition of the machine (102) based on clustering of the current magnitude values further comprises:
grouping the current magnitude values into the plurality of clusters using a K-Means clustering technique, wherein each cluster is defined by a centroid and corresponds to a bounded interval of current magnitude values determined based on distances of the current magnitude values from the centroid;
assigning each of the current magnitude values to one of the plurality of clusters based on a distance between the current magnitude values and corresponding cluster centroids; and
determining the operating condition of the electrical machine (102) based on the cluster to which the current magnitude values are assigned.

15. The computer-implemented method as claimed in claim 14, further comprising:
assigning each current magnitude value to one of the plurality of clusters based on a minimum distance between the current magnitude value and the corresponding cluster centroids;
generating a cluster identifier corresponding to the assigned cluster; and
determining the operating condition of the electrical machine (102) in real-time based on the cluster identifier.

16. The computer-implemented method as claimed in claim 11, wherein the machine (102) comprises a multi-phase machine (102), including a three-phase machine (102), wherein the sensor (104) is disposed on any one of the single-phase power supply lines associated with the machine (102), the method further comprising:
acquiring the stream of current samples corresponding to the current signal from the single-phase power supply line;
determining the operating condition of the machine (102) based on the stream of current samples acquired from the single-phase power supply line; and
determining that the operating condition derived from the stream of current samples acquired from the single-phase power supply line corresponds to an operating condition of the multi-phase electrical machine (102).

17. The computer-implemented method as claimed in claim 11, further comprising:
recording a timestamp corresponding to the stream of current samples, wherein the timestamp comprises a date and a time value associated with acquisition of the stream of current samples; and
linking the timestamp with at least one of the current magnitude values, the operating condition, or the transmitted data, wherein the timestamp enables identification of the occurrence of the operating condition with respect to a specific date and time.

18. The computer-implemented method as claimed in claim 11, further comprising:
determining the operating condition of the machine (102) based on variation in the current signal independent of a plurality of predefined threshold levels;
computing a measure of variation in the current signal over a predefined time window;
setting the operating condition to a running state when the measure of variation exceeds a predefined variation value over the predefined time window;
setting the operating condition to an idle state when the measure of variation remains below the predefined variation value over the predefined time window; and
setting the operating condition to a surge state when the current signal exceeds a predetermined surge threshold continuously for at least a predetermined duration, wherein determination of the operating condition is independent of predefined threshold levels, model training, and historical data storage.

19. The computer-implemented method as claimed in claim 18, further comprising:
determining the operating condition of the machine (102) based on a measure of variation in the current signal computed over a predefined time window, wherein the measure of variation represents a magnitude of change in the current signal within the predefined time window;
setting the operating condition to a running state when the measure of variation indicates continuous fluctuation in the current signal over the predefined time window;
setting the operating condition to an idle state when the measure of variation indicates stability in the current signal over the predefined time window;
setting the operating condition to a surge state when the current signal exceeds a predetermined surge threshold for at least a predetermined duration; and
determining the operating condition based on the measure of variation independent of absolute current magnitude values for the running state and the idle state.

20. The computer-implemented method as claimed in claim 11, further comprising training the machine (102) learning classifier, wherein the machine (102) learning classifier is configured to determine the operating condition of the electrical machine (102) based on relationships between the current magnitude values derived from the filtered current samples and the operating condition of the electrical machine (102), wherein the training comprises:
identifying a production cycle from the stream of current samples based on variation in the current signal corresponding to a start and an end of operation of the electrical machine (102);
determining one or more current magnitude values for the production cycle from the filtered current samples;
linking the current magnitude values with a corresponding operating condition of the electrical machine (102) to form stored training data;
storing the training data corresponding to a plurality of production cycles as historical data; and
training the machine (102) learning classifier using the historical data to establish a relationship between the current magnitude values and the operating condition, wherein the method further comprises determining the operating condition for a subsequent production cycle using the trained classifier and the current magnitude values derived from the subsequent production cycle.

21. A non-transitory computer-readable medium having stored thereon, computer- executable instructions which, when executed by a computer, cause the computer to execute operations, the operations comprising:
sampling the current signal at a predefined sampling interval to generate a stream of current samples;
filtering the stream of current samples using a transient-exclusion filter to remove samples corresponding to spikes having a duration less than a predefined minimum duration, thereby generating filtered current samples;
measuring one or more current magnitude values from the filtered current samples for a predefined duration;
applying a clustering technique to the measured current magnitude values to group recurring current patterns into a plurality of behaviour clusters, each behaviour cluster corresponding to an operational state of the machine (102); and
classifying, using a trained machine learning classifier, an operating condition of the machine (102) into one of a plurality of operating states.