Claims:1. A method, comprising:
receiving machine data corresponding to a fleet of machines, wherein the machine data comprises operational data, non-operational data, and historical data corresponding to the fleet of machines;
generating an operational data model corresponding to a fault condition of a machine among the fleet of machines based on the historical data;
generating a machine learning model corresponding to the fleet of machines based on the non-operational data;
determining a fault detection model based on the operational data model and the machine learning model; and
determining a life duration of the machine based on the fault detection model and the operational data and the non-operational data.
2. The method of claim 1, wherein generating the operational data model comprises determining a physics based mathematical model.
3. The method of claim 1, further comprising receiving the non-operational data from a plurality of databases, wherein the plurality of databases comprises manufacturing data, inspection data, usage data, environment data, geographic data, customer data, and expert opinion data, or combinations thereof.
4. The method of claim 3, wherein determining the fault detection model further comprises calibrating the fault detection model based on the inspection data.
5. The method of claim 3, wherein generating the machine learning model comprises analyzing the non-operational data based on a machine learning technique.
6. The method of claim 5, wherein analyzing the non-operational data comprises integrating data from the plurality of databases.
7. The method of claim 5, wherein analyzing the non-operational data comprises determining a plurality of data clusters of the fleet of machines based on an unsupervised learning technique.
8. The method of claim 7, further comprising generating a plurality of fault detection models corresponding to the fault condition based on the plurality of data clusters and the operational data model.
9. The method of claim 5, wherein analyzing the non-operational data comprises determining a plurality of feature parameters of the fleet of machines based on a supervised learning technique.
10. The method of claim 9, wherein determining the fault detection model comprises:
determining the plurality of key feature parameters of the fleet of machines; and
combining the key parameters with the operational data model
11. The method of claim 1, further comprising communicating the life duration to an operator.
12. A system, comprising:
a data acquisition unit configured to receive machine data corresponding to a fleet of machines, wherein the machine data comprises operational data, non-operational data and historical data corresponding to the fleet of machines;
a model generator unit communicatively coupled to the data acquisition unit and configured to generate an operational data model corresponding to a fault condition of a machine among the fleet of machines based on the historical data;
a machine learning unit communicatively coupled to the data acquisition unit and configured to generate a machine learning model corresponding to the fleet of machines based on the non-operational data;
a fault management unit communicatively coupled to the model generating unit, the machine learning unit, and the data acquisition unit and configured to:
determine a fault detection model based on the operational data model and the machine learning model; and
determine a life duration of the machine based on the fault detection model and the operational data and the non-operational data.
13. The system of claim 12, wherein the machine learning unit is further configured to receive data from a plurality of databases, and wherein the plurality of databases comprises a manufacturing database, an inspection database, a usage database, an environment database, a geographic database, a customer database, an expert opinion database, or combinations thereof.
14. The system of claim 13, wherein the machine learning unit is further configured to calibrate the fault detection model based on inspection data obtained from the inspection database.
15. The system of claim 13, wherein the machine learning unit is configured to analyze the non-operational data based on a machine learning technique.
16. The system of claim 15, wherein the machine learning unit is further configured to determine a plurality of data clusters of the fleet of machines based on an unsupervised learning technique.
17. The system of claim 16, wherein the fault management unit is further configured to generate a plurality of fault detection models corresponding to the fault condition based on the plurality of data clusters and the operational data model.
18. The system of claim 15, wherein the machine learning unit is further configured to determine a plurality of feature parameters of the fleet of machines based on a supervised learning technique.
19. The system of claim 18, wherein the fault management unit is further configured to:
determine the plurality of key feature parameters of the fleet of machines; and
combine the key parameters with the operational data model.
, Description:BACKGROUND
[0001] Embodiments of the present specification relate generally to life management of a machine among a fleet of machines, and more specifically to estimation of remaining life of the machine based on machine learning techniques.
[0002] Accurate and cost-effective maintenance planning of a machine such as a turbine engine and an aircraft engine calls for accurate prediction of lifespan of engine components based on acute degradations and failures modes. In many cases, the machine is part of the fleet of machines or one of a plurality of near-identical machines operating under similar or varied operating conditions. Conventional maintenance practices are based on collection of operational data from various sensors and inspection of the engines during outages or maintenance. The operational data is compared with data patterns associated with known lifespan durations determined at a design stage for arriving at maintenance related decisions. However, the progression of the failure modes as determined at the design stage may not match accurately with the actual progression of failure modes in the field. This mismatch may result in increased costs to the engine manufacturer, reduced availability of the engines, and reduced revenue to the engine owner.
[0003] Currently available solutions for determining the lifespan of the engine components entail use of diagnostics, prognostics, or health monitoring. Diagnostics, prognostics and/or health monitoring rely primarily on data collected from on-board sensors and fusion algorithms to combine the sensor data. Poor sensor performance may yield an inaccurate picture of the state of the engine component. Also, the sensor data may fail to accurately convey the underlying physics of any degradation of the engine components. Moreover other parameters not captured by on-board sensors play an active role on part damage. Some of these factors include geographical parameters (like location in coastal regions v/s location in deserts), environmental factors (like dust particle size, salinity content, sulphate content, fuel quality), manufacturing factors (like manufacturing vendor). Many of these factors are very difficult to translate in terms of damage as they need very complex physics based analytical models.
[0004] Other component lifespan-prediction techniques combine design models, remote monitoring and diagnostic (RMD) data, and inspection data into a single, probabilistic, total lifespan predictive model to calibrate and improve the predictions of the failure modes. However, inspection data corresponding to the machine may not be available immediately after installation. Although the quantum of data available increases with time, determining an effective model for a failure mode of a component may not be effective with the available data. Further, validation of the failure models may be difficult in the absence of actual failure data. In addition, it is desirable to schedule the inspection of critical components of the machine based on times associated with the predicted failure of the components to minimize the time the machine is taken off-line for inspection.
BRIEF DESCRIPTION
[0005] In accordance with one aspect of the present specification, a method is disclosed. The method includes receiving machine data corresponding to a fleet of machines. The machine data comprises operational data, non-operational data, and historical data corresponding to the fleet of machines. The method further includes generating an operational data model corresponding to a fault condition of a machine among the fleet of machines based on the historical data. The method includes generating a machine learning model corresponding to the fleet of machines based on the non-operational data. The method also includes determining a fault detection model based on the operational data model and the machine learning model. The method includes determining a life duration of the machine based on the fault detection model and the operational data and the non-operational data.
[0006] In accordance with another aspect of the present specification, a system is disclosed. The system includes a data acquisition unit configured to receive machine data corresponding to a fleet of machines. The machine data comprises operational data, non-operational data and historical data corresponding to the fleet of machines. The system further includes a model generator unit communicatively coupled to the data acquisition unit and configured to generate an operational data model corresponding to a fault condition of a machine among the fleet of machines based on the historical data. The system also includes machine learning unit communicatively coupled to the data acquisition unit and configured to generate a machine learning model corresponding to the fleet of machines based on the non-operational data. The system further includes a fault management unit communicatively coupled to the model generating unit, the machine learning unit, and the data acquisition unit and configured to determine a fault detection model based on the operational data model and the machine learning model. The fault management unit is also configured to determine a life duration of the machine based on the fault detection model and the operational data and the non-operational data.
DRAWINGS
[0007] These and other features and aspects of embodiments of the present invention will become better understood when the following detailed description is read with reference to the accompanying drawings in which like characters represent like parts throughout the drawings, wherein:
[0008] FIG. 1 illustrates a system for maintaining a machine among a plurality of machines, in accordance with aspects of the present specification;
[0009] FIG. 2 is a schematic representation of sources of data used for fault detection, in accordance with aspects of the present specification;
[0010] FIG. 3 is a flow chart of a method for determining a life of a machine, in accordance with aspects of the present specification;
[0011] FIG. 4 is a flow chart of a method for determining a plurality of data clusters, in accordance with aspects of the present specification;
[0012] FIG. 5 is a flow chart of a method for determining a plurality of feature parameters, in accordance with aspects of the present specification;
[0013] FIGs. 6(a)-6(b) are schematics illustrating performance of a machine learning based fault detection, in accordance with aspects of the present specification; and
[0014] FIG. 7 is a flow chart of a method for maintaining a machine among a plurality of machines, in accordance with aspects of the present specification.
DETAILED DESCRIPTION
[0015] The term ‘machine learning’ refers to an artificial intelligence technique that employs statistical methods to study and develop data processing algorithms and data modelling. Also, the machine learning technique is used to learn from and make predictions on the data. Moreover, the term ‘supervised learning’ refers to a machine learning technique that uses a set of input datasets and corresponding output datasets to determine a data model or an algorithm for performing numerous tasks related to estimation, prediction, and analysis. Furthermore, the term ‘unsupervised learning’ refers to a machine learning technique that determines a pattern in a given dataset. The term ‘life duration’ refers to a time estimate of remaining life of a component of a machine or a machine in a fleet of machines. The term ‘feature’ is used interchangeably with the term ‘feature parameter’ to denote an important parameter among a plurality of parameters associated with the machine data. In addition, the term ‘data cluster’ refers to a subset of machine data that has some common characteristics. The machine data is typically viewed as including a plurality of data clusters.
[0016] FIG. 1 is a diagrammatic representation of a system 100 for maintaining a machine among a fleet of machines 102, in accordance with aspects of the present specification. The fleet of machines 102 includes a plurality of similar machines 104, 106 operating in similar or varied operating conditions. Each machine 104, 106 may include a plurality of components 108. In one embodiment, the fleet of machines 102 corresponds to a plurality of turbines deployed in a plurality of locations around the globe. In another embodiment, the fleet of machines 102 corresponds to a plurality of aircraft engines flying across different continents.
[0017] It may be desirable to accurately determine the life of the components 108 in each machine 104, 106 in the fleet of machines 102. In accordance with aspects of the present specification, the system 100 is configured to determine the remaining life of the plurality of components in the fleet of machines 102. In accordance with aspects of the present specification, the system 100 receives machine data 114 from the fleet of machines 102 and generates an indicator representative of a life duration 122 of one or more components in the fleet of machines 102.
[0018] In a presently contemplated configuration, the system 100 includes a data acquisition unit 124, a model generator unit 126, a machine learning unit 128, a fault management unit 130, a memory unit 132, and a processor unit 134 communicatively connected with each other by a communication bus 112. The data acquisition unit 124 is communicatively coupled to the fleet of machines 102 and configured to receive the machine data 114. In one embodiment, the machine data 114 includes operational data, non-operational data and historical data. The operational data is representative of a plurality of operating parameters such as phase currents and phase voltages. Also, the non-operational data is representative of a plurality of parameters related to manufacturing data, environment data, geographic data, usage data, and inspection data. In addition, the historical data is representative of previously acquired operational and non-operational data corresponding to the fleet of machines 102. In another embodiment, the data acquisition unit 124 is configured to receive machine data 114 having operational data and non-operational data, store the machine data in the memory unit 132, and provide the stored machine data 116 at a future instant of time.
[0019] The model generator unit 126 is communicatively coupled to the data acquisition unit 124 and configured to determine an operational data model corresponding to a given fault condition of a machine based on physics-based mathematical models and historical data. The operational data model is representative of a failure mode of a component 108 of the machine 106. Accordingly, the model generator unit 126 may be configured to generate a plurality of models corresponding to a plurality of fault conditions for one or more components 108 of the machine 106.
[0020] Furthermore, the machine learning unit 128 is communicatively coupled to the data acquisition unit 124 and configured to generate a machine learning model corresponding to the fleet of machines 102 based on the non-operational data. In one embodiment, the machine learning unit 128 is configured to generate a plurality of feature parameters based on a supervised learning technique. In another embodiment, the machine learning unit 128 is configured to generate the plurality of feature parameters based on an unsupervised learning technique.
[0021] Moreover, the fault management unit 130 is communicatively coupled to the model generator unit 126 and the machine learning unit 128. The fault management unit 130 is configured to determine a fault detection model based on the operational data model 118 and the machine learning model 120 respectively generated by the model generation unit 126 and the machine learning unit 128. The fault management unit 130 is further configured to determine a life duration of the machine 106 based on the fault detection model and the plurality of operational data and non-operational data.
[0022] In one embodiment, the fault management unit 130 determines a damage index representative of the extent of damage of one or more components of a machine using the fault detection model. Further, the fault management unit 130 also determines a rate of damage corresponding to the damage index. The rate of damage may be representative of a change in damage of a component with respect to time. Further, the fault management unit 130 determines a time duration representative of remaining life based on the damage index and the rate of damage using extrapolation techniques. In another embodiment, the fault management unit 130 determines a first damage index at a first time instant and a second damage index at a second instant of time. The fault management unit 130 determines the remaining life duration based on the first damage index, the second damage index, the first time instant and the second time instant.
[0023] The processor unit 134 is communicatively coupled to the fault management unit 130 and may include one or more of an arithmetic logic unit, a microprocessor, a general purpose controller, and a processor array to perform desired computations. In one embodiment, the processor unit 134 may be configured to perform tasks associated with one or more of the data acquisition unit 124, the processor unit 134, the model generator unit 126, the machine learning unit 128, and the fault management unit 130. While the processor unit 134 is shown as a single unit, in certain embodiments, the processor unit 134 may include more than one processor co-located or distributed in different locations.
[0024] In addition, the memory unit 132 is communicatively coupled to the processor unit 134. In certain embodiments, the memory unit 132 may include a plurality of memory subunits. The memory unit 132 may be a non-transitory storage medium. For example, the memory may be a dynamic random access memory (DRAM) device, a static random access memory (SRAM) device, flash memory or other memory devices. In one embodiment, the memory may include a non-volatile memory or similar permanent storage device, media such as a hard disk drive, a floppy disk drive, a compact disc read only memory (CD-ROM) device, a digital versatile disc read only memory (DVD-ROM) device, a digital versatile disc random access memory (DVD-RAM) device, a digital versatile disc rewritable (DVD-RW) device, a flash memory device, or other non-volatile storage devices. In one embodiment, a non-transitory computer readable medium may be encoded with a program to instruct the processor to perform functions of one or more of the data acquisition unit 124, the model generator unit 126, the machine learning unit 128, and the fault management unit 130.
[0025] FIG. 2 is a schematic 200 illustrating data flow in the system 100, in accordance with aspects of present specification. The data flow of FIG. 2 will be described with reference to the components of FIG. 1. The machine data 114 that corresponds to the fleet of machines 102 comprises of operational data 202, non-operational data 204 and historical data 206. The operational data 202 is representative of a plurality of parameters acquired from the fleet of machines 102 continuously via sensors. In one example, the operational data 202 includes a plurality of phase currents and a plurality of phase voltages corresponding to the machine 106 in the fleet of machines 102. In another embodiment, the operational data 202 includes a plurality of flight parameters such as altitude, temperature, speed, and humidity. It may be noted that the plurality of flight parameters is measured in real time.
[0026] The non-operational data 204 includes data acquired from a plurality of databases 218. Each database in the plurality of databases includes data corresponding to one or more machines in the fleet of machines 102. The plurality of databases 218 includes, but is not limited to, a manufacturing database 220, an inspection database 222, an usage database 224, an environment database 226, an expert database 228, a geographical database 230 and a customer database 232. The manufacturing database 220 includes design parameters and tolerances, machine specifications, vendor/component historical reliability corresponding to the fleet of machines 102 and other manufacturing related data. The inspection database 222 includes data generated during inspection and/or scheduled servicing of the plurality of machines in the fleet of machines 102. Furthermore, the environment database 226 includes parameters related to the environmental conditions of the fleet of machines 102. The expert database 228 includes patterns that are observed in the other databases and decisions related to failure modes of the components 108 in the machine 106. The geographical database 230 includes data related to a geographical region where the machine 106 or the fleet of machines 102 is deployed during a normal course of operation. In addition, the customer database 232 includes information regarding customers like fuel quality checks, operating procedures, and historical reliability while using the machine 106. The usage database 224 includes data related to operating regimes of the machines 106 in the fleet of machines 102. In some embodiments, the plurality of databases 218 includes internal and proprietary databases related to repair shop reports, design reviews, life data management, and diagnostics data, and the like.
[0027] Furthermore, as previously noted with reference to FIG. 1, the machine learning unit 128 of FIG. 1 is configured to generate the machine learning model 120 based on the non-operational data 204 corresponding to a failure mode of a component 108 or the machine 106. In one embodiment, the machine learning model 120 is determined based on an analysis of the non-operational data 204 using machine learning algorithms. By way of example, the analysis of non-operational data may entail fusing data corresponding to the plurality of databases 218 to generate combined data. Moreover, exploratory data analysis of the combined data may be used to generate a data pattern. However, in certain embodiments, the exploratory data analysis of each of the databases may precede the data fusion operation of the plurality of databases 218.
[0028] In one embodiment, the exploratory data analysis may be performed on each of the plurality of databases 220, 222, 224, 226, 228, 230, 232. In another embodiment, the exploratory data analysis may be performed on a subset of the plurality of databases 218. Furthermore, in certain embodiments, plurality of database parameters corresponding to each of the databases may be generated consequent to the exploratory data analysis. In another embodiment, the exploratory data analysis may generate a plurality of datasets corresponding to each of the databases. Moreover, in one example, the data fusion operation may be performed by integrating data from the plurality of databases 218. In other embodiments, the data fusion operation may be performed by concatenating the plurality of database parameters corresponding to the plurality of databases 218. In yet another embodiment, the data fusion operation may be performed by combining the plurality of datasets corresponding to the plurality of databases 218.
[0029] In accordance with further aspects of the present specification, a supervised machine learning technique is employed to generate the machine learning model 120. However, in certain other embodiments, an unsupervised learning technique may be employed to generate the machine learning model 120. Use of the supervised learning technique entails the generation of the machine learning model 120 based on a plurality of input datasets, and corresponding output datasets from the fleet of machines 102. The input datasets and the output datasets are data selected from one of the plurality of databases. It may be noted that use of the unsupervised learning technique to generate the machine learning model 120 calls for use of statistical techniques including, but not limited to, clustering, method of moments, and matrix factorization techniques.
[0030] Furthermore, a plurality of data clusters in the non-operational data corresponding to the fleet of machines 102 may be determined based on an unsupervised learning technique. A plurality of operational data models corresponding to a fault condition is determined based on the plurality of data clusters. In another embodiment, a plurality of vital feature parameters is determined from the non-operational data 204 using a supervised learning technique. A classifier model is generated based on the plurality of feature parameters. The operational data model is modified based on the classifier model.
[0031] Historical data 206 is generated by storing operational data 202 and the non-operational data 204 in a memory unit such as the memory unit 132 of FIG. 1. In one embodiment, a time delay element 208 may be used to generate and store the historical data 206. The historical data may include data related to the component 108, the machine 106, the fleet of machines 102 or combinations thereof. The historical data 206 would be used in combination with real-time operational data, 202 as inputs into the operational data model 118. As previously noted, the operational data model 118 is used to determine the damage state for different failure modes of the component 108 and/or the machine 106. The operational data model 118 includes a physics based mathematical model or a simulation model capable of generating the physical state of the plurality of components, 108 in the plurality of machines 106 in the fleet of machines 102 using the operational data 202 and historical data, 206.
[0032] A fault detection model 210 is determined based on the operational data model 118 and the machine learning model 120. In one embodiment, the fault detection model 210 is calibrated based on the inspection data. The calibration of the fault detection model involves determining differences between an outcome of an inspection and a corresponding output of the fault detection model. The calibration further involves modifying the fault detection model when the differences are not negligible. A life duration122 of the machine 106 is determined based on the fault detection model 210. Furthermore, the life duration 122 may be communicated to an operator.
[0033] FIG. 3 is a flow chart 300 of a method for determining a life of a machine, in accordance with aspects of present specification. The method includes receiving operational data, as indicated by step 302. The operational data includes data obtained from one or more of a component of a machine, the machine, and a fleet of machines during operation. The operational data also includes previously acquired data and the presently acquired data.
[0034] At step 304, a fault model is generated based on the operational data. In one embodiment, the fault model is an operational data model representative of working of the machine and is determined based on the operational data. The fault model is also configured to indicate one or more faults of the machine. In one embodiment, the fault model may be generated based on underlying physical principles that use the operational data. The fault model may be in the form of a simulation model, an executable program, or a set of mathematical equations. Also, inspection data is generated during regular inspection and maintenance schedules, as depicted by step 306. Further, as indicated by step 308, a machine learning model is generated based on non-operational data. Subsequently, at step 310, the fault detection model generated at 304 is modified based on the machine learning model generated at step 308 and the inspection data generated at step 306.
[0035] Further, at step 312, a fault in machine may be predicted using either or both the operational data and non-operational data. In particular, the fault may be predicted based on modified fault detection model. Additionally, at step 310, a damage index corresponding to the predicted fault is determined based on modified fault model. Moreover, at step 314, a life duration of the machine may be determined based on the damage index and the modified fault model. Further, the life duration of the machine is communicated to a user, as depicted by step 316.
[0036] FIG. 4 is a flow chart 400 of a method for determining an enhanced fault detection model, in accordance with aspects of present specification. The method includes integrating non-operational data from a plurality of databases to generate fused data, as depicted by step 402. Furthermore, at step 404, an exploratory data analysis of the fused non-operational data is performed to extract a plurality of parameters from the fused data.
[0037] Moreover, at step 406, a plurality of feature parameters is selected from the plurality of parameters extracted at step 404. In one embodiment, a supervised learning technique may be used to select the plurality of feature parameters. Also, at step 408, an operational data model is generated based on operational data. In one embodiment, the operational data model may be generated using physical principles associated with the design of the machine and the generation of the operational data from the machine.
[0038] Subsequently, at step 410, a fault detection model is generated based on the plurality of feature parameters selected at the step 406 and the operational data model generated at step 408. Further, at step 412, a fault is determined based on the fault detection model. In addition, at step 412, a corresponding damage index is determined. Further, in some embodiments, the fault condition at a future time instant may be predicted and a corresponding damage index is determined by extrapolation techniques. As indicated by step 414, a life duration of the machine is determined based on the damage index and/or the fault detection model. Further, the life duration of the machine is communicated to a user, as depicted by step 416.
[0039] Referring now to FIG. 5, a flow chart 500 of a method of determining a plurality of data clusters, in accordance with aspects of present specification, is depicted. The method includes integrating non-operational data from a plurality of databases to generate fused data, as indicated by step 502. Subsequently, at step 504, an exploratory data analysis of the fused non-operational data is performed.
[0040] Moreover, a plurality of data clusters is generated based on the non-operational data using an unsupervised learning technique, as indicated by step 506. Also, at step 508, an operational data model is generated based on operational data. In one embodiment, the operational data model may be generated using physical principles associated with the design of the machine and the generation of the operational data from the machine.
[0041] Additionally, at step 510, a fault detection model is determined based on the plurality of data clusters and the operational data model. In certain embodiments, a fault detection model corresponding to each of the plurality of data clusters may be generated.
[0042] Further, at step 512, a fault is determined based on the fault detection model. In addition, a corresponding damage index is also determined at step 512. The fault detection model and the damage index are employed to determine a life duration of the machine, as indicated by step 514. Moreover, the life duration of the machine is communicated to a user in step 516.
[0043] FIGs. 6(a) and 6(b) are schematics illustrating performance of a fault detection system, in accordance with aspects of present specification, presented on a set of 25 machine units, 6 of which had observed component faults at field inspection. An observation or model prediction of this fault is termed as fail status in these charts and absence of the fault is termed as no fail status in these charts. FIG. 6(a) is representative of one confusion matrix 602 that represents performance of an operational data model for fault detection Also, FIG. 6(b) is representative of another confusion matrix 604 that represents performance of a combined operational data and machine learning based model for fault detection.
[0044] The confusion matrix 602 of FIG. 6(a) includes an axis 606 corresponding to a prediction status and an axis 608 corresponding to an actual status of a machine, which was observed in the field inspection. Also, as depicted in the confusion matrix 602, six faults are correctly classified as faults, eleven cases are correctly classified as non-faults, and eight non-faulty cases are incorrectly classified as faults by the model. The prediction accuracy of the operational data model as shown by the confusion matrix 602 is about 68%. Probability of (Fault) Detection (POD) is 100% as all six faults have been classified correctly. The False Detection Rate (FDR) is about 42% as eight cases are incorrectly classified as faults among nineteen non-fault conditions.
[0045] Also, the confusion matrix 604 of FIG. 6(b) includes an axis 610 corresponding to a prediction status and an axis 612 corresponding to an actual status of the machine, which was observed in the field inspection. The confusion matrix 604 has six faults correctly classified as faults, eighteen cases correctly classified as non-faults, and one non-faulty case incorrectly classified as a fault. The prediction accuracy of the physics based model as shown by the confusion matrix 604 is about 96%. Probability of (Fault) Detection (POD) is 100% as all six faults are classified correctly as faults. The False Detection Rate (FDR) is about 5.3% as only one case is incorrectly classified as a fault among 19 non-fault conditions. Based on the confusion matrices 602, 604 depicted in FIGs. 6(a)-6(b), it may be observed that the performance of the combined operational data and machine learning based fault detection technique of FIG. 6(b) is much better than the performance of the conventional techniques that are based on only operational data models and depicted in FIG. 6(a).
[0046] FIG. 7 is a flow chart 700 of a method for maintaining a machine among a plurality of machines, in accordance with aspects of present specification. The method includes receiving machine data corresponding to a fleet of machines, as indicated by step 702. The machine data includes operational data, non-operational data, and historical data corresponding to the fleet of machines. The operational data corresponds to a plurality of operating parameters of the machine and a plurality of fleet related parameters measured during operation of the fleet of machines. Also, the non-operational data refers to manufacturing data, usage data, design data, inspection data, and environment data. It should be noted that the non-operational data is not limited to the types of data enumerated here. The historical data is representative of operational data and non-operational data stored in memory.
[0047] Further, an operational data model corresponding to a fault condition of a machine among the fleet of machines is generated, as indicated by step 704. The operational data model may be generated based on the historical data, in one embodiment. Also, the operational data model may be in the form of a mathematical equation, a simulation model, or a combination thereof. In one embodiment, the operational data model may be modified periodically based on the inspection data. In another embodiment, a predicted fault may be used to calibrate the operational data model.
[0048] Moreover, at step 706, a machine learning model corresponding to the fleet of machines is generated based on the non-operational data. In one embodiment, the machine learning model is based on a plurality of feature parameters corresponding to the non-operational data. In particular, the plurality of feature parameters may be selected/extracted from a plurality of parameters associated with the non-operational data. Also, the machine learning model may be determined using a supervised learning technique. In another embodiment, the machine learning model may be generated based on a plurality of data clusters that correspond to the non-operational data. In this example, the machine learning model may be determined using a unsupervised learning technique.
[0049] Subsequently, at step 708, a fault detection model is determined based on the operational data model and the machine learning model. In the embodiment where a supervised learning technique is used to generate the machine learning model, the plurality of feature parameters is combined with the operational data model to determine the fault detection model. In the embodiment where an unsupervised learning technique is used to generate the machine learning model, the plurality of data clusters is used to refine the operational data model to determine the fault detection model. Specifically, when the plurality of data clusters is available, the fault detection model is determined by refining the operational data model with reference to each of the plurality of data clusters. A life duration of the machine is determined based on the fault detection model and the operational data, as indicated by step 710. Also, at step 712, the life duration determined at step 710 is communicated to a user. In some embodiments, the life duration of the machine may be communicated to the user/operator via use of a display, an audio signal, a video signal, an alarm, and the like.
[0050] The systems and methods disclosed hereinabove provide an enhanced prediction of life of a component in a machine. In particular, the prediction of the life duration is based on a fault detection model corresponding to a given component of a machine. The fault detection model is determined based on the operational data and non-operational data corresponding to a given machine, thereby enhancing the prediction of the life duration of that component or machine. A life duration representative of expended damage or remaining life and relative risk across components in a fleet of machines is determined based on the fault detection model. The predicted life duration is used in to perform timely condition-based maintenance. Accurate prediction of the life duration provides opportunities for optimizing the operation of the fleet of machines and for extending the life of components of the machine.
[0051] Not necessarily all such objects or advantages described above may be achieved in accordance with any particular embodiment. Thus, for example, those skilled in the art will recognize that the systems and techniques described herein may be embodied or carried out in a manner that achieves or improves one advantage or group of advantages as taught herein without necessarily achieving other objects or advantages as may be taught or suggested herein.
[0052] While the technology has been described in detail in connection with only a limited number of embodiments, it should be readily understood that the specification is not limited to such disclosed embodiments. Rather, the technology can be modified to incorporate any number of variations, alterations, substitutions or equivalent arrangements not heretofore described, but which are commensurate with the spirit and scope of the claims. Additionally, while various embodiments of the technology have been described, it is to be understood that aspects of the specification may include only some of the described embodiments. Accordingly, the specification is not to be seen as limited by the foregoing description, but is only limited by the scope of the appended claims.