Claims:
1. A processor implemented method (600) of monitoring an industrial plant, comprising:
collecting (602) real-time and non-real-time data from a plurality of data sources for each of a plurality of components of the industrial plant, as input, via one or more hardware processors; and
performing health monitoring of each component of interest from among the plurality of components, via the one or more hardware processors, comprises:
determining (606) a plurality of feature importance metrics of a plurality of sensors of each of a plurality of preceding components of the component of interest;
identifying all the preceding components that are contributing for degradation of the component of interest, from among the plurality of preceding components, by comparing the feature importance metrics of the plurality of sensors of each preceding component with a plurality of threshold values, wherein the component of interest is identified to have a dependency on the preceding components that contribute to the degradation of the component of interest;
estimating (608) Remaining Useful Life (RUL) of the component of interest, based on the components that have been identified as contributing to the degradation of the component of interest, using a self-ensemble model; and
detecting (610) a change point of the component of interest, based on the components that have been identified as contributing to the degradation of the component of interest, using the self-ensemble model.

2. The method as claimed in claim 1, wherein determining the dependency comprises:
performing for each component preceding the component of interest:
identifying the feature importance metric of each of the plurality of sensors associated with the preceding component;
generating a data-driven model using data from the plurality of sensors;
determining accuracy of the data driven model and comparing the determined accuracy with a first threshold;
labeling the preceding component as a component contributing for the degradation of the component of interest if the determined accuracy exceeds the first threshold;
comparing the determined feature importance metric of each of the plurality of sensors with a second threshold, and selecting all sensors from among the plurality of sensors for which the determined feature importance metric is greater than the second threshold;
comparing the determined feature importance metric of each of the plurality of sensors for which the determined accuracy is less than the first threshold, with a third threshold, and selecting all sensors for which the feature importance metric is more than the third threshold; and
labeling the preceding component as a component contributing for the degradation of the component of interest, if the feature importance metric of at least one of the plurality of sensors exceeds the third threshold.

3. The method as claimed in claim 1, wherein obtaining the self-ensemble model comprises:
fetching (702) a plurality of available prediction models of each component of interest;
fetching (704) at least one model selection criteria, for a RUL estimation requirement, as an input;
selecting (706) all prediction models from the available prediction models that satisfy the fetched selection criteria;
tuning (708) a plurality of parameters of each of the plurality of ensemble methods with each of the selected predictions models and generating predictions using each of the tuned ensemble methods;
comparing (710) the generated predictions with a measured value of predictions to determine an error value between the predicted value and the measured value;
selecting (716) all ensemble methods having the error value less than an error threshold, as candidate ensemble methods, from the plurality of ensemble methods; and
selecting (718) an ensemble method from among the candidate ensemble methods, based on the selection criteria.

4. A system (102) for monitoring an industrial asset or component, comprising:

one or more hardware processors;
a communication interface; and
a memory coupled to the one or more hardware processors via the communication interface, wherein a plurality of instructions stored in the memory when executed, cause the one or more hardware processors to:
collect real-time and non-real-time data from a plurality of data sources for each of a plurality of components of the industrial plant, as input; and
perform health monitoring of each component of interest from among the plurality of components, by:
determining a plurality of feature importance metrics of a plurality of sensors of each of a plurality of preceding components of the component of interest, via the one or more hardware processors;
identifying all the preceding components that are contributing for degradation of the component of interest, from among the plurality of preceding components, by comparing the feature importance metrics of the plurality of sensors of each preceding component with a plurality of threshold values, wherein the component of interest is identified to have a dependency on the preceding components that contribute to the degradation of the component of interest;
estimating Remaining Useful Life (RUL) of the component of interest, based on the components that have been identified as contributing to the degradation of the component of interest, using a self-ensemble model; and
detecting a change point of the component of interest, based on the components that have been identified as contributing to the degradation of the component of interest, using the self-ensemble model.

5. The system as claimed in claim 4, wherein the system determines the dependency by:
performing for each component preceding the component of interest:
identifying the feature importance metric of each of the plurality of sensors associated with the preceding component;
generating a data-driven model using data from the plurality of sensors;
determining accuracy of the data driven model and comparing the determined accuracy with a first threshold;
labeling the preceding component as a component contributing for the degradation of the component of interest if the determined accuracy exceeds the first threshold;
comparing the determined feature importance metric of each of the plurality of sensors with a second threshold, and selecting all sensors from among the plurality of sensors for which the determined feature importance metric is greater than the second threshold;
comparing the determined feature importance metric of each of the plurality of sensors for which the determined accuracy is less than the first threshold, with a third threshold, and selecting all sensors for which the feature importance metric is more than the third threshold; and
labeling the preceding component as a component contributing for the degradation of the component of interest, if the feature importance metric of at least one of the plurality of sensors exceeds the third threshold.

6. The system as claimed in claim 4, wherein the system obtains the self-ensemble model by:
fetching a plurality of available prediction models of each component of interest;
fetching at least one model selection criteria, for a RUL estimation requirement, as an input;
selecting all prediction models from the available prediction models that satisfy the fetched selection criteria;
tuning a plurality of parameters of each of the plurality of ensemble methods with each of the selected predictions models and generating predictions using each of the tuned ensemble methods;
comparing the generated predictions with a measured value of predictions to determine an error value between the predicted value and the measured value;
selecting all ensemble methods having the error value less than an error threshold, as candidate ensemble methods, from the plurality of ensemble methods; and
selecting an ensemble method from among the candidate ensemble methods, based on the selection criteria.
, Description:FORM 2

THE PATENTS ACT, 1970
(39 of 1970)
&
THE PATENT RULES, 2003

COMPLETE SPECIFICATION
(See Section 10 and Rule 13)

Title of invention:
METHOD AND SYSTEM FOR REMAINING USEFUL LIFE (RUL) ESTIMATION BASED ON DEPENDENCY BETWEEN INDUSTRIAL COMPONENTS

Applicant:
Tata Consultancy Services Limited
A company Incorporated in India under the Companies Act, 1956
Having address:
Nirmal Building, 9th Floor,
Nariman Point, Mumbai 400021,
Maharashtra, India

The following specification particularly describes the invention and the manner in which it is to be performed.
TECHNICAL FIELD
The disclosure herein generally relates to industrial plant monitoring, and, more particularly, to a method and system for change point detection and Remaining Useful Life (RUL) estimation of components in the industrial plant by considering dependency between different components.

BACKGROUND
Manufacturing or process industry consists of large number of movable and immovable assets or units comprising of multiple components that work in tandem to produce or manufacture a product of interest. The production or manufacturing environment is considered as an evolving environment due to deterioration of components and sensors, maintenance activities, upgrading plans involving new components and system architectures, and the change in the operational and environmental conditions. The environment further complicates when the performance of individual units depends on the degradation of multiple components in the unit or other units. For example, as the components are connected to each other, and as working of different components may be interdependent, degradation of a component can accelerate the degradation processes of the other components. As a result, the degradation in single component can modify/affect Remaining Useful Life (RUL) of the dependent components.
Current practice in the industry is to monitor the key performance indicators (KPIs) of either assets or components or both. When the KPIs deviate farther away from corresponding desired values, it is considered as an indication of component degradation. However, change point of degradation can be detected based on early signs in the sensor data. This also helps in identification of triggering conditions that lead to degradation of asset and also in prediction of remaining useful life of the given asset or component.
Existing methods used for change point detection and RUL estimation have the disadvantage that they perform the RUL estimation of a component by considering values and parameters representing functioning of that component alone. As a result, they have limited applicability i.e. for RUL estimation of the systems with independent asset or components. Further, different plants are deployed in different fields of applications, hence characteristics of the components as well as the data being handled by each of the components may be different. For such plants deployed in different fields of applications, performance may need to be assessed based on different criteria, and hence a single monitoring system may not be able to perform the monitoring and RUL estimation of different plants.

SUMMARY
Embodiments of the present disclosure present technological improvements as solutions to one or more of the above-mentioned technical problems recognized by the inventors in conventional systems. For example, in one embodiment, a processor implemented method of monitoring an industrial plant is provided. In this method, real-time and non-real-time data from a plurality of data sources for each of a plurality of components of the industrial plant are collected as input, via one or more hardware processors. Further, health monitoring of each component of interest from among the plurality of components is performed, via the one or more hardware processors. During the health monitoring, a plurality of feature importance metrics of a plurality of sensors of each of a plurality of preceding components of the component of interest are determined. Further, all the preceding components that are contributing for degradation of the component of interest are identified, from among the plurality of preceding components, by comparing the feature importance metrics of the plurality of sensors of each preceding component with a plurality of threshold values, wherein the component of interest is identified to have a dependency on the preceding components that contribute to the degradation of the component of interest. Further, Remaining Useful Life (RUL) of the component of interest is estimated, based on the components that have been identified as contributing to the degradation of the component of interest, using a self-ensemble model. Further, a change point of the component of interest is detected based on the components that have been identified as contributing to the degradation of the component of interest, using the self-ensemble model.
In another aspect, a system for monitoring an industrial asset or component is provided. The system includes one or more hardware processors, a communication interface, and a memory coupled to the one or more hardware processors via the communication interface. The memory comprises a plurality of instructions which when executed, cause the one or more hardware processors to initially collect real-time and non-real-time data from a plurality of data sources for each of a plurality of components of the industrial plant, as input. Further, health monitoring of each component of interest from among the plurality of components is performed. During the health monitoring, a plurality of feature importance metrics of a plurality of sensors of each of a plurality of preceding components of the component of interest are determined. Further, all the preceding components that are contributing for degradation of the component of interest are identified, from among the plurality of preceding components, by comparing the feature importance metrics of the plurality of sensors of each preceding component with a plurality of threshold values, wherein the component of interest is identified to have a dependency on the preceding components that contribute to the degradation of the component of interest. Further, Remaining Useful Life (RUL) of the component of interest is estimated, based on the components that have been identified as contributing to the degradation of the component of interest, using a self-ensemble model. Further, a change point of the component of interest is detected based on the components that have been identified as contributing to the degradation of the component of interest, using the self-ensemble model.
In yet another aspect, a non-transitory computer readable medium for monitoring an industrial asset is provided. The non-transitory computer readable medium includes a plurality of instructions, which when executed, cause one or more hardware processors to perform the health monitoring of the industrial assets by performing the following steps. Initially real-time and non-real-time data from a plurality of data sources for each of a plurality of components of the industrial plant are collected as input, via one or more hardware processors. Further, health monitoring of each component of interest from among the plurality of components is performed, via the one or more hardware processors. During the health monitoring, a plurality of feature importance metrics of a plurality of sensors of each of a plurality of preceding components of the component of interest are determined. Further, all the preceding components that are contributing for degradation of the component of interest are identified, from among the plurality of preceding components, by comparing the feature importance metrics of the plurality of sensors of each preceding component with a plurality of threshold values, wherein the component of interest is identified to have a dependency on the preceding components that contribute to the degradation of the component of interest. Further, Remaining Useful Life (RUL) of the component of interest is estimated, based on the components that have been identified as contributing to the degradation of the component of interest, using a self-ensemble model. Further, a change point of the component of interest is detected based on the components that have been identified as contributing to the degradation of the component of interest, using the self-ensemble model.
It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention, as claimed.

BRIEF DESCRIPTION OF THE DRAWINGS
The accompanying drawings, which are incorporated in and constitute a part of this disclosure, illustrate exemplary embodiments and, together with the description, serve to explain the disclosed principles:
FIG. 1 illustrates an example network implementation of a system for Remaining Useful Life (RUL) estimation of an industrial component, according to some embodiments of the present disclosure.
FIG. 2 depicts an example implementation of the system of FIG. 1, according to some embodiments of the present disclosure.
FIG. 3 is a block diagram depicting components of a RUL estimation module of the system of FIG. 2, according to some embodiments of the present disclosure.
FIG. 4 is a flow diagram depicting steps involved in the process of performing component dependency analysis for RUL estimation using the system of FIG. 2, according to some embodiments of the present disclosure.
FIG. 5 depicts flow of data between different components of the relationship determination module in the process of determining dependency between components of an industrial asset being monitored, according to some embodiments of the present disclosure
FIG. 6 depicts data flow between different components of the prediction module while predicting failure of a component using the system of FIG. 2, according to some embodiments of the present disclosure.
FIGS. 7A and 7B (collectively referred to as FIG. 7) is a flow diagram depicting steps involved in the process of selecting methods to generate an ensemble of methods for RUL estimation, using the system of FIG. 2, according to some embodiments of the present disclosure.
FIG. 8 is an example diagram depicting feature extraction by the system of FIG. 2, according to some embodiments of the present disclosure.

DETAILED DESCRIPTION OF EMBODIMENTS
Exemplary embodiments are described with reference to the accompanying drawings. In the figures, the left-most digit(s) of a reference number identifies the figure in which the reference number first appears. Wherever convenient, the same reference numbers are used throughout the drawings to refer to the same or like parts. While examples and features of disclosed principles are described herein, modifications, adaptations, and other implementations are possible without departing from the scope of the disclosed embodiments. It is intended that the following detailed description be considered as exemplary only, with the true scope being indicated by the following claims.
Referring now to the drawings, and more particularly to FIG. 1 through FIG. 8, where similar reference characters denote corresponding features consistently throughout the figures, there are shown preferred embodiments and these embodiments are described in the context of the following exemplary system and/or method.
FIG. 1 illustrates an example network implementation of a system for Remaining Useful Life (RUL) estimation of an industrial component, according to some embodiments of the present disclosure.
Although the present disclosure is explained considering that the system 102 is implemented on a server, it may be understood that the system 102 may also be implemented in a variety of computing systems 104, such as a laptop computer, a desktop computer, a notebook, a workstation, a cloud-based computing environment and the like. It will be understood that the system 102 may be accessed through one or more devices 106-1, 106-2... 106-N, collectively referred to as devices 106 hereinafter, or applications residing on the devices 106. Examples of the devices 106 may include, but are not limited to, a portable computer, a personal digital assistant, a handheld device, a Smartphone, a tablet computer, a workstation and the like. The devices 106 are communicatively coupled to the system 102 through a network 108.
In an embodiment, the network 108 may be a wireless or a wired network, or a combination thereof. In an example, the network 108 can be implemented as a computer network, as one of the different types of networks, such as virtual private network (VPN), intranet, local area network (LAN), wide area network (WAN), the internet, and such. The network 106 may either be a dedicated network or a shared network, which represents an association of the different types of networks that use a variety of protocols, for example, Hypertext Transfer Protocol (HTTP), Transmission Control Protocol/Internet Protocol (TCP/IP), and Wireless Application Protocol (WAP), to communicate with each other. Further, the network 108 may include a variety of network devices, including routers, bridges, servers, computing devices, storage devices. The network devices within the network 108 may interact with the system 102 through communication links.
As discussed above, the system 102 may be implemented in a computing device 104, such as a hand-held device, a laptop or other portable computer, a tablet computer, a mobile phone, a PDA, a smartphone, and a desktop computer. The system 102 may also be implemented in a workstation, a mainframe computer, a server, and a network server. In an embodiment, the system 102 may be coupled to a data repository, for example, a repository 112. The repository 112 may store data processed, received, and generated by the system 102. In an alternate embodiment, the system 102 may include the data repository 112.
The network environment 100 supports various connectivity options such as BLUETOOTH®, USB, ZigBee and other cellular services. The network environment enables connection of devices 106 such as Smartphone with the server 104, and accordingly with the database 112 using any communication link including Internet, WAN, MAN, and so on. In an exemplary embodiment, the system 102 is implemented to operate as a stand-alone device. In another embodiment, the system 102 may be implemented to work as a loosely coupled device to a smart computing environment. An example implementation of the system of FIG. 1 is depicted in FIG. 2. Further, various steps involved in the process of RUL estimation being performed by the system 102 are depicted in flow diagrams depicted in FIG. 3 through FIG. 7, and the steps are explained with reference to the hardware components depicted in FIG. 1 and FIG. 2.
According to an embodiment of the disclosure, the system 102 further comprises one or more hardware processors, a memory in communication with the one or more hardware processors and a data repository. The one or more hardware processors are configured to execute programmed instructions stored in the memory, to perform various functions as explained in the later part of the disclosure. The repository may store data processed, received, and generated by the system 102.
Further, the system 102 is configured to collect and automatically preprocess the industrial data to extract the feature sets and their corresponding preprocessed values that are required for various models that are used for change point detection and RUL estimation. The system 102 is also configured to consider the dependency of a component on other components on the plant, to understand the degradation of any component being considered at a time instance. The term ‘component’ may refer to an industrial asset or an equipment within an asset. The system 102 is also configured to select a set of methods to form an ensemble of methods and use the ensemble of methods to generate RUL predictions. The system 102 is further configured to monitor the performance of each model and the ensemble method. The system 102 is also configured to identify a root cause if any deviation of prediction performance is detected. The system 102 is further configured to either retune individual models or ensemble method being used, depending on the determined root cause of the deviation.
The processing or manufacturing plant or any process industry can be represented in terms of a plurality of assets and plurality of equipment. The industries can be iron and steel making, power generation, pharma manufacturing, refineries, cement making, oil and gas production, fine chemical production, paper making, automotive production and so on. Each asset is a combination of multiple equipment, designed to perform specific operations in the plant. Each equipment may be a building block or part of an asset, and may have dedicated functions contributing to overall functioning of the plant. The system 100 can be configured to perform the health monitoring (including RUL estimation and change point estimation) at an asset level or equipment level. For explanation purpose, a term ‘component’ is used to represent assets as well as equipment. When the health monitoring is being performed at an asset level, the term ‘component’ refers to the asset and all equipment that form part of the asset. When the health monitoring is being performed at an equipment level, the term ‘component’ refers to the equipment. A few examples of the assets and equipment that may be monitored for data mining in manufacturing and process industries are, but not limited to, valves, compressors, blowers, pumps, steam turbines, gas turbines, heat exchangers, chemical reactors, bio-reactors, condensers, and boilers. The assets may be either movable assets such as a motor vehicle or immovable assets such as a reactor or a mixer. The components may or may not be interacting with each other. In case the plurality of these components interact with each other and the degradation of a component depends on other components, the asset/equipment is considered as interacting type. The plurality of these components may or may not be identical. In case, the plurality of components are identical with each other, the components are considered as identical components. The interaction among the components may or may not be physical. The interaction among the components can be direct such as interaction among gears in a gear box. The interaction among the components can be indirect such interaction among various idlers and pulleys in a conveyor belt. The plurality of components may be from unit operations such as pulverized coal fired boiler, rougher flotation cells, wind and gas turbines, semi-permeable membrane tubes in series and parallel, with applications in process industries, refineries, drugs and pharmaceuticals, power plants, steel plants, pulp and paper mills, cement plans, mineral processing plants, etc. In an example, degradation of hydrodynamic bearings in wind turbines may lead to increasing the looseness of primary transmission shafts, which in turn increase the vibration levels in the gearbox and accelerate the degradation of the gears.
FIG. 2 depicts an example implementation of the system of FIG. 1, according to some embodiments of the present disclosure. The system 102 comprises an automation system or distributed control system (DCS) 202, a plurality of asset data sources 203, a server 204, RUL estimation module 205, a model repository, a knowledge database, a plurality of databases and the I/O interface/user interface. It should be appreciated that the model repository, the knowledge database and the plurality of databases could be the part of a data repository in a memory module of FIG. 1. Various components of the system 102 work in combination for real-time monitoring, change point detection and RUL prediction in interacting and non-interacting systems. Functions of various components of the system 102 are explained below.
The knowledge database stores process knowledge and user knowledge. In an embodiment, the process knowledge and the user knowledge are derived from data being handled by one or more components of the RUL estimation module 205. In another embodiment, at least part of the process knowledge and the user knowledge is obtained via the user interface as an input. The process knowledge and the user knowledge comprise knowledge derived from multitude of simulations being performed by the RUL estimation module 205, knowledge of dependency of each component on other components, performance information of the plurality of physics-based and data-driven models of the component, and diagnostics information or fault trees for common process and the components. Data in the knowledge database is accessible to necessary components of the system 100, at various stages of the data processing involved in the plant monitoring and RUL estimation being performed by the system 102.
According to an embodiment of the disclosure, the plurality of databases comprise static and dynamic information pertaining to the components. The databases comprise a material database that contains static properties of raw materials, intermediate products, byproducts, final product and emissions, etc., a component database that contains component design data, details of construction materials, etc., a process configuration database that contains of process flowsheets, equipment layout, control and instrumentation diagrams, etc., an algorithm database that contains algorithms and techniques of data-driven and physics-based models, and solvers for physics-based models and optimization problems. The databases further comprise an operations database that contains sensor data, a laboratory database that contains of properties of raw materials, intermediate products, byproducts, final products and emissions obtained via tests at the laboratories, a maintenance database that consists of condition of the process, health of a component, maintenance records indicating corrective or remedial actions on various components, etc., an environment database that consists of weather and climate data such as ambient temperature, atmospheric pressure, humidity, etc.
According to an embodiment of the disclosure, referring to FIG. 2, the model repository comprises physics-based models, data-driven models, soft sensor models and RUL prediction models, change point detection models of the components, ensemble models of change point detection and relationship determination models.
FIG. 3 is a block diagram depicting components of a RUL estimation module of the system of FIG. 2, according to some embodiments of the present disclosure. The RUL estimation module 205 includes a receiving module 301, a data preprocessing module 302, a prediction module 303, an offline simulation module 304, a monitoring module 305, a diagnosis module 306, a self-ensemble module 307, a model building module 309, a feature extraction module 310, and a relationship determination module 311. Functions of various modules are explained below. Also, functions of some of the modules of the RUL estimation module 205 are depicted using the flow diagrams from FIG. 4 through FIG. 7.
The receiving module 301 is configured to receive the plurality of real-time data from the server 204 and real-time and non-real-time data from various data sources of the component, at a pre-determined frequency as an input. The real-time and non-real-time data may include operational data of various components, and information on failures of various components. The input data may be configured to be received at a pre-configured frequency (for example, once in every 1 min, once in every 5 min, and so on). Real-time data comprises operations data such as temperatures, pressures, flow rates, levels, density, valve opening percentages, vibrations, chemical composition of gases, dust levels, power consumption, motor currents, motor RPM, and so on, measured for various components. The input data may also comprise environment data such as ambient temperature, atmospheric pressure, ambient humidity, rainfall, and so on. The non-real time data includes data from laboratory tests and maintenance activities. Laboratory data consists of parameters such as chemical composition, particle size distribution, density, microstructure information, impurity levels, material assay, dissolved oxygen levels, turbidity, plasticity, volatile matter, ash and so on. The maintenance data includes details of planned and unplanned maintenance activities performed on one or more components, and condition and health of the process and various components. Real-time data is obtained from plant automation systems such as distributed control system (DCS) via a communication server such as OPC server or via an operations data source such as a historian. The non-real-time data is obtained from Laboratory Information Management System (LIMS), manufacturing execution system (MES), historian and other plant maintenance databases. Both real-time as well as non-real time data with respect to the failure of the components also are collected as input.
The data preprocessing module 302 is configured to pre-process the input data. The preprocessing of data comprises identification and removal of outliers, imputation of missing data, and synchronization and integration of a plurality of variables from one or more data sources using the frequency of various units in the plant. The sampling frequency of real-time and non-real-time data may be unified to, for example, once every 1 min, where the real-time data is averaged as necessary and the non-real-time data is interpolated or replicated as necessary.
The relationship determination module 311 is configured to determine dependency of each components on degradation of any other component. Various steps involved in the process of determining the dependency are depicted in FIG. 5, and are explained hereafter. In an embodiment, the system 102 may perform the RUL estimation for one component at a time. In another embodiment, the system 102 may perform the RUL estimation for multiple components at a time. The component(s) for which the RUL estimation is being done at any point of time is referred to as ‘component of interest’. Depending whether the monitoring and estimation is being carried out at asset level or equipment level, the component of interest may be an asset of an equipment. If the monitoring is done at asset level, then the component of interest may be an asset, and all equipment associated with the asset. The components that provide the inputs such as but not limited to material inputs, information inputs, that are needed for production or operation or processing at the component of interest are called upstream components/ preceding components. If the component of interest is an asset, then the preceding components also are assets. If the component of interest is an equipment, then the preceding components also are equipment. On the other hand, the components where outputs, such as but not limited to product outputs or information output, of the component of interest gets distributed are called downstream components/ subsequent components. If the component of interest is an asset, then the subsequent components also are assets. If the component of interest is an equipment, then the subsequent components also are equipment. The relationship determination module 311 identifies all preceding and subsequent components for the component of interest. To determine if any of the preceding components is contributing for the degradation of the component of interest, the relationship determination module 311 initially performs a feature selection using the sensors of the preceding component being considered at an instance, with respect to the degradation of the component of interest. The relationship determination module 311, further calculates a feature importance metric of all the sensors of the particular preceding component with respect to the degradation in the component of interest.
To determine whether the particular preceding component is contributing for degradation of the component of interest, the relationship determination module 311 then generates a data driven model to predict the change point of the component of interest, using selected sensors of the preceding component, wherein the sensors are selected based on the calculated feature metric. The relationship determination module 311 may use any suitable machine learning technique (a known in the art technique) to process the sensor data and generate the data driven model. Further, the relationship determination module 311 determines accuracy of the generated data driven model. The relationship determination module 311 may determine the accuracy by comparing predictions generated by the data driven model with actual measured values. The relationship determination module 311 then compares the measured accuracy with a first threshold, wherein value of the first threshold is pre-defined and configured with the relationship determination module 311. If the determined accuracy of the data driven model exceeds the first threshold, then the relationship determination module 311 labels the particular preceding component as a component that is contributing for the degradation of the component of interest. The relationship determination module 311 further determines the sensors of the preceding component that are contributing for the degradation of component of interest. The relationship determination module 311 achieves this by comparing the calculated feature importance metric of each of the sensors with a second threshold value. All the sensors with feature importance metric below the second threshold value are ignored and all the sensors with the feature importance metric greater than the second threshold are tagged as a features of a change point detection model of the component of interest. The second threshold is chosen such that majority of the sensors of the preceding component are tagged as features as the accuracy of the model built by relationship determination module 311 is well above the first threshold. This helps in making sure that most of the sensors from the identified preceding component are considered for subsequent feature extraction.
However, in case the accuracy of the generated data driven model is less than the first threshold, the particular upstream component is subjected to a second level of verification for dependency. The relationship determination module 311 performs this second level of verification based on feature importance. This level of verification helps in identifying those components where the models are not accurate as the particular preceding component is not self-sufficient to identify the degradation or change point in the component of interest. To perform this verification, the relationship determination module 311 compares the feature importance metric of each sensor of the preceding component with a third threshold. If the feature importance metric is less than the third threshold, then the sensor is ignored as a feature. If the feature importance metric is greater than the third threshold, then the sensor is tagged as a feature of a change point detection model of the component of interest, and in turn, the corresponding preceding component labeled as a component that is contributing for the degradation of the component of interest. The third threshold is chosen such that only those sensors of the preceding component that are contributing or correlated to the degradation in the component of interest are tagged as features. This is because of low accuracy of the model built by relationship determination module 311. This helps in making sure that only few sensors from the identified preceding component are considered for subsequent feature extraction.
In an embodiment, a second approach that may be used by the relationship determination module 311 is explained below. In this approach, dependency of a change point of a component of interest on corresponding preceding components is identified based on change point of other components. Historical data consisting of data of sequence of components that failed is extracted from associated database(s) by the relationship determination module 311. The relationship determination module 311 further uses one or more suitable rule induction algorithms such as association mining, to extract patterns in the historical data. The patterns contain information on failures of certain components that leads to failure in another component. However, if data available is insufficient for the system 102 to identify such patterns with a desired accuracy, then the system 102 may collect additional data such as but not limited to times of failures of each asset, and status of health of each component before starting the operations on a given day, and use this information to identify the patterns. This additional data is used along with the equipment failure data to identify the patterns and one or more classification models. This helps operator in identifying the critical components that needs close monitoring as these critical components might degrade beyond their thresholds for given operating conditions.
By means of one or more of the aforementioned processes, the relationship determination module 311 determines and provides information on the components that contribute for deterioration in other components. The relationship determination module 311 also helps in identifying the sensors of each component that are necessary to identify the change point of other components. The relationship determination module 311 further identifies the patterns in the sequence of failures of components that are being monitored.
The feature extraction module 310 is configured to estimate the feature set required for each RUL model or a change point detection model of each component in the plant in real-time. FIG. 8 illustrates an example of the feature-set extraction by the feature extraction module 310. The feature extraction module 310 collects the preprocessed data from the data preprocessing module 302, and sensors information as suggested by the relationship determination module 311 as inputs for the feature set extraction. The feature extraction module 310 applies one or more predefined feature extraction methods to create the feature-sets that are later used for model building and model predictions. The feature extraction module 310 may calculate the mean of a values of a sensor for a given time window. The feature extraction module 310 may use 1D-CNN method to extract the spatial features of the time series data that can be further used for model predictions. The feature extraction module 310 may use filters to obtain residual values from sensors after removing the trend and seasonality in the data. The different feature set generated using the different methods may be then used to generate a combined feature set, which in turn is used for change point detection and RUL prediction.
The offline simulation module 304 is configured to perform simulation tasks on the plant, to generate data that is required for the estimation of RUL and change point detection, but cannot be measured due to practical difficulties owing to the complexity of the components. The offline simulation module 310 generates specific test instances for simulation that are simulated using high fidelity physics-based models and data-driven models. These models provide insights into overall performance of the components. The simulation from the offline simulation module 310 may be performed upon receiving request from one or more of the modules such as the prediction module 303 and so on. The offline simulation module 304 interacts with the plurality of databases, the knowledge database and the model repository. The information generated by the offline simulation module 304 is stored in one or more of the plurality of databases.
The prediction module 303 is configured to utilize the real-time data from the data preprocessing module 302 for the RUL estimation and the change point detection. The prediction module 303 is further configured to collect and utilize data from the offline simulation module 304 for offline RUL estimation and change point detection for a given offline data. These models can be either physics based models or data driven models. These models captures the behavior of that specific component in either absence or presence of influence from other components. The prediction module 303 further comprises prediction of those models that consider the prediction of individual or independent models along with the data from the preprocessing module 302 as input. The prediction module 303 further comprises prediction of these models that consider the data from the preprocessing module 302 of multiple preceding components as suggested by the relationship determination module 311. These models capture the influence of other components in the change point detection and RUL estimation of a given component. These dependent models can be either physics based models or data driven models. The prediction module 303 is configured to use an ensemble method for generating the predictions. Use of the ensemble method involves using a combination of the dependent and independent models to generate the predictions on degradation of the component and then uses these predictions to obtain a single prediction value using various ensemble techniques. The prediction module 303 may be further configured to be in communication with the user interface, and the plurality of databases, to store the generated predictions and/or to provide the predictions to a user via a suitable user interface.
The monitoring module (also referred to as ‘performance monitoring module’) 305 is configured to determine the performance of the ensemble method used by the prediction module 303, by comparing the predictions from the ensemble method with their respective measured or actual values. The measured values may be received intermittently from the laboratory database or periodically from sensors, or may be fed by a user via the user interface provided. The performance of ensemble method is computed using performance index that utilizes the predictions of the ensemble method and the corresponding measured variables. The model performance index may be computed based on false alarms, or missed alarms, or based on deviation of the detected change point or predicted RUL from the measured values, or using combination of any of the three factors. The performance index can be one of plurality of error metrics such as precision, specificity, sensitivity, F1 score, true positives, true negatives, false positives, and false negatives, and so on. The performance index can also be one of the plurality of mean error, mean absolute error, absolute percentage error, mean absolute percentage error, mean square error, root mean square error, etc. The performance index can also be one of the plurality of Chebyshev distance or Euclidean distance or Mahalanobis distance or Manhattan distance or Minkowski distance between predictions of the models and the corresponding measured variables. If the performance index exceeds a predefined threshold for any model, the monitoring module 305 triggers the diagnosis module 306.
The diagnosis module 306 is configured to identify one or more reasons behind the drift in the performance index of the change point detections and RUL predictions by the data driven models. The diagnosis module 306 verifies whether the drift in performance index is due to deterioration of the accuracy in independent and dependent models that are used in ensemble modeling, or is due to the inapplicability of the ensemble method, or due to necessity for changing values of different coefficients in the ensemble method used for the predictions. The diagnosis module 306 verifies whether the drift in accuracy of ensemble method is due to change in accuracy of any of the dependent and independent models, by comparing the accuracy of these models obtained with the recent data to the accuracy of the same models when they are built. In an embodiment, information on the accuracy of the models when they were built, is stored as historical information in one of the databases of the system 102, and can be accessed by the diagnosis module 306 and any other authorized module. If the accuracy has deteriorated, the diagnosis module 306 tags these models for retuning and relationship determination module 311 is triggered. The diagnosis module 306 further updates statistics of distribution of data that is applicable for one or more of the dependent or independent models that are triggered for retuning in the knowledge base. The diagnosis module 306 is further configured to trigger the self-ensemble module 307 when the accuracy of all the dependent and independent models are within the predefined threshold and hence the performance of the ensemble method is determined as the reason behind the drift in the performance index. The diagnosis module 306 then updates the statistics of the distribution of data that is applicable for the ensemble method in the knowledge base.
The self-ensemble module 307 is configured to generate the ensemble method for generating the predictions. The self-ensemble module 307 is configured to select a plurality of available models to tune the coefficients in the available ensemble methods, using the process depicted in FIG. 7. The self-ensemble module 307 fetches one or more suitable model selection criteria (also referred to as ‘selection criteria’) for given requirement as an input. Here the term ‘given requirement’ at any instance refers to type of accuracy in the RUL estimation and change point being performed for a given set of input data. Various selection criterion used may be defined in such a way that factors such as but not limited to over-fitting, under-fitting, false alarms, missed alarms, and so on can be reduced. The self-ensemble module 307 further identifies all the available models from among the available prediction models, which satisfy the fetched selection criteria. Further, the self-ensemble module 307 estimates the coefficients of the available ensemble methods with each of the identified individual predictions models and generates predictions using each of the trained ensemble methods. The generated predictions are compared with a measured value of predictions, to determine error between the predicted value and the measured value. If the determined error is less than a defined error threshold, for prediction generated by the trained ensemble method, that means the ensemble method can generate predictions with required accuracy, and hence is selected as a candidate ensemble method. If the error exceeds the error threshold, then the ensemble method is identified as not good enough for considering for prediction and hence is discarded. Further, from among the selected ensemble methods, an ensemble method is selected based on the specified selection criteria. In an embodiment, if more than one of the selected ensemble methods satisfy the selection criteria, then self-ensemble module 307 may consider a model index which represents performance of each of the models for different selection criteria in past instances, and selects an ensemble method from among the candidate ensemble methods, having highest index value for the selection criteria being considered.
The model building module 309 is configured to build a plurality of physics-based and data-driven independent and dependent models of the plant and retune/retrain the models in case of a drift in their performance. For the physics-based models, the model building module 309 identifies respective mathematical model as suggested using user interface and tunes its parameters such as but not limited to heat transfer coefficients, tuning parameters for reaction rates, and tuning parameters for specific heat capabilities of various species using the selected data from the available databases. The model building module 309 further re-tunes these tunable parameters when the diagnosis module 306 identifies that there is a performance drift due to these models. For the data-driven models, the model building module 309 builds multiple models for each of the applicable algorithm in the database or by using the algorithms suggested using user interface, by optimizing various hyper-parameters of these algorithms. The model building module 309 further re-tunes these hyper-parameters of the models when diagnosis module 306 identifies that there is a performance drift due to these models. The re-tuned and re-trained models are stored in the model repository and may be activated for prediction in the prediction module 303. The model building module 309 further triggers re-tuning or re-training of the models of the components that have dependency on other components for which the performance index deteriorated.
FIG. 4 is a flow diagram depicting steps involved in the process of performing component dependency based RUL estimation using the system of FIG. 2, according to some embodiments of the present disclosure. At step 402, the system 102 collects real-time and non-real information on operational data of the components being monitored, as well as information on failure of various components, as input. A component for which the RUL estimation and change point detection needs to be done is selected as component of interest, at step 404. Further, at step 406, dependency of other components on degradation of the component of interest is determined. Various steps involved in the process of determining the dependency are depicted in FIG. 5 & FIG. 6 and details are explained in description of relationship determination 311 as well as in description of FIG. 5 & FIG. 6. Further, at step 408, the RUL estimation of the component of interest is estimated. In various embodiments, various steps in method 400 may be performed in the same order as depicted in method 400 or in any alternate order that is technically feasible. In another embodiment, one or more steps in method 400 may be omitted if required.
FIG. 5 depicts flow of data between different components of the relationship determination module in the process of determining dependency between components being monitored, according to some embodiments of the present disclosure. For a component of interest selected at an instance, a feature scoring module 502 of the relationship determination module 311 performs feature selection using the sensors of the particular upstream unit or component and calculates a feature importance metric to each sensor of that particular preceding component. Further, a model accuracy determination module 504 of the relationship determination module 311 creates a model and then accuracy of the model is calculated by processing data from one of the associated databases and then by verifying accuracy of predictions made in comparison with actual measured values. A model accuracy verification module 506 determines whether the accuracy of the generated model is beyond the first threshold value. Then, a feature score verification module 508 determines whether any of a plurality of sensors is contributing for the deterioration of the component of interest by comparing the feature importance metrics of each of the sensors with second and third threshold values respectively. The first, second and third threshold values may be passed or changed using the user interface. The feature assignment module 510 then assigns all the relevant sensors of all the preceding components that are contributing for the degradation of component of interest as features that need to be considered for feature extraction.
FIG. 6 depicts data flow between different components of the prediction module while predicting failure of a component using the system of FIG. 2, according to some embodiments of the present disclosure. A rule induction module 602 of the prediction module 303 identifies all the data with respect to sequence of components that failed. The rule induction module 602 further identifies one or more suitable rule induction algorithms such as association mining that are used to identify patterns in the historical data. The patterns contain information on failures of certain components that leads to failure in another component. Further, a model accuracy verification module 604 of the prediction module 303 calculates accuracy of these patterns using the data from the databases. If the available data is insufficient for the system 102 to create models containing such patterns with a desired accuracy, then additional data such as but not limited to times of failures of each component, and health status of each component before starting the operations on a given day, are collected for a classification model for each component. Such additional data may be verified for authenticity using the data verification module 606 of the prediction module 303. This additional data is used along with the equipment failure data to identify the patterns and one or more classification models using model building module 309 to predict failure of the component of interest. In various embodiments, various steps in method 600 may be performed in the same order as depicted in method 600 or in any alternate order that is technically feasible. In another embodiment, one or more steps in method 600 may be omitted if required.
FIG. 7 is a flow diagram depicting steps involved in the process of selecting ensemble models to generate an ensemble model for RUL estimation, using the system of FIG. 2, according to some embodiments of the present disclosure. At step 702, a plurality of available prediction models for the component of interest are fetched. At step 704, one or more model selection criteria (also referred to as ‘selection criteria’) is fetched as input. Various selection criterion used may prioritize reducing over-fitting, under-fitting, false alarms, missed alarms, and so on. Further, at step 706, all prediction models which satisfy the fetched selection criteria are identified from among the available prediction models. Further, predictions are generated using each of the identified ensemble models that utilize prediction models that satisfied the selection criteria, after tuning a plurality of parameters of each of the ensemble models at step 708. The generated predictions are compared with a measured value of predictions at step 710, to determine error between the predicted value and the measured value. If the determined error is less than a defined error threshold, at step 712, for prediction generated by the ensemble model that signifies that the ensemble model can generate predictions with required accuracy, and hence is selected as a candidate ensemble model at step 716. Further, at step 718, one of the candidate ensemble models is selected as an ensemble model satisfying the selection criteria, which may be then used for generating the predictions. If the error exceeds the threshold of error, the accuracy of the ensemble model is less, then at step 714, the ensemble model is discarded. In various embodiments, various steps in method 700 may be performed in the same order as depicted in method 700 or in any alternate order that is technically feasible. In another embodiment, one or more steps in method 700 may be omitted if required.
Practical Example:
For example, consider a wind turbine. Wind turbines used for practical applications may be of different types in terms of sizes, however are similar in terms of the components present. Some of the major components are:

Rotor Blades - The rotor blades of a wind turbine operate under the same principle as aircraft wings. One side of the blade is curved while the other is flat. The wind flows more quickly along the curved edge, creating a difference in pressure on either side of the blade. The blades are “pushed” by the air in order to equalize the pressure difference, causing the blades to turn.

Nacelle – The nacelle contains a set of gears and a generator. The turning blades are linked to the generator by the gears. A series of gears increase the rotation of the rotor from about 18 revolutions a minute to roughly 1,800 revolutions per minute -- a speed that allows the turbine’s generator to produce AC electricity.

A streamlined enclosure called a nacelle houses key turbine components -- usually including the gears, rotor and generator -- are found within a housing called the nacelle. Sitting atop the turbine tower, some nacelles are large enough for a helicopter to land on.

Another key component is the turbine’s controller, which keeps the rotor speeds from exceeding 55 mph to avoid damage by high winds. An anemometer continuously measures wind speed and transmits the data to the controller. A brake, also housed in the nacelle, stops the rotor mechanically, electrically or hydraulically in emergencies.

Tower – The blades and nacelle are mounted on top of a tower. The tower is constructed to hold the rotor blades off the ground and at an ideal wind speed. Towers are usually between 50-100 m above the surface of the ground or water. Offshore towers are generally fixed to the bottom of the water body, although research is ongoing to develop a tower that floats on the surface.

Further, the rotor blades are connected with various other components within Nacelle such as multiple set of gears (or a transmission system), generator, brake etc.
In any normal operation, controller within existing systems takes care of the drift in the wind speed by appropriately selecting suitable gear for a fixed frequency power generation from wind Turbine unit, which is always monitored at the generator end.
Though, considering a case, where anemometer (wind sensor) on which controller depends starts giving erroneous readings. Controller will be using this reading to engage gears accordingly and hence will be generating power with inconsistencies than the desired (AC) power. Here, controller have no way of correcting the output power as it can only specify combination of gear set which will not be able to produce desired power.
With proposed architecture, all the subsystems: rotor, gear set, generator are considered as interacting systems. At rotor level, performance of the rotor vary over time, owing to its placement in harsh open environment. Similarly, transmission box degrades and even get failed under sudden high wind speeds.
The relationship determination module 311 utilizes local component level features. It can sequentially monitor respective outputs based on sub-system level interaction. Considering rotor as component 1, the relationship determination module 311 checks for the suspected change in the output at rotor level utilizing anemometer reading and pre built features/models, and continues to do so sequentially till last component which is the generator. As the sensed reading of wind speed itself may be wrong, a change is be detected between the observed and calculated power value. All the intermediate component level features clearly signify no suspected drift from expected values and hence the reason of inaccuracy can be marked because of sensor.
Example 2:
The industrial component considered for this example includes 4 interconnected identical equipment. An equipment enters an abnormal condition when D_t^j exceeds a threshold df, where D_t^jis time-continuous degradation value of jth equipment. This abnormal condition state does not correspond to the equipment failure but makes the operating conditions of the system harsher. The equipment failure occurs at time Tf, when all four equipment start operating in abnormal conditions. The level of degradation of the equipment is estimated through 10 sensors installed on each equipment. The sensor signals, s_t^(j,1),…,s_t^(j,K), are influenced by both the degradation levels D_t^j, j=1, …, 4, and the operating condition E_t^ . At any given time t, label of the equipment with 0 indicates normal conditions and 1 indicates abnormal conditions.

Model selection criteria:
The error of a particular equipment in estimating the time of failure ?? j with t ^j,m is defined by:
"?" ^"j,m" "=" {¦("t" ^"j,m" "- " "t" ^^"j,m" &"t" ^"j,m" "?NAN," "t" ^^"j,m" "?NAN" @"0" &"t" ^"j,m" "=NAN," "t" ^^"j,m" "=NAN" @¦("k" _"false" @?"-k" ?_"missed" )&¦("t" ^"j,m" "=NAN," "t" ^^"j,m" "?NAN" @?" t" ?^"j,m" "?NAN," "t" ^^"j,m" "=NAN " ))¦ (1)
The error equal to ????????????>?? is assigned for the false alarms, and an error of ??????????????>?? is assigned for missed alarms. The following metric is used to as an aggregate error for M different test values.

"A= " "1" /?"4M" ?_"Test" ?_"m=1" ^("M" _"Test" )¦?_"j=1" ^"4" ¦?"?(" ?" ?" ?^"j.m" " )" ? "?[0,1]" (2)
where "?(" "?" ^"j,m" ")=" {¦(" 1" "?" ^"j,m" "<-T" @" " ("1-" "e" ^("?" ^"j,m" /"a" _"1" ) ) "b" _"1" " -T=" "?" ^"j,m" "<0" @" " ("1-" "e" ^("?" ^"j,m" /"a" _"2" ) ) "b" _"2" " " ?"0=?" ?^"j,m" "=T" @" 1" "?" ^"j,m" ">T" )¦ (3)

Parameters b1 and b2 in the equation (3) are set to obtain ?(T)=1 and ?(-T) =1, respectively. Parameters a1 and a2 are set such that missed alarms are more penalized compared to false alarms. The metric ‘A’ is ideally close to 0 and should be as less as possible.

Prediction models
To detect the operational change point in each of the four equipment of the interconnected system, the following Long Short Term Memory (LSTM) based models were developed for each equipment.
Model A: LSTM based regression for predicting change point, assuming no interdependence between all the 4 equipment
Model B: LSTM based regression for predicting change point, assuming interdependence between the 4 equipment using the relationship determination module 311.
Model C: LSTM based semi-supervised change point prediction
Model D: Ensemble LSTM to reduce the variance in the predictions of the multiple models that are trained with same training data.
Model results:
Model Names Train Error Test Error Total Error
Model A 0.0081 0.021 0.0117
Model B 0.0585 0.2385 0.1089
Model C 0.4423 0.417 0.435
Model D 0.0084 0.0086 0.0085
Table. 1

Among all the above developed models, the ensemble model that utilizes multiple other LSTM models provided better results.
The written description describes the subject matter herein to enable any person skilled in the art to make and use the embodiments. The scope of the subject matter embodiments is defined by the claims and may include other modifications that occur to those skilled in the art. Such other modifications are intended to be within the scope of the claims if they have similar elements that do not differ from the literal language of the claims or if they include equivalent elements with insubstantial differences from the literal language of the claims.
The embodiments of present disclosure herein addresses unresolved problem of RUL estimation by considering dependency between industrial components being monitored. The embodiment, thus provides a mechanism to determine dependency between industrial components being monitored. Moreover, the embodiments herein further provide a mechanism to perform the RUL estimation, based on the determined dependency between the components.
It is to be understood that the scope of the protection is extended to such a program and in addition to a computer-readable means having a message therein; such computer-readable storage means contain program-code means for implementation of one or more steps of the method, when the program runs on a server or mobile device or any suitable programmable device. The hardware device can be any kind of device which can be programmed including e.g. any kind of computer like a server or a personal computer, or the like, or any combination thereof. The device may also include means which could be e.g. hardware means like e.g. an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or a combination of hardware and software means, e.g. an ASIC and an FPGA, or at least one microprocessor and at least one memory with software processing components located therein. Thus, the means can include both hardware means and software means. The method embodiments described herein could be implemented in hardware and software. The device may also include software means. Alternatively, the embodiments may be implemented on different hardware devices, e.g. using a plurality of CPUs.
The embodiments herein can comprise hardware and software elements. The embodiments that are implemented in software include but are not limited to, firmware, resident software, microcode, etc. The functions performed by various components described herein may be implemented in other components or combinations of other components. For the purposes of this description, a computer-usable or computer readable medium can be any apparatus that can comprise, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device.
The illustrated steps are set out to explain the exemplary embodiments shown, and it should be anticipated that ongoing technological development will change the manner in which particular functions are performed. These examples are presented herein for purposes of illustration, and not limitation. Further, the boundaries of the functional building blocks have been arbitrarily defined herein for the convenience of the description. Alternative boundaries can be defined so long as the specified functions and relationships thereof are appropriately performed. Alternatives (including equivalents, extensions, variations, deviations, etc., of those described herein) will be apparent to persons skilled in the relevant art(s) based on the teachings contained herein. Such alternatives fall within the scope of the disclosed embodiments. Also, the words “comprising,” “having,” “containing,” and “including,” and other similar forms are intended to be equivalent in meaning and be open ended in that an item or items following any one of these words is not meant to be an exhaustive listing of such item or items, or meant to be limited to only the listed item or items. It must also be noted that as used herein and in the appended claims, the singular forms “a,” “an,” and “the” include plural references unless the context clearly dictates otherwise.
Furthermore, one or more computer-readable storage media may be utilized in implementing embodiments consistent with the present disclosure. A computer-readable storage medium refers to any type of physical memory on which information or data readable by a processor may be stored. Thus, a computer-readable storage medium may store instructions for execution by one or more processors, including instructions for causing the processor(s) to perform steps or stages consistent with the embodiments described herein. The term “computer-readable medium” should be understood to include tangible items and exclude carrier waves and transient signals, i.e., be non-transitory. Examples include random access memory (RAM), read-only memory (ROM), volatile memory, nonvolatile memory, hard drives, CD ROMs, DVDs, flash drives, disks, and any other known physical storage media.
It is intended that the disclosure and examples be considered as exemplary only, with a true scope of disclosed embodiments being indicated by the following claims.