Systems and methods for remaining useful life (RUL) detection and controls based thereon

Abstract

Systems and methods for detection of a health state and / or RUL of a vehicle component can include one or more processors obtaining a plurality of data sets of one or more vehicles such that each data set is recorded from a corresponding vehicle at a corresponding RUL of a vehicle component and includes parameter values of a set of parameters of the vehicle. The one or more processors assign to each data set a corresponding health state of the vehicle component based on the corresponding RUL, and select a subset of parameters of the set of parameters based on impact of the parameters on the health state and / or RUL of the vehicle component. The one or more processors train, using values of the subset of parameters, a machine learning model for detection of the health state and / or RUL of the vehicle component, and provide the machine learning model for deployment.

Description

SYSTEMS AND METHODS FOR REMAINING USEFUL LIFE (RUL) DETECTION AND CONTROLS BASED THEREONCROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This PCT Application claims the benefit and priority to Indian Patent Application No. 202441000823, filed on January 4, 2024, entitled SYSTEMS AND METHODS FOR REMAINING USEFUL LIFE (RUL) DETECTION AND CONTROLS BASED THEREON, which is incorporated herein by reference in its entirety and for all purposes.TECHNICAL FIELD

[0002] The present disclosure relates to monitoring health states of systems and / or parts of vehicles and / or other pieces of equipment. More particularly, the present disclosure relates to systems and methods for detecting the remaining useful life (RUL) of components or systems of a vehicle or a piece of equipment using at least one machine learning model. The systems and methods can include developing a virtual RUL sensor using one or more machine learning models, and using the virtual RUL sensor to detect the RUL or a health state of one or more systems or parts of a vehicle or a piece of equipment.BACKGROUND

[0003] Emission regulations for internal combustion engines have become more stringent over recent years. Environmental concerns have motivated the implementation of stricter emission requirements for internal combustion engines throughout much of the world. Governmental agencies carefully monitor the emission quality of engines and set emission standards to which engines must comply. At least some emission standards call for proper and timely maintenance of aftertreatment systems, engine systems, transmission systems and / or other systems of vehicles or pieces of equipment.

[0004] When a system, component and / or part of a vehicle or a piece of equipment get old, such system, component and / or part may not operate in compliance with existing emission standards and / or other standards or regulations. In particular, the performance of the system, component and / or part of the vehicle or piece of equipment usually degrades gradually overtime. As the performance degradation progresses over time, e.g., gradually and / or abruptly, the system, component and / or part can reach a failure point (or failure state) at whichthe performance does not comply with a given standard, regulation and / or norm set by a government agency and / or a manufacturer of the vehicle (or piece of equipment) or systems thereof. For instance, performance of a treatment system or components thereof can degrade to pint where the gas released by the vehicle (or piece of equipment) does not meet one or more standards set by an equipment manufacturer.SUMMARY

[0005] One embodiment relates to a system for training machine learning models for detecting health states of vehicle components. The system can include one or more processors and a memory storing executable instructions. The executable instructions, when executed by the one or more processors, cause the one or more processors to obtain a plurality of data sets of one or more vehicles. Each data set is recorded from a corresponding vehicle of the one or more vehicles at a corresponding remaining useful life (RUL) of a vehicle component and includes parameter values of a predefined set of parameters of the vehicle. The one or more processors can assign to each data set of the plurality of data sets a corresponding health state of the vehicle component based on the corresponding RUL of the vehicle component; select, using the plurality of data sets, a subset of parameters of the predefined set of parameters based on variations of the predefined set of parameters in response to a change in the health state of the vehicle component; train, using parameter values of the subset of parameters in the plurality of the data sets, a machine learning model for detection of the health state of the vehicle component; and provide the machine learning model for use to detect the health state of the vehicle component.

[0006] In some implementations, the vehicle component is a diesel oxidation catalyst (DOC). The subset of parameters may include at least one of a minimum model base soot load in aftertreatment (MBSLE) parameter, a maximum combine soot load in aftertreatment (CSLE) parameter, a maximum MBSLE parameter, a mean delta pressure soot load estimate (DPSLE) offset adjustment parameter, or a maximum Urea pump pressure (related to DEF pressure) parameter.

[0007] In some implementations, the vehicle component is a selective catalytic reduction (SCR) system. The subset of parameters may include at least one of a maximum CSLE parameter, a maximum MBSLE parameter, a parameter representing average urea pump command while pressure is in closed loop but no dosing, a maximum diesel particulate filter (DPF) delta pressure parameter, or a parameter representing average compressor inlet pressure while equipment (or vehicle) is working is the fifth most predictive parameter.

[0008] In some implementations, the one or more processors are configured to select a number of parameters with largest variation ranges responsive to the change in the health state of the vehicle component. The one or more processors are configured to use an extreme gradient boosting (XGBoost) algorithm in selecting the number of parameters with the largest variation ranges responsive to the change in the health state of vehicle component.

[0009] In some implementations, the machine learning model includes at least one of: a random forest model; a statistical learning model; or a neural network.

[0010] In some implementations, the health state of the vehicle component includes an unhealthy state when the RUL of the vehicle component is less than a predefined threshold and a healthy state when the RUL of the vehicle component is greater than the predefined threshold. The predefined threshold may be defined as an additional driving distance to be driven by a vehicle associated with the vehicle component before performance of the vehicle component reaches a failure state.

[0011] Another embodiment relates to a method of training machine learning models for detecting health states of vehicle components, the method including: obtaining, by a computer system, a plurality of data sets regarding one or more vehicles, each data set recorded from a corresponding vehicle of the one or more vehicles at a corresponding remaining useful life (RUL) of a vehicle component and including parameter values of a predefined set of parameters of the vehicle; assigning, by the computer system, to each data set of the plurality of data sets a corresponding health state of the vehicle component based on the corresponding RUL of the vehicle component; selecting, by the computer system, using the plurality of data sets, a subset of parameters of the predefined set of parameters based on variations of the predefined set of parameters in response to a change in the health state of the vehicle component; training, by the computer system, using parameter values of the subset of parameters in the plurality of the data sets, a machine learning model for detection of thehealth state of the vehicle component; and providing, by the computer system, the machine learning model for use to detect the health state of the vehicle component.

[0012] In some implementations, the vehicle component is a diesel oxidation catalyst (DOC). The subset of parameters may include at least one of a minimum model base soot load in aftertreatment (MBSLE) parameter, a maximum combine soot load in aftertreatment (CSLE) parameter, a maximum MBSLE parameter, a mean delta pressure soot load estimate (DPSLE) offset adjustment parameter, or a maximum Urea pump pressure (related to DEF pressure) parameter.

[0013] In some implementations, the vehicle component is a selective catalytic reduction (SCR) system. The subset of parameters may include at least one of a maximum CSLE parameter, a maximum MBSLE parameter, a parameter representing average urea pump command while pressure is in closed loop but no dosing, a maximum diesel particulate filter (DPF) delta pressure parameter, or a parameter representing average compressor inlet pressure while equipment (or vehicle) is working is the fifth most predictive parameter.

[0014] In some implementations, the method includes selecting a number of parameters with largest variation ranges responsive to the change in the health state of the vehicle component. The method includes using an extreme gradient boosting (XGBoost) algorithm in selecting the number of parameters with the largest variation ranges responsive to the change in the health state of vehicle component.

[0015] In some implementations, the machine learning model includes at least one of: a random forest model; a statistical learning model; or a neural network.

[0016] In some implementations, the vehicle component is defined to be in the unhealthy state based on the RUL of the vehicle component being less than a predefined threshold. The predefined threshold is defined as an additional driving distance to be driven by a vehicle corresponding to the vehicle component before performance of the vehicle component reaches a failure criterion.

[0017] Another embodiment relates to a system for detecting health states of vehicle components. The system includes one or more processors and a memory storing executable instructions, which when executed by the one or more processors, cause the one or more processors to perform operations including: receiving parameter values of a plurality ofparameters of a vehicle; determining, using a machine learning model and the parameter values, a health state of a component of a vehicle, the health state indicative of whether a remaining useful life (RUL) of the vehicle component is less than a predefined threshold; and providing an indication of the health state for display on a user interface.

[0018] In some implementations, the vehicle component is a diesel oxidation catalyst (DOC). The subset of parameters includes at least one of a minimum model base soot load in aftertreatment (MBSLE) parameter, a maximum combine soot load in aftertreatment (CSLE) parameter, a maximum MBSLE parameter, a mean delta pressure soot load estimate (DPSLE) offset adjustment parameter, or a maximum Urea pump pressure (related to DEF pressure) parameter.

[0019] In some implementations, the vehicle component is a selective catalytic reduction (SCR) system. The subset of parameters includes at least one of a maximum CSLE parameter, a maximum MBSLE parameter, a parameter representing average urea pump command while pressure is in closed loop but no dosing, a maximum diesel particulate filter (DPF) delta pressure parameter, or a parameter representing average compressor inlet pressure while equipment (or vehicle) is working is the fifth most predictive parameter.

[0020] In some implementations, the machine learning model includes at least one of: a random forest model; a statistical learning model; or a neural network. The health state is an unhealthy state based on the RUL of the vehicle component being less than a predefined threshold.

[0021] Another embodiment relates to a method of detecting health states of vehicle components. The method includes: receiving, by one or more processors, parameter values of a plurality of parameters of a vehicle; determining, by the one or more processors, using a machine learning model and the parameter values, a health state of a component of a vehicle, the health state indicative of whether a remaining useful life (RUL) of the vehicle component is less than a predefined threshold; and providing, by the one or more processors, an indication of the health state for display on a user interface..

[0022] In some implementations, the vehicle component is a diesel oxidation catalyst (DOC). The subset of parameters includes at least one of a minimum model base soot load in aftertreatment (MBSLE) parameter, a maximum combine soot load in aftertreatment (CSLE) parameter, a maximum MBSLE parameter, a mean delta pressure soot load estimate (DPSLE)offset adjustment parameter, or a maximum Urea pump pressure (related to DEF pressure) parameter.

[0023] In some implementations, the vehicle component is a selective catalytic reduction (SCR) system. The subset of parameters includes at least one of a maximum CSLE parameter, a maximum MBSLE parameter, a parameter representing average urea pump command while pressure is in closed loop but no dosing, a maximum diesel particulate filter (DPF) delta pressure parameter, or a parameter representing average compressor inlet pressure while equipment (or vehicle) is working is the fifth most predictive parameter.

[0024] In some implementations, the machine learning model includes at least one of: a random forest model; a statistical learning model; or a neural network. The health state is an unhealthy state based on the RUL of the vehicle component being less than a predefined threshold.

[0025] This summary is illustrative only and is not intended to be in any way limiting. Other aspects, inventive features, and advantages of the devices or processes described herein will become apparent in the detailed description set forth herein, taken in conjunction with the accompanying figures, wherein like reference numerals refer to like elements. Numerous specific details are provided to impart a thorough understanding of embodiments of the subject matter of the present disclosure. The described features of the subject matter of the present disclosure may be combined in any suitable manner in one or more embodiments and / or implementations. In this regard, one or more features of an aspect of the invention may be combined with one or more features of a different aspect of the invention. Moreover, additional features may be recognized in certain embodiments and / or implementations that may not be present in all embodiments or implementations.BRIEF DESCRIPTION OF THE FIGURES

[0026] FIG. 1 is a schematic diagram of a computing and network environment for detection of RULs and / or health states of components and / or systems in vehicles and / or pieces of equipment, according to an example embodiment.

[0027] FIG. 2 is a schematic diagram of a controller of the RUL and / or health state detection system of FIG. 1, according to an example embodiment.

[0028] FIG. 3 is a flow chart of a method for generating or training a virtual RUL or health state detector, according to an example embodiment.

[0029] FIG. 4 is a flow diagram depicting data acquisition from vehicles or pieces of equipment, according to an example embodiment.

[0030] FIG. 5 is a diagram illustrating determination of the RUL and the health state of a vehicle component from recorded data, according to an example embodiment.

[0031] FIG. 6 is a chart depicting SHAP values of multiple selected vehicle parameters in relation with the health state of a vehicle component or system, shown as a diesel oxidation catalyst (DOC), according to an example embodiment.

[0032] FIG. 7 is a chart depicting SHAP values of multiple selected vehicle parameters in relation with the health state of a vehicle component or system, shown as a selective catalytic reduction (SCR) system, according to example embodiments.

[0033] FIG. 8 is a chart depicting SHAP values of multiple selected vehicle parameters in relation with the health state of a vehicle component or system, shown as a particulate filter and particularly a diesel particulate filter (DFP), according to example embodiments.

[0034] FIG. 9 is a flow chart of a method of detecting health states and / or RULs of components of vehicles and / or pieces of equipment using the virtual health state and / or RUL detector generated in FIG. 3, according to an example embodiment.DETAILED DESCRIPTION

[0035] Following below are more detailed descriptions of various concepts related to, and implementations of, methods, apparatuses, and systems for generating one or more virtual remaining useful life (RUL) detectors or sensors and detecting the RUL(s) of one or more systems and / or components of a vehicle or a piece of equipment using the virtual RUL detectors or sensors. Before turning to the figures, which illustrate certain exemplary embodiments in detail, it should be understood that the present disclosure is not limited to the details or methodology set forth in the description or illustrated in the figures. It should also be understood that the terminology used herein is for the purpose of description only and should not be regarded as limiting.

[0036] Some emission standards or regulations, e.g., set by government agencies, impose specific requirements on gasses released by vehicles and / or other pieces of equipment. The vehicles and / or other pieces of equipment are expected to comply with the requirements during operation. However, the performance of components of the vehicles and / or other pieces of equipment, e.g., components of a corresponding aftertreatment system, usually degrades over time and fails or reaches a failure state at some point of time. As used herein, the failure, failure event, or failure state of a component refers to a state when the performance of the component does not comply with predefined desired operating characteristics. In one embodiment, the predefined desired operating characteristics may correspond to at least one requirement or regulation. In another embodiment, the predefined desired operating character! stic(s) may correspond with a value(s) different from the at least one requirement or regulation (e.g., a value less than a predefined maximum value in order to provide a buffer to aid in desired operation). Enhanced compliance with existing operations standards and / or existing requirements or regulations calls for automatic or nearly automatic monitoring of health states of relevant components and / or systems of vehicles or pieces of equipment and detection of respective RULs.

[0037] Regular monitoring of the health or RUL of a component or system over time allows for prediction of failure events ahead of time, and the taking of appropriate action before the component or system reaches a state of failure. As used herein, the RUL of the component or system refers to a remaining period of operation (which may be expressed as a predefined amount of time and / or a predefined additional distance value to be driven, e.g., by the corresponding vehicle when the systems, computer-readable media, and methods are applied to a vehicle), before the failure event occurs. The remining period of operation can be a remaining operation time (e.g., during which the corresponding piece of equipment or vehicle is turned ON or in operation) before the failure event occurs. Thus, the smaller the RUL value indicates the closer the component and / or system is to likely experience the failure event. When the failure event of the component or system is determined to be “close”, a detection system can send a warning to an operator, owner and / or maintenance facility of the vehicle or piece of equipment to request a maintenance or replacement of the monitored component or system.

[0038] Systems, computer-readable media and methods described herein enable virtual detectors, using at least one machine learning model, to detect RULs and / or health states ofcomponents or systems in vehicles or pieces of equipment. The health state of a component or system can be defined based on the corresponding RUL. For example, the health state can be defined as an “unhealthy state” when the RUL of the component or system is below a predefined threshold and defined as a “healthy state” when the RUL is greater than the predefined threshold. Systems, computer-readable media, and methods described herein can train at least one machine learning model to estimate the RUL and / or predict the health state of the component or system. The systems and methods described herein can train the machine learning model using a subset of parameters selected from a larger set of available parameters of one or more vehicles or pieces of equipment. The selection can be based on variations of the parameters in response to change in the health state and / or RUL of the component or system.

[0039] Once the machine learning model is trained, the systems and methods described herein can deploy the trained machine learning model to monitor the RUL or health state of the component of system in one or more vehicles. The systems and methods described herein can acquire data corresponding to the selected subset of parameters from a vehicle or other piece of equipment and feed the data as input to the trained machine learning model. In response, the machine learning model can provide a health state and / or RUL of the component or system as an output. The systems and methods described herein can provide or send the health state and / or RUL of the component or system to a dashboard, graphical user interface and / or computing device to trigger a proactive maintenance action before the component or system reaches a failure state.

[0040] While some of the example embodiments described below relate to RUL and / or health state detection for selective catalytic reduction (SCR) systems, particulate filters such as diesel particulate filters (DPFs), and diesel oxidation catalysts (DOCs), it is to be noted that the methods and systems, for training the machine learning models and / or detection of RUL or health state, described herein are applicable to other components of aftertreatment systems or more generally to other components and / or systems of vehicles or pieces of equipment (e.g., power generators). For instance, similar approaches for training or generating machine learning models or virtual sensors and / or using the machine learning models or virtual sensors to detect component RULs and / or health states can be used for various types of components and / or systems, such as diesel exhaust fluid (DEF) dosers,mufflers, power steering systems, catalytic converters, transmission systems, batteries, alternators, radiators, clutches, and brakes among others.

[0041] Technically and beneficially, the systems and methods described herein enable reliable and early detecting / prognosticating of failure conditions for various components, parts or systems of vehicles and / or pieces of equipment. The failure conditions can be detected or predicted before they occur allowing for early and proactive maintenance or repair. Also, the machine learning based approach allows for a relatively high failure detection / prediction accuracy and customization for various types of components, parts or systems of vehicles and / or pieces of equipment. Furthermore, the systems and methods described herein provide computationally efficient failure predictors. In particular, the machine learning models are trained or generated based on a selected set of parameters that are relatively more correlated or more sensitive to a potential future failure and are more effective in predicting future failure events. The selection of a relatively small number of parameters (instead of using a total set of available parameters) reduces the computational complexity of the machine learning models and the training process without jeopardizing prediction or detection efficacy or reliability. Finally, the early detection or prediction of a failure event allows for early servicing (e.g., before the failure event occurs) and leads to increased up-time and longer lifetime of the vehicles and / or pieces of equipment. These and other features and benefits are described more fully herein below.

[0042] Referring now to FIG. 1, a schematic diagram of a computing and network environment 100 for detection of RULs and / or health states of components or systems in vehicles and / or pieces of equipment is shown, according to various example embodiments. In brief overview, the computing and network environment 100 can include a health state detection system 102, one or more vehicles (or pieces of equipment) 104 and one or more computer systems 106. The health state detection system (also referred to herein as RUL detection system or RUL / health state detection system, a health state detection computing system, and / or provider computing system) 102, the one or more vehicles (or pieces of equipment) 104 and the one or more computer systems 106 can be communicatively connected via a communication network 108. The health state detection system 102 can include a data acquisition circuit 110, a virtual detector (or sensor) generator 112 and one or more virtual RUL / health state detectors 114a - 104n, referred to herein either individually or in combination as RUL / health state detector(s) 114, health state detector(s) 114 or RULdetector 114. The vehicle (or piece of equipment) 104 can include, among other components and / or systems, a control system 116 and a telematics unit 118. The computer system 106 can include a graphical user interface (or dashboard) for displaying information, such as data or notifications received from the health state detection system 102.

[0043] The vehicle 104 may be an on-road or an off-road vehicle including, but not limited to, line-haul trucks, mid-range trucks (e.g., pick-up truck), cars (e.g., sedans, hatchbacks, coupes, etc.), buses, vans, refuse vehicles, fire trucks, concrete trucks, delivery trucks, locomotives, marine vehicles, aviation vehicles, and other types of vehicles. In some embodiments, the vehicle 104 shown in FIG. 1 can be a stationary or substantially stationary piece of equipment, such as a power generator, genset, crane, oil rig equipment and / or other types of equipment. In general, embodiments disclosed in the present disclosure can be applicable to vehicles and / or pieces of equipment.

[0044] The vehicle 104 can include at least one control system or controller 116 configured or structured to monitor and / or manage the performance of one or more systems or subsystems of the vehicle, and a telematics unit 118 for communicating with the health state detection system 102 and / or other remote systems or devices.

[0045] The control system 116 may include any type of control system included in a vehicle including, but not limited to, an engine control module or unit (ECM or ECU), a powertrain control module, a transmission control module, an aftertreatment system control module, and / or a combination therewith. The control system 116 can include a processing circuit having one or more processors and one or more memories. The processor may be implemented as a single- or multi-chip processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. The processor may be a microprocessor. The processor also may be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a DSP core, or any other such configuration. The memory 206 (e.g., memory unit and / or storage device) may include one or more devices (e.g., RAM, ROM, Flash memory, hard disk storage) for storing data and / or computer code for completing or facilitating the various processes, layers and modules described in the present disclosure. The memory 206 may be communicablyconnected to the processor 204 to provide computer code or instructions to the processor 204 for executing at least some of the processes described herein. Moreover, the memory 206 may be or include tangible, non-transient volatile memory or non-volatile memory. Accordingly, the memory 206 may include database components, object code components, script components, or any other type of information structure for supporting the various activities and information structures described herein.

[0046] The control system 116 may provide commands and / or instructions to one or more actuators within the vehicle 104 (e.g., timing and quantity of fuel injected, exhaust gas recirculation percentage, turbo-compressor speed, etc.). Further, the control system 116 is also structured to receive, acquire, read, record, keep track of and / or interpret data from one or more components and / or systems in the vehicle 104. As such, the control system 116 may be communicably coupled to one or more sensors included within the vehicle 104. For example, the control system 116 may receive engine speed and engine torque data from each of an engine speed sensor and an engine torque sensor, respectively. The control system 116 may receive measurements from other sensors, such as one or more NOx sensors, a diesel exhaust fluid (DEF) dosing sensor and / or an exhaust system thermometer among other vehicle sensors. For tracking, compartmentalization, and analytics, each piece of data may correspond with a data identifier (DID) (e.g., a code, value, etc.). The control system 116 may also receive data from an on-board diagnostics system (e.g., OBD II, OBD I, EOBD, JOBD, etc.). As such, the control system 116 may receive diagnostic trouble codes (DTCs) based on one or more operating characteristics of a component in the vehicle 104. The DTCs may include fault codes, parameter IDs, etc.

[0047] The telematics unit 118 is communicatively coupled to the control system 116 in the vehicle 104 and to a telematics platform, e.g., associated with or communicatively coupled to the health state detection system 102. The telematics unit 118 may be structured as any type of telematics control unit. Accordingly, the telematics unit 118 may include, but is not limited to, a location positioning system (e.g., global positioning system) to track the location of the vehicle (e.g., latitude and longitude data, elevation data, etc.), one or more memory devices for storing data, one or more electronic processing units for processing the data, and a communication interface for facilitating the exchange of data between the telematics unit 118 and the telematics platform or the health state detection system 102. In this regard, the communication interface may be configured as any type of mobile communications interfaceor protocol including, but not limited to, Wi-Fi, WiMAX, Internet, Radio, Bluetooth, Zigbee, satellite, radio, Cellular, GSM, GPRS, LTE, and the like. The telematics unit 118 may also include a communications interface for communicating with the control system 116 of the vehicle 104. The communication interface for communicating with the control system 116 may include any type and number of wired and wireless protocols (e.g., any standard under IEEE 802, etc.). In some implementations, the communication between the telematics unit 118 and the control system 116 can be via the unified diagnostic services (UDS) protocol.

[0048] In some embodiments, the telematics unit may be excluded / omitted and the controller 116 may include or be coupled to a communication interface structured to enable communications with remote computing systems.

[0049] Components or systems of the vehicle 104 may communicate with each other or remote components using any type and any number of wired or wireless connections. For example, a wired connection may include a serial cable, a fiber optic cable, a CAT5 cable, or any other form of wired connection. Wireless connections may include the Internet, Wi-Fi, cellular, radio, Bluetooth, ZigBee, etc. In one embodiment, a controller area network (CAN) bus provides the exchange of signals, information, and / or data. The CAN bus includes any number of wired and / or wireless connections. In some implementations, the control system 116 communicates with other components of the vehicle 104 via the CAN bus.

[0050] The computer system 106 can include any combination of computing devices, such as computer server(s), desktop(s), laptop(s), tablet(s) and / or handheld device(s) among others. The computer system(s) 106 can be associated with an owner of the vehicle 104 and / or a third party that is monitoring the health state of one or more components of the vehicle 104. For example, the computer system(s) 106 can be associated with an operator of the vehicle, an original equipment manufacturer (OEM), an entity managing a fleet of vehicles 104 and / or a vehicle maintenance facility, among others. The computer system 106 can be communicatively coupled to the health state detection system 102 and / or the vehicle 104 via the communications network 108. The communications network 108 can include the Internet, one or more wireless communications networks, one or more Wi-Fi networks, one or more local area networks (LANs), one or more wide area networks (WANs), a landline network or a combination thereof.

[0051] The health state detection system 102 can include one or more computing devices, such as computer server(s), desktop(s), laptop(s), tablet(s) and / or handheld device(s) among others. In some implementations, the health state detection system 102 or portions thereof can reside on a cloud. According to such implementations, functionalities or operations associated with the health state detection system 102 can be implemented or executed via one or more hardware servers and / or one or more virtual servers. The health state detection system 102 includes a data acquisition circuit 110, a virtual RUL detector generator 112 (also referred to herein as health state detector generator 112 or RUL detector generator 112) and one or more health state detectors 114. Each of the components or circuits 110, 112 and 114 can be implemented as hardware, firmware, software or a combination thereof. For instance, each of the components or circuits 110, 112 and 114 can be implemented, either fully or partially, as executable instructions that are executed by one or more processors. The health state detection system 102 can be owned, operated and / or managed by, or associated with, a provider institution (e.g., of services, equipment, etc.). For example, the provider may be an original equipment manufacturer (OEM). The provider may also provide a service for tracking and maintaining a health of one or more components and / or systems. In this instance, the provider may be an engine manufacturer that also provides various services.

[0052] The data acquisition circuit 110 can be structured or configured to acquire training data, validation data and detection data. The data acquisition circuit 110 can include a communications interface to communicate, e.g., via the communication network 108, with the vehicle 104, the computer system 106 and / or other remote systems or devices. The data acquisition circuit 110 can include one or more circuits to generate requests for data and / or keep track of timing of data requests. The data acquisition circuit 110 can be communicatively coupled to the health state / RUL detector generator 112 and to the health state detector(s) 114. The data acquisition circuit 110 can receive training data, e.g., including validation (or test) data, from the vehicle(s) 104, the computer system 106 or from some other data sources, and provide the training data to the health state detector generator 112. The data acquisition circuit 110 can receive or acquire sensor data from the vehicle(s) 104 and provide the received sensor data to the health state detector(s) 114. The data acquisition circuit 110 can include a memory or database to store training data, validation (or test) data and / or detection data.

[0053] The virtual health state / RUL detector generator 112 can be configured or structured to generate one or more virtual health state / RUL detectors 114 for detecting RULs or health states of one or more specific components and / or system of one or more of the vehicle(s) 104. In particular, the virtual health state / RUL detector generator 112 can receive training data from the data acquisition circuit 110 or a database, and train one or more machine learning models using the training data. The training data can include parameter values (or parameter measurements if measured values from one or more physical sensors) of a set of parameters of the vehicle 104 or subsystems thereof. The virtual health state / RUL detector generator 112 can receive validation (or test) data from the data acquisition circuit 110 or a database thereof and validate or test the trained machine learning model(s) using the validation (or test) data. The validation data can be separate and different from the training data. In some implementations, the virtual health state / RUL detector generator 112 can use a first portion of the training data for training the machine learning model(s) and use a second portion for validating or testing the trained model(s). The machine learning model is structured or configured, once trained, to detect a health state or RUL of a corresponding vehicle component of the vehicle 104 based on parameter values of the vehicle 104.

[0054] The virtual health state / RUL detector generator 112 can preprocess the training data prior to use for training the machine learning model. The preprocessing can include editing the data, labeling data sets and / or filtering the data or selecting a subset of parameters of the vehicle whose parameter values are to be used for training the machine learning model. The virtual health state / RUL detector generator 112 can apply the same preprocessing to the validation (or test) data. A more detailed description of the virtual health state / RUL detector generator 112 is provided below in relation with FIGS. 2 - 9.

[0055] The heath state detector 114 is a virtual detector and can include one or more machine learning models that are trained and validated by the virtual health state / RUL detector generator 112. The heath state detector 114 can receive sensor data (also referred to as detection data), e.g., parameter values of vehicle parameters, from the vehicle 104 via the data acquisition circuit 110 or a database. The heath state detector 114 can feed the received detection data as input to the machine learning model(s) (or virtual detector), and in response, the machine learning model(s) outputs a health state of a component of the vehicle 104, e.g., an indication of whether the RUL of the vehicle component is less than a predefined threshold. The heath state detector 114 can provide or send the indication of the health state of thevehicle component to the vehicle 104, the computer system(s) 106, another remote system or device or a combination thereof. A more detailed description of the health state detector 114 is provided below in relation with FIGS. 2 - 9.

[0056] While FIG. 1 shows a single vehicle 104 and a single computer system 106, in general, the health state / RUL detection system 102 can be structured or configured to monitor vehicle components (e.g., of one or more given types) in a plurality of vehicles (and / or pieces of equipment) 104 and can serve (e.g., report health state detection results to) a plurality of computer systems 106 and / or a plurality of other remote systems or devices. For instance, the heath state / RUL detection system 102 or the virtual detector generator 112 can generate a plurality of health state detectors 114 (or machine learning models) using a plurality of training data sets, e.g., for detecting the RULs or health states of vehicle components of various types. The health state / RUL detection system 102 or the virtual health state / RUL detector generator 112 can generate a separate health state detector 114 (or machine learning model) for each type of vehicle components of the vehicles 104. For instance, each health state detector 114 can be trained or structured to detect the RUL or health state of vehicle components of a given type, e.g., a SCR system or a DOC. In some implementations, the health state / RUL detection system 102 can provide the generated health state detector 114 (or machine learning model) to the vehicle 104, the computer system 106 or to other remote system or device. In such implementations, the health state / RUL detection can be performed by the vehicle 104, the computer system 106 and / or the other remote system or device.

[0057] Referring now to FIG. 2, a schematic diagram of a controller 200 of the health state / RUL detection system 102 of FIG. 1 is shown, according to example embodiments. The controller 200 can represent an example implementation of the health state / RUL detection system 102, or a portion thereof. Additionally, a more detailed block diagram depiction of the vehicle 104 is provided, according to an example embodiment.

[0058] The vehicle 104 is shown to include an engine 222, a fuel system 224 coupled to the engine 222, an aftertreatment system 226 coupled to the engine 222, a transmission system 228 coupled to the engine 222, and one or more sensors 230.

[0059] The engine 222 may be any type of engine, such as a gasoline, natural gas, hydrogen, diesel engine, a hybrid engine (e.g., a combination of an internal combustion engine and anelectric motor), and / or any other suitable engine. In the example depicted, the engine 222 can be a diesel-powered compression-ignition engine.

[0060] The aftertreatment system 226 is in exhaust-gas receiving communication with the engine 222. The aftertreatment system 226 includes a diesel particulate filter (DPF), a diesel oxidation catalyst (DOC), a selective catalytic reduction (SCR) system, and an ammonia slip catalyst (ASC). The DOC is structured to receive the exhaust gas from the engine 222 and to oxidize hydrocarbons and carbon monoxide in the exhaust gas. The DPF is structured to remove particulates, such as soot, from exhaust gas flowing in the exhaust gas stream. The DOC can be fluidly coupled to the exhaust gas conduit system to oxidize hydrocarbons and carbon monoxide in the exhaust gas. In order to properly assist in this reduction, the DOC may be required to be at a certain operating temperature. In some embodiments, this certain operating temperature is approximately between 200-500 °C. In other embodiments, the certain operating temperature is the temperature at which the conversion efficiency of the DOC exceeds a predefined threshold (e.g., the conversion of HC to less harmful compounds, which is known as the HC conversion efficiency). The SCR is configured to assist in the reduction of NOx emissions by accelerating a NOx reduction process between the ammonia and the NOx of the exhaust gas into diatomic nitrogen and water. If the SCR catalyst is not at or above a certain temperature, the acceleration of the NOx reduction process is limited and the SCR may not be operating at a level of efficiency to meet regulations. In some embodiments, this certain temperature is approximately 200-600°C. In some implementations, the aftertreatment system 226 may include one or more additional components or less components.

[0061] The transmission system 228 is structured to transmit power generated by the engine 222 to the wheels via a mechanical system of gears and gear trains. In other implementations, the final drive may be omitted or different relative to the mentioned wheels. The transmission system 228 allows a driver to apply power to vehicle 104 in a controlled manner.

[0062] The sensors 230 can include an engine-out nitrogen oxide (NOx) sensor, a system- out NOx sensor, a DEF dosing sensor, a temperature sensor, a pressure sensor and / or other sensors to monitor operational parameters or states of one or more systems or components of the vehicle 104.

[0063] The controller 200 includes a processing circuit 202 having a processor 204 and a memory 206. The controller 200 includes one or more data acquisition circuits 110, one or more preprocessing (or data preprocessing) circuits 210, one or more model training circuits 212, one or more model validation (or model testing) circuits 214, one or more machine learning model circuits 216, one or more notification (or output) circuits 218 and a communication interface 220. The data acquisition circuit(s) 110, the preprocessing circuit(s) 210, the model training circuit(s) 212, the model validation circuit(s) 214, the machine learning model circuit(s), 216 and / or the notification circuit(s) 218 can be part of the processing circuit 202 or separate components within the controller 200. In some implementations, the preprocessing circuit(s) 210, the model training circuit(s) 212 and the model validation circuit(s) 214 represents circuits of the virtual RUL detector generator 112 shown in FIG. 1. In some implementations, the machine learning model circuit(s) 216 and the notification circuit(s) represent circuits of the health state / RUL detector 114 of FIG. 1.

[0064] The controller 200 is structured or configured to receive vehicle training data from one or more vehicles 104 or pieces of equipment, from the computer system(s) 106 and / or from other remote system(s) or device(s). The training data can include parameter values for a large set of parameters of the vehicle 104. The controller 200 can use the training data to train a machine learning model for detecting health state and / or RUL of a corresponding vehicle component in the vehicle(s) 104. The controller 200 may use a first portion of the training data for training the at least one machine learning model, and use a second portion of the training data to validate or test the trained model. Once the machine learning model is generated (e.g., trained and validated), the controller 200 can use the model to detect health state and / or RUL and predict a “close” failure event of the vehicle component in the vehicle(s) 104 based on detection data received from the vehicle(s) 104. The controller 200 can provide or send an indication of the predicted or detected health state, RUL and / or failure event of the vehicle component to the vehicle 104, the computer system(s) 106 and / or other computer system(s) or device(s).

[0065] In some implementations, the data acquisition circuit(s) 110, the preprocessing circuit(s) 210, the model training circuit(s) 212, the model validation circuit(s) 214, the machine learning model circuit(s) 216 and / or the notification circuit(s) 218 can be embodied or implemented as machine-executable instructions that are stored in the memory 206 (or other memory or storage device) and executed by the processor 204. The data acquisitioncircuit(s) 110, the preprocessing circuit(s) 210, the model training circuit(s) 212, the model validation circuit(s) 214, the machine learning model circuit(s) 216 and / or the notification circuit(s) 218 can be embodied or implemented as machine or computer-readable media storing instructions that are executable by the processor 204. As described herein and amongst other uses, the machine-readable media facilitates performance of certain operations to enable reception and transmission of data. For example, the machine-readable media may provide an instruction (e.g., command, etc.) to, e.g., acquire data. In this regard, the machine- readable media may include programmable logic that defines the frequency of acquisition of the data (or, transmission of the data). The computer readable media instructions may include code, which may be written in any programming language including, but not limited to, Java or the like and any conventional procedural programming languages, such as the "C" programming language or similar programming languages. The computer readable program code may be executed on one processor or multiple remote processors. In the latter scenario, the remote processors may be connected to each other through any type of network (e.g., CAN bus, etc.).

[0066] In some implementations, the data acquisition circuit(s) 110, the preprocessing circuit(s) 210, the model training circuit(s) 212, the model validation circuit(s) 214, the machine learning model circuit(s) 216 and / or the notification circuit(s) 218 can be embodied as hardware units, such as electronic control units. As such, the data acquisition circuit(s) 110, the preprocessing circuit(s) 210, the model training circuit(s) 212, the model validation circuit(s) 214, the machine learning model circuit(s) 216 and / or the notification circuit(s) 218 may be embodied as one or more circuit components including, but not limited to, processing circuitry, network interfaces and / or other circuitry components. In some implementations, the data acquisition circuit(s) 110, the preprocessing circuit(s) 210, the model training circuit(s) 212, the model validation circuit(s) 214, the machine learning model circuit(s) 216 and / or the notification circuit(s) 218 may take the form of one or more analog circuits, electronic circuits (e.g., integrated circuits (IC), discrete circuits, system on a chip (SOCs) circuits, microcontrollers, etc.), telecommunication circuits, hybrid circuits, and any other type of “circuit.” In this regard, the data acquisition circuit(s) 110, the preprocessing circuit(s) 210, the model training circuit(s) 212, the model validation circuit(s) 214, the machine learning model circuit(s) 216 and / or the notification circuit(s) 218 may include any type of component for accomplishing or facilitating achievement of the operations described herein. For example, a circuit as described herein may include one or more transistors, logic gates(e.g., NAND, AND, NOR, OR, XOR, NOT, XNOR, etc.), resistors, multiplexers, registers, capacitors, inductors, diodes, wiring, and so on). The data acquisition circuit(s) 110, the preprocessing circuit(s) 210, the model training circuit(s) 212, the model validation circuit(s) 214, the machine learning model circuit(s) 216 and / or the notification circuit(s) 218 may also include programmable hardware devices such as field programmable gate arrays, programmable array logic, programmable logic devices or the like. The data acquisition circuit(s) 110, the preprocessing circuit(s) 210, the model training circuit(s) 212, the model validation circuit(s) 214, the machine learning model circuit(s) 216 and / or the notification circuit(s) 218 may include one or more memory devices for storing instructions that are executable by one or more processor(s) of these circuits. The one or more memory devices and processor(s) may have the same definition as provided above with respect to the memory 206 and processor 204. In some hardware unit configurations, the data acquisition circuit(s) 110, the preprocessing circuit(s) 210, the model training circuit(s) 212, the model validation circuit(s) 214, the machine learning model circuit(s) 216 and / or the notification circuit(s) 218 may be geographically dispersed throughout separate locations in cloud. In some implementations, the data acquisition circuit(s) 110, the preprocessing circuit(s) 210, the model training circuit(s) 212, the model validation circuit(s) 214, the machine learning model circuit(s) 216 and / or the notification circuit(s) 218 may be embodied in or within a single unit / housing, which is shown as the controller 200 in FIG. 2.

[0067] In the example shown, the controller 200 includes the processing circuit 202 having the processor 204 and the memory device 206. The processing circuit 202 may be structured or configured to execute or implement the instructions, commands, and / or control processes described herein with respect to the data acquisition circuit(s) 110, the preprocessing circuit(s) 210, the model training circuit(s) 212, the model validation circuit(s) 214, the machine learning model circuit(s) 216 and / or the notification circuit(s) 218. The depicted configuration represents the circuit 110 and 210 - 218 as machine or computer-readable media. However, as mentioned above, this illustration is not meant to be limiting as the present disclosure contemplates other embodiments where the circuits 110 and 210 - 218, or at least one circuit of the circuits 110 and 210 - 218, is configured as a hardware unit. All such combinations and variations are intended to fall within the scope of the present disclosure.

[0068] The processing circuit 202 or the processor 204 may be structured or configured to execute or implement instructions, commands, and / or control processes described herein with respect to the data acquisition circuit(s) 110, the preprocessing circuit(s) 210, the model training circuit(s) 212, the model validation circuit(s) 214, the machine learning model circuit(s) 216, and / or the notification circuit(s) 218. The processor 204 may be implemented as a single- or multi-chip processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. The processor 204 may be a microprocessor. The processor 204 also may be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a DSP core, or any other such configuration. In some implementations, the controller 200 may include one or more processors 204, which may be shared by multiple circuits. For example, the data acquisition circuit(s) 110, the preprocessing circuit(s) 210, the model training circuit(s) 212, the model validation circuit(s) 214, the machine learning model circuit(s) 216 and / or the notification circuit(s) 218 may comprise or otherwise share the same processor which, in some example embodiments, may execute instructions stored, or otherwise accessed, via different areas of memory. In some implementations, the one or more processors may be structured to perform or otherwise execute certain operations independent of one or more co-processors. Two or more processors may be coupled via a bus to enable independent, parallel, pipelined, or multithreaded instruction execution. All such variations are intended to fall within the scope of the present disclosure.

[0069] The memory 206 (e.g., memory unit and / or storage device) may include one or more devices (e.g., RAM, ROM, Flash memory, hard disk storage) for storing data and / or computer code for completing or facilitating the various processes, layers and modules described in the present disclosure. The memory 206 may be communicably connected to the processor 204 to provide computer code or instructions to the processor 204 for executing at least some of the processes described herein. Moreover, the memory 206 may be or include tangible, nontransient volatile memory or non-volatile memory. Accordingly, the memory 206 may include database components, object code components, script components, or any other type of information structure for supporting the various activities and information structures described herein.

[0070] The communications interface 220 may include any combination of wired and / or wireless interfaces (e.g., antennas, transmitters, receivers, transceivers, wire terminals) for conducting data communications with various systems, devices, or networks structured to enable communications with vehicles 104, computer systems 106 and / or other remote systems, devices or databases. For example, the communications interface 220 may include an Ethernet card and port for sending and receiving data via an Ethernet-based communications network and / or a Wi-Fi transceiver for communicating via a wireless communications network. The communications interface 220 may be structured to communicate via local area networks or wide area networks (e.g., the Internet) and may use a variety of communications protocols (e.g., IP, LON, Bluetooth, ZigBee, radio, cellular, near field communication).

[0071] The controller 200 can communicate with the vehicle 104, the computer system(s) 106, and / or other computer systems or devices via the communications interface 220 and the communication network 108. For example, the controller 200 can communicate with the control system 116, the telematics unit 118, the engine 222, the fuel system 224 or components thereof, the aftertreatment system 226 or components thereof, the transmission system 228, sensors 230 and / or other systems or components of the vehicle 104. In some implementations, the control system 116 of the vehicle 104 monitors and / or manages various systems of the vehicle 104, and the controller 200 communicates with control system 116 of the vehicle 104 via the telematics unit 118 to acquire data about the systems or components thereof. For example, the control system 116 can collect or record parameter values or measurements from various sensors 230 of the vehicle 104, and the controller 200 can request or receive the collected or recorded parameter values from the control system 116 via the telematics unit 118, the communication network 108 and / or the communication interface 220. In some implementations, the controller 200 can command or instruct the control system 116 regarding parameters for which to send or provide corresponding parameter values. In some implementations, the controller 200 can communicate with various systems of the vehicle 104 or components thereof directly, not via the control system 116 and / or the telematics unit 118.

[0072] The data acquisition circuit(s) 110 is configured or structured to receive training data, validation data and / or detection data from remote systems or devices, such as vehicle 104, computer system(s) 106 or other computer systems or device. The data acquisition circuit(s) 110 corresponds to or can be part of the data acquisition component 110. Thepreprocessing circuit(s) 210, the model training circuit(s) 212 and the model validation circuit(s) 214 are circuits or modules of the virtual health state / RUL detector generator 112. The preprocessing circuit(s) 210 is configured or structured to preprocess training and validation data. The model training circuit(s) 212 is configured or structured to train the machine learning model for detecting the health state and / or RUL of a vehicle component. The model validation circuit(s) 212 is configured or structured to validate or test the machine learning model using validation data. The machine learning model(s) circuit(s) 216 and the notification circuit(s) 218 can be circuits or modules of the health state detector 114. The machine learning model(s) circuit(s) 216 implements the trained (and validated) machine learning model, and is configured or structured to use detection data received from the vehicle 104 to determine a health state and / or RUL of the vehicle component. The notification circuit(s) 218 is structured or configured to send or provide an indication of the detected health state to the vehicle 104, the computer system(s) 106 and / or other remote system(s) or device(s). The functionalities of these circuits are described in further detail below in relation with FIGS. 3 - 9.

[0073] In some implementations, the controller 200 can be a controller of the vehicle(s) 104. In such implementations, the training and validation of the machine learning model can be performed by the health state / RUL detection system 102, e.g., residing on the cloud. As such, the data acquisition circuit(s) 110, the preprocessing circuit(s) 210, the model training circuit(s) 212 and the model validation circuit(s) 214 can be part of the health state / RUL detection system 102. The machine learning model circuit(s) 216 and the notification circuit(s) 218, however, can be implemented in the controller 200 of the vehicle(s) 104. Upon validating the machine learning model(s) and deciding on a final model for deployment (e.g., based on validation results for multiple trained models), the model validation circuit(s) 214 can provide the final machine learning model for deployment in the controller of the vehicle(s) 104. As such, the RUL and / or the health state detection can be performed in the vehicle(s) 104.

[0074] Referring now to FIG. 3, a flow chart of a method 300 for generating or training a virtual health state / RUL detector is shown, according to example embodiments. In brief overview, the method 300 can include data sets corresponding to various RULs of a vehicle component (STEP 302) and assigning to each data set a corresponding health state of the vehicle component based on vehicle component RUL corresponding to the data set (STEP304). The method 300 can include selecting a subset of parameters, from a predefined set of vehicle parameters of the data sets, based on respective variations responsive to change in the health state and / or RUL (STEP 306). The method 300 can include generating, using parameters values of the selected subset of parameters in the data sets, a machine learning model for detecting health state and / or RUL of the vehicle component (STEP 308), and providing the machine learning model for use to detect health state and / or RUL of the vehicle component (STEP 310).

[0075] Referring now to FIGS. 1-3, the method 300 can include the one or more processors 204 or the data acquisition circuit(s) 110 obtaining or acquiring training data sets of the vehicle 104 (STEP 302). The one or more processors 204 or the data acquisition circuit(s) 110 can receive or acquire data sets of one or more vehicles 104. Each data set is recorded from a corresponding vehicle 104 of the one or more vehicles at a corresponding RUL of the vehicle component. Each data set includes parameter values of a predefined set of parameters of the vehicle 104. The processor(s) 204 or the data acquisition circuit(s) 110 can receive training data from a remote data source, such as the computer system(s) 106, a database or the vehicle 104, among other data sources. For example, the computer system(s) 106 can collect parameter values for a set of parameters from one or more vehicles 104 over time. In some implementations, the vehicles 104 can be of the same make and model. In other embodiments, the vehicles 104 can be of different makes and models yet the components and / or systems under investigation are of the same make and model. The computer system(s) 106 can store the collected data in a memory or database of, or coupled to, the computer system(s) 106. The computer system(s) 106 can provide or send the collected data to the controller 200 or the health state / RUL detection system 102 via the communication network 108. The computer system(s) 106 can provide or send the collected data proactively or upon request from the data acquisition circuit(s) 110. In some implementations, the data acquisition circuit(s) 110 can collect the data directly from the vehicle(s) 104 over time, and store the collected data in the memory 206 or a database of the health state / RUL detection system 102. In some implementations, a cloud database can collect and / or maintain parameter values for the predefined set of parameters of the vehicle(s) 104. The data acquisition circuit(s) 110 can acquire the parameter values for the set of parameters from the cloud database.

[0076] Referring now to FIG. 4, a diagram depicting a process 400 of data acquisition or data collection from vehicle 104 is shown, according to an example embodiment. The control system 116, of vehicle 104 collects data from vehicle sensors 230. In some implementations, the control system 116 can collect a separate data set for each trip made by the vehicle 104 (or for each time duration, or another operation interval). As used herein, a trip can be defined as a time interval during which the vehicle 104 is continuously turned on. A new trip starts each time the vehicle 104 is turned on (or power on) and ends when the vehicle is turned off (or powered off).

[0077] Each data set can include parameter values for the predefined set of vehicle parameters collected during a corresponding vehicle trip. A data set can be referred to as embedded field performance analytics (eFPA) data set and includes a set of parameters values for a corresponding trip. The predefined set of vehicle parameters can vary based on the country or state where the vehicle 104 operates. In some implementations, the data set for each vehicle trip can include parameter values for about 450 predefined parameters. In some implementations, the number of predefined vehicle parameters can be about 350 parameters, about 400 parameters or other number. Each data set may include additional information or parameters indicative of health and / or performance trends. The size of each data set can be about 2Kilo bytes. For a given vehicle 104, the control system 116 can collect training data sets for a plurality of trips over a time period until at least a failure event for a vehicle component of interest is detected. For example, the control system 116 can collect data sets from for a plurality of trips, e.g., over months, while the vehicle component to be monitored, e.g., SCR system or DOC, is in a healthy state or condition and until the vehicle component fails. As such, the data sets include at least one data set collected or recorded during one or more trips when the vehicle component is in a failed state and multiple data sets collected or recorded during multiple trips before the vehicle component fails. The training data can include data sets collected or recorded from multiple vehicles 104, where for each vehicle multiple data sets are collected or recorded until failure of the vehicle component of interest is detected or reached. For each vehicle 104, the corresponding data sets include a plurality of data sets recorded over a time period until a failure event or failure state of the vehicle component is detected. At least one of the data sets of the vehicle corresponds to the failure event or failure state of the vehicle component.

[0078] The control system 116 can encrypt recorded data sets and store them in a local memory. The control system 116 can transfer or send the encrypted data sets to the health state / RUL detection system 102 or a cloud system hosting the health state / RUL detection system 102. Data transfer can be achieved using various communication technologies, such telematics, EDGE or mobile applications, e.g., GUIDANZ MOBILE. When using a mobile application, a mobile device can receive data from the control system 116 via BLUETOOTH or other wireless technology. The mobile device can send the data received from the control system 116 via a mobile application installed in the mobile device. When using telematics, the data can be transferred or sent to the health state / RUL detection system 102 via another cloud system 404, e.g., a cloud system hosting the telematics platform. In some implementations, data collected or recorded by the control system 116 can be transferred to the health state / RUL detection system 102 via a computing device 406, e.g., a laptop, that receives the data from the control system 116, e.g., via CLIP protocol, and send the data to the health state / RUL detection system 102, e.g., via the Internet. The control system 116 can transmit or communicate recorded data sets to the health state / RUL detection system 102 on a regular basis, e.g., one or few times daily or weekly. In some implementations, the control can store up to about 750 data sets for about 750 trips, or more. The control system 116 can transmit or communicate recorded data sets to the health state / RUL detection system 102 in real-time or near real-time, where each data set can be transmitted to the health state / RUL detection system 102 immediately after the data set is recorded. The training data received from the control system 116 can be stored in a database 408 of the health state / RUL detection system 102 in encrypted form.

[0079] The set of parameters can include parameters associated with various systems (or components thereof) of the vehicle(s) 104. For instance, the set of parameters can include parameters associated with engine 222, parameters associated with the fuel subsystem 224, parameters associated with the aftertreatment system 224, parameters associated with the transmission system 226, parameters associated with other systems of the vehicle(s) 104 or a combination thereof. The training data received or acquired by the data acquisition circuit(s) 110 can include, for each parameter of the set of parameters, a plurality of corresponding parameter values across multiple data sets (e.g., corresponding to various vehicle trips). Data sets for a given vehicle 105 are recorded at different time instances corresponding to different RULs of the vehicle component of interest (or to be monitored by the health state detector114). As such, each vehicle parameter has respective values in the training data corresponding to different RULs of the vehicle component.

[0080] In some implementations, the set of parameters can include all parameters or almost all of the vehicle(s) 104 for which measurements or values are recorded, e.g., by sensors and / or other components of the vehicle(s) 104. For example, the set of parameters can be a comprehensive set of parameters of the vehicle(s) 104. The set of parameters can include parameters that are correlated with or related to one or more specific systems of the vehicle 104, e.g., fuel system 224, aftertreatment system 226 and / or transmission system 228. The health state / RUL detection system 102 and / or the controller 200 may not initially know which parameters are relevant with respect to detection of the health state and / or RUL of the vehicle component.

[0081] The method 300 can include the one or more processors 204 or the preprocessing circuit(s) 210 of the computing system 102 assigning to each data set a corresponding health state of the vehicle component based on a RUL of the vehicle component corresponding to the data set (STEP 304). This assignment of the health states can be viewed as labeling of the training data sets. In some implementations, the training data sets received by the health state / RUL detection system 102 can be already labeled, where each data set includes an indication of a corresponding health state of the vehicle component. The health state assigned or mapped to each data set is defined based on the RUL of the vehicle component at the time the data set was recorded in the corresponding vehicle 104.

[0082] Referring now to FIG. 5, a diagram 500 depicting how the health state is defined in terms of the RUL is shown, according to an example embodiment. The RUL according to FIG. 5 is defined in terms of total travel distance, e.g., recorded by the odometer, of the vehicle 104. The RUL of the vehicle component is equal to a predefined value (particularly zero) when the vehicle component fails or reaches a failure state. If the vehicle component (e.g., SCR system or POC) fails at a total traveled distance equal to a predefined value, shown as X (which may be expressed in miles or kilometers), the travel distance X can be denoted as the failure travel distance (or vehicle mileage at failure point) and the RUL of the vehicle component is equal to the predefined value (e.g., zero) at the failure travel distance. In other words, the failure travel distance (or vehicle mileage at failure point) for the vehicle component represents a reference point for the RUL of the vehicle component.

[0083] As depicted in FIG. 5, the RUL of the vehicle component before the failure point is defined as the difference between the failure travel distance X and the current total traveled distance of the vehicle 104. For instance, the RUL can be defined or computed as RUL = Failure Kms (Mileage) - trip Kms (Mileage), where trip the Kms (Mileage) represents the total distance traveled by the vehicle up to the corresponding trip or the distance provided by the vehicle odometer at the corresponding trip. For example, if the current total traveled distance (e.g., recorded by the odometer) of the vehicle 104 is 10,000 kilometers less than the failure travel distance X, then the RUL of the vehicle component is 10,000 kilometers (Km). In other words, the RUL represents the additional distance to be traveled or driven by the vehicle before the vehicle component fails or reaches a failure state.

[0084] For training data sets of a vehicle 104, e.g., recorded over a plurality of trips of the vehicle 104, each data set can include an indication of the total traveled distance, e.g., travel distance read from the odometer, at the time the data set is recorded. The data set corresponding to a failure of the vehicle component (e.g., the data set corresponding to a trip when the vehicle component failed or reached a failure state) can be indicated or labeled as a failure data set. The one or more processors 204 or the preprocessing circuit(s) 210 can determine the RUL corresponding to each data set as the difference between the travel distance associated with the failure data set and the travel distance of the data set.

[0085] The one or more processors 204 or the preprocessing circuit(s) 210 can determine or set a threshold for the RUL to indicate that a failure event is “close”. For example, the threshold in FIG. 5 is 10,000 Km. The one or more processors 204 or the preprocessing circuit(s) 210 can define or determine the health state of the vehicle component as “unhealthy” or “unhealthy state” when the RUL of the vehicle component is less than (or less than or equal to) the RUL threshold. When the RUL is greater than the RUL threshold, the one or more processors 204 or the preprocessing circuit(s) 210 can determine or define the health state of the vehicle component as “healthy” or “healthy state”. The one or more processors 204 or the preprocessing circuit(s) 210 can assign to (or label) each data set recorded prior to failure of the vehicle component as either “healthy state” or “unhealthy state” based on the corresponding RUL. The labeling of the data sets with health states of the vehicle component allows for training a machine learning model to predict the health stat of the vehicle component.

[0086] In some implementations, the RUL can be defined as the remaining operation time of the vehicle component before the vehicle component fails or reaches its failure state. The operation time can be defined as the time when the vehicle 104 or piece of equipment is turned on or when the vehicle is moving. In other words, the RUL can be defined in units of time instead of units of distance.

[0087] The one or more processors 204 or the preprocessing circuit(s) 210 can preprocess the data sets to estimate, fill in or populate missing data points. For example, for some vehicle parameters, no corresponding parameter values will be recorded if the parameter or a corresponding event is not triggered. For instance, a data set will not include a derate time if there is no corresponding fault triggered in the aftertreatment system 226. In such case, the one or more processors 204 or the preprocessing circuit(s) 210 can assign a predefined value (e.g., a zero value) to the parameter indicative of derate time. In general, the one or more processors 204 or the preprocessing circuit(s) 210 can clean the data sets so that all parameters have corresponding values in all data sets. In some implementations, the one or more processors 204 or the preprocessing circuit(s) 210 may remove data sets determined to be corrupted (e.g., missing a relatively large number of parameter values or depicting unreasonable parameter values) from the training data.

[0088] The method 300 can include the one or more processors 204 or the preprocessing circuit(s) 210 selecting a subset of parameters of the predefined set of parameters in the data sets, based on respective variations responsive to change in the health state and / or RUL of the vehicle component (STEP 306). In selecting the subset of parameters, the one or more processors 204 or the preprocessing circuit(s) 210 can identify or determine the parameters that depict higher sensitivity to change in the health state and / or RUL of the vehicle component. For instance, vehicle parameters depicting relatively higher variation ranges, responsive to change(s) in the health state and / or RUL of the vehicle component, are better predictors of the health state and / or RUL of the vehicle component.

[0089] The preprocessing circuit(s) 210 may initially filter the training data maintained based on at least one of geographical location, a vehicle make and / or model and / or one or more other characteristics of the vehicle or properties of the data. This initial filtering step may depend on characteristics of the training data available to the health state / RUL detection system. For instance, the initial filtering step may be optional or omitted depending on characteristics or span of the training data maintained and / or provided by the data source.

[0090] The preprocessing circuit(s) 210 can filter the parameters based on cross-correlation or multicollinearity between different parameters of the vehicle(s) 104. The preprocessing circuit(s) 210 can reduce the number of parameters to be considered by eliminating redundancy associated with substantially correlated or collinear parameters of the vehicle(s) 104. The preprocessing circuit(s) 210 can compute cross-correlation metrics (e.g., correlation coefficients) between sequences of parameter values of different parameters of the vehicle(s) 104. The preprocessing circuit(s) 210 can maintain a subset of the parameters (or corresponding sequences of parameter values) of the vehicle(s) 104 having relatively low (e.g., below a predefined threshold) correlation coefficients. For example, if a pair of parameters or corresponding sequences of parameter values have a correlation coefficient above the predefined threshold, the preprocessing circuit(s) 210 can eliminate one of them. The predefined threshold for the correlation coefficient can be a predefined value, such as 0.9, 0.85 or 0.8, among other possible values. Applying the filtering based on the crosscorrelation or multi-collinearity between different parameters leads to reducing the number of parameters (or corresponding sequences or parameter values) from 365 to 205.

[0091] As discussed above, the training data can be labeled per data set (or per trip). For instance, each data set (e.g., apart from data sets associated with failure events) can include an indicator indicative of whether the data set is associated with a “healthy state” or “unhealthy state”. As such, for a given vehicle parameters, the corresponding values across multiple data sets are associated with corresponding health states and / or corresponding RULs of the vehicle component. The preprocessing circuit(s) 210 can assess the variation or range of variation of each parameter as a function of the health state and / or the RUL of the vehicle component of interest. The preprocessing circuit(s) 210 can filter the parameters of the vehicle(s) 104 based on variations or variation ranges of the corresponding values as a function of (or responsive to change in) the RUL and / or the health state of the vehicle component. For a given parameter, if the corresponding values exhibit relatively significant or substantial variations as the health state and / or the RUL of the vehicle component changes, that’s an indication that the parameter depends on or is a relatively good predictor of the health state and / or the RUL of the vehicle component.

[0092] The preprocessing circuit(s) 210 can select the subset of parameters as the parameters having the largest variation ranges, responsive to changes in the RUL and / or health state of the vehicle component, from the predefined set of parameters of the vehicle(s)104. The preprocessing circuit(s) 210 can rank and / or classify the parameters of the vehicle(s) based on the respective variations or variation ranges in relation with changes in the RUL and / or health state of the vehicle component. In some implementations, the preprocessing circuit(s) 210 can use an unsupervised machine learning model to rank and / or classify the parameters of the vehicle(s) 104 based on respective variations or variation ranges responsive to change(s) in the RUL and / or health state of the vehicle component. In some implementations, the preprocessing circuit(s) 210 can use an extreme gradient boosting (XGBoost) algorithm in selecting the subset of parameters with the largest variation ranges responsive to change(s) in the RUL and / or health state of the vehicle component or having the highest impact on the RUL and / or health state of the vehicle component. In particular, the preprocessing circuit(s) 210 can use the XGBoost algorithm to classify and / or rank the parameters based on corresponding variations or variation ranges (responsive to change(s) in the RUL and / or health state of the vehicle component) and / or based on respective impact on the RUL and / or health state of the vehicle component, and select a group of parameters having the highest variation ranges or highest impact. In some implementations, the preprocessing circuit(s) 210 can use other unsupervised machine learning models (e.g., other than the XGBoost algorithm), other ranking or classification algorithms (e.g., not based on learning machine models) or can use some other techniques to identify parameters depicting relatively high variation in response to a change in the RUL and / or health state of the vehicle component. By applying the XGBoost algorithm to the parameter values of the parameters in the training data sets, the preprocessing circuit(s) 210 can identify a relatively smaller number of relevant parameters compared to the total number of parameters in the predefined set.

[0093] Referring now to FIG. 6, a chart 600 depicting Shapley additive explanations (SHAP) values of multiple selected vehicle parameters in relation with the health state of a DOC is shown, according to an example embodiment. In particular, FIG. 6 illustrates the distribution of SHAP values for 20 selected parameters of the vehicle(s) 104 based on field data. The left side of the chart or graph represents SHAP values corresponding to a healthy state of DOC, while the right side of the chart or graph depicts SHAP values corresponding to the unhealthy state of DOC. The range of variation(s) of the SHAP values for a given parameter as the health state of DOC changes is a good indicator of how effective the parameter can be in detecting the health state of DOC. The preprocessing circuit(s) 210 canselect the subset of parameters based on the corresponding SHAP values to train a machine learning model for detecting the health state of DOC.

[0094] Table 1 below depicts a ranked list of 18 parameters of the 20 parameters shown in FIG. 6, with corresponding descriptions. The ranking represents an order of parameter effectiveness with respect in detecting the health state of a DOC. For example, the minimum model base soot load in aftertreatment (MBSLE) parameter is the most predictive parameter (based on SHAP values) of the health state of DOC, the maximum combine soot load in aftertreatment (CSLE) parameter is the second most predictive parameter, the maximum MBSLE is the third most predictive parameter, the average delta pressure soot load estimate (DPSLE) offset adjustment parameter for measuring the ash load adjustment is the fourth most predictive parameter, and the maximum Urea pump pressure (related to DEF pressure) is the fifth maximum predictive parameter. In some implementations, the selected set of parameters can include at least one of these five parameters.Table 1. Parameters for predicting DOC health and / or DOC RUL.

[0095] FIG. 7 shows a chart 620 depicting SHAP values of multiple selected vehicle parameters in relation with the health state of SCR system is shown, according to example embodiments. In particular, FIG. 7 illustrates the distribution of SHAP values for 20 selected parameters of the vehicle(s) 104 in relation with the health state of SCR system. The left side of the chart or graph represents SHAP values corresponding to a healthy state of the SCR system, while the right side of the chart or graph depicts SHAP values corresponding to theunhealthy state of the SCR system. The range of variation(s) of the SHAP values for a given parameter as the health state of the SCR system changes is a good indicator of how effective the parameter can be in detecting the health state of the SCR system. The preprocessing circuit(s) 210 can select the subset of parameters based on the corresponding SHAP values to train a machine learning model for detecting the health state of SCR system.

[0096] Table 2 below depicts a ranked list of 19 parameters of the 20 parameters shown in FIG. 7, with corresponding descriptions. The ranking represents an order of parameter effectiveness with respect in detecting the health state of SCR. For example, the maximum CSLE parameter is the most predictive parameter (based on SHAP values) of the health state of SCR, the maximum MBSLE is the second most predictive parameter, the parameter representing average urea pump command (UL2 System) while pressure is in closed loop but no dosing is the third most predictive parameter, the maximum diesel particulate filter (DPF) delta pressure parameter is the fourth most predictive parameter, and the parameter representing average compressor inlet pressure while equipment (or vehicle) is working is the fifth most predictive parameter. In some implementations, the selected set of parameters can include at least one of these five parameters.Table 2. Parameters for predicting SCR health and / or SCR RUL.

[0097] FIG. 8 shows a chart 620 depicting SHAP values of multiple selected vehicle parameters in relation with the health state and / or RUL of a filter, such as a particular filter(e.g., a diesel particulate filter), is shown, according to example embodiments. In this example, the filter is a DPF. In particular, FIG. 8 illustrates the distribution of SHAP values for 20 selected parameters of the vehicle(s) 104 in relation with the health state and / or RUL of a DPF. The left side of the chart or graph represents SHAP values corresponding to a healthy state of the filter, while the right side of the chart or graph depicts SHAP values corresponding to the unhealthy state of the filter. The range of variation(s) of the SHAP values for a given parameter as the health state of the SCR system changes may be a good indicator of how effective the parameter can be in detecting the health state of the DPF. The preprocessing circuit(s) 210 can select the subset of parameters based on the corresponding SHAP values to train a machine learning model for detecting the health state of the DPF.

[0098] Table 3 below depicts a ranked list of the 20 parameters shown in FIG. 8, with corresponding descriptions. The ranking represents an order of parameter effectiveness with respect in detecting the health state and / or RUL of the filter. For example, a minimum model base soot load parameter (MBSLEMIN) may be the most predictive parameter (based on SHAP values) of the health state and / or RUL of the filter in some situations, the maximum particulate filter pressure change (DPFDPMAX) may be the second most predictive parameter, the maximum pressure change soot load (DPSLEMAX) parameter may be the third most predictive parameter, the mean model base soot load (MBSLEMEAN) parameter may be the fourth most predictive parameter, and the maximum combine soot load (CSLEMAX) parameter may be the fifth most predictive parameter. In some implementations, the selected set of parameters can include at least one of these five parameters.

[0099] This filtering of the parameters based on respective variations or respective SHAP values allows for identifying the parameters of the vehicle(s) 104 that would be the most effective in detecting the health state of the vehicle component of interest. Also, the reduction in the number of parameters to be used for health state detection leads to computationally efficient training of the virtual health state / RUL detector 114, a simpler health state / RUL detector 114 and a significant reduction in the amount of data that is to be communicated to the health state / RUL detection system 102 or the controller 200. For instance, during the detection phase, only parameter values corresponding to the selected parameters can be communicated to the controller 200 or the health state / RUL detection system 102.

[0100] It is to be noted that that the preprocessing circuit(s) 210 can perform any combination of the above-described data preprocessing steps. The preprocessing circuit(s) 210 can perform the combination data preprocessing steps in any order. In other words, the order according to which the data preprocessing steps are described above is not limiting and represents an illustrative example preprocessing order among other possible orders. Also, it is to be noted that the parameters of the vehicle(s) 104 are referred to herein as features of the vehicle(s) 104 or subsystems thereof.

[0101] Referring back to FIGS. 1-3, the method 300 can include the processor(s) 204 or the model training circuit(s) 212 generating a machine learning model for health state and / or RUL detection using parameter values of the selected subset of parameters (STEP 308). The generating of the machine learning model can include the processor(s) 204 or the model training circuit(s) 212 training and validating at least one machine learning model. In some implementations, a first portion of the parameter values of the selected parameters can be used for training the machine learning model(s) by the model training circuit(s) 212, and a second portion can be used for validating the machine learning model(s) by the model validation circuit(s) 214. In some other implementations, the validation data can be obtained from another set of data (e.g., a set of validation data) based on the identified or selected subset of parameters by the preprocessing circuit(s) 210.

[0102] The model training circuit(s) 212 can train one or more machine learning models using the parameter values of the subset of parameters identified or selected by the preprocessing circuit(s) 210. The machine learning model(s) can include a random forest model, a statistical learning model, a neural network (e.g., a deep neural network) and / or a machine learning model of another type. Training a machine learning model can include the model training circuit(s) 212 selecting initial coefficients or parameters of the model. The model training circuit(s) 212 can select an architecture or configuration of the model and assign initial values for the parameters of the model to be estimated. The model training circuit(s) 212 can apply the parameter values of the selected subset of parameters (e.g., the 20 parameters illustrated in FIGS. 6, 7, or 8) as input to the machine learning model, and compare the outputs of the machine learning model(s) (e.g., predicted health state of the vehicle component) to the labels or health states associated with the parameter values of the selected parameters or the corresponding data sets.

[0103] For instance, a first group of parameter values of the selected subset of parameters can be associated with data set corresponding to a healthy state of the vehicle component of interest. When applying the first group of parameter values of the selected subset of parameters as input to the machine learning model, the model training circuit(s) 212 can compare the output to the model to the healthy state. Similarly, a second group of parameter values of the selected subset of parameters can be associated with data set corresponding to a healthy state of the vehicle component. When applying the second group of parameter values of the selected subset of parameters as input to the machine learning model, the model training circuit(s) 212 can compare the output to the model to the unhealthy state. Note that the machine learning model is structured or configured to provide an estimate of health state of the in response to a group of parameter values of the subset of selected parameters provided as input.

[0104] If the output of the machine learning model does not match the health state associated with the input values, the model training circuit(s) 212 can update or modify one or more coefficients or parameters of the machine learning model. The model training circuit(s) 212 can update or modify the coefficient(s) or parameter(s) of the machine learning model separately after each non-matching output of the machine learning model or once after feeding all available groups of parameter values of the selected subset of parameters of the vehicle(s) 104. The model training circuit(s) 212 can iterate the feeding of the groups ofparameter values of the selected subset of parameters of the vehicle(s) 104 and the updating of the coefficients or parameters of the machine learning model until, for example, all outputs of the machine learning model match corresponding labels or health states associated with the input data. In some implementations, the model training circuit(s) 212 can iterate the feeding of the groups of parameter values of the selected subset of parameters of the vehicle(s) 104 and the updating of the coefficients or parameters of the machine learning model until some other convergence criterion (or criteria) is / are satisfied.

[0105] In some implementations, the model training circuit(s) 212 can train multiple machine learning models for health state detection using the parameter values of the selected subset of parameters. For instance, the model training circuit(s) 212 can train multiple random forest models (or random decision forests), e.g., with distinct architectures and / or different numbers of total nodes, using the parameter values of the selected subset of parameters. Training multiple machine learning models provides a plurality of trained models to choose from, e.g., at the validation stage.

[0106] The model validation circuit(s) 214 can validate the trained machine learning model(s) provided by the model training circuit(s) 212, using validation or test data. The validation or test data includes parameter values for the subset of parameters selected by the preprocessing circuit(s) 210. The validation or test data can include some groups of parameter values for the subset of parameters selected by the preprocessing circuit(s) 210 associated with the healthy state of the vehicle component and other groups associated with the unhealthy state of the vehicle component. The validation or test data is different from the data used for training the machine learning model(s). For example, the parameter values for the selected subset of parameters used for validation can be associated with different vehicle trips compared to the parameter values for the subset of parameters used for training the machine learning model(s). In some implementations, the parameter values for the selected subset of parameters used for validation may be recorded at different time instances and / or by different vehicles compared to the parameter values for the selected subset of parameters used for training the machine learning model(s).

[0107] Validating or testing the trained machine learning model(s) can include the model validation model 214 feeding various groups of parameter values for the subset of parameters selected by the preprocessing circuit(s) 210 (e.g., groups of 20 parameter values corresponding to 20 selected parameters), and determining whether the output for each groupmatches the health state for corresponding data set. The validation or testing process provides an assessment of how reliable the trained machine leaning model is in detecting the health state of the vehicle component based on an input data set different of the training data set. In some implementations, the model validation model 214 can validate or test each of a plurality of trained machine learning models, and select the best performing model for deployment to detect the health state of the vehicle component of interest.

[0108] The method 300 can include the processor(s) 204 or the model validation circuit(s) 214 providing the machine learning model for use to detect the health state of the vehicle component of interest (STEP 310). For instance, the processor(s) 204 or the model validation circuit(s) 214 can provide the validated model (or the model selected based validation results for multiple trained models) for deployment to detect the health state of the vehicle component as described below in relation with FIG. 9.

[0109] While the method 300 is described mainly with regard to training and deployment of a machine learning model or virtual detector for detecting the health state of a vehicle component, the same approach can be used to train machine learning models for estimating the RUL of the vehicle component. In some implementations, the health state / RUL detection system 102 or the controller 200 can train a machine learning model to estimate the RUL instead of or in addition to detecting the health state of the vehicle component. The machine learning model can be trained to provide estimates the RUL of the vehicle component as output and the model training circuit(s) 212 can adjust the coefficient and / or parameters of the model based on matching or discrepancies between outputs of the model during training and RUL values associated with different data sets. The preprocessing circuit(s) 210 can select the subset of parameters based on variations of the predefined set of parameters responsive to change(s) in the RUL. In some implementations, the health state / RUL detection system 102 or the controller 200 can train a machine learning model to detect or estimate the RUL and the health state of the vehicle component.

[0110] The method 300 can be used to train machine learning models for detecting or estimating the RUL and / or the health state of various types of vehicle components or systems, such as components of the aftertreatment system and / or other systems of the vehicle 104 or components thereof. In some implementations, the health state / RUL detection system 102 or the controller 200 can train or generate multiple machine learning models for detecting or estimating the RUL and / or the health state of various vehicle component. For example, thehealth state / RUL detection system 102 or the controller 200 can train or generate a first machine learning model for detecting or estimating the RUL and / or the health state of the DOC and a second machine learning model for detecting or estimating the RUL and / or the health state of the SCR system.

[0111] Referring now to FIG. 9, a flow chart of a method 800 of detecting health states of components and / or systems of vehicles 104 and / or pieces of equipment using the virtual health state / RUL detector (or the machine learning model) generated in FIG. 3 is shown, according to example embodiments. In brief overview, the method 800 can include acquiring or receiving parameter values of a plurality of parameters of the vehicle (or piece of equipment) 104 (STEP 802), determining a health state and / or RUL of the vehicle (or piece of equipment) 104 using a machine learning model and the parameter values of the plurality of parameters of the vehicle (or piece of equipment) 104 (STEP 804). The method 800 can include providing an indication of the health state and / or RUL for display on a user interface (STEP 806).

[0112] Referring now to FIGS. 1, 2 and 9, the method 800 can include the processor(s) 204 or the data acquisition circuit(s) 110 acquiring or receiving parameter values of the plurality of parameters of the vehicle (or piece of equipment) 104 (STEP 802). The plurality of parameters for which can be the subset of parameters of the vehicle selected or identified as described above in relation with the preprocessing circuit(s) 210, STEP 306 and FIGS. 6, 7, and / or 8. In other words, the plurality of parameters are identified or selected during the development or training phase of the machine learning model. In some embodiments, the plurality of parameters relate to a vehicle component, such as a diesel oxidation catalyst (DOC), a selective catalytic reduction (SCR) system, or a filter, such as a diesel particulate filter (DPF). In one example arrangement, when the plurality of parameters relate to the DOC, the subset of parameters includes at least one of a minimum model base soot load in aftertreatment (MBSLE) parameter, a maximum combine soot load in aftertreatment (CSLE) parameter, a maximum MBSLE parameter, a parameter representing average delta pressure soot load estimate (DPSLE) offset adjustment, or a maximum urea pump pressure parameter. In another example, when the plurality of parameters relate to the SCR, the subset of parameters includes at least one of a maximum combine soot load in aftertreatment (CSLE) parameter, a maximum model base soot load in aftertreatment (MBSLE) parameter, a parameter representing average urea pump command while pressure is in closed loop but nodosing, a maximum diesel particulate filter (DPF) delta pressure parameter; or a parameter representing average compressor inlet pressure while equipment is working.

[0113] In some implementations, the data acquisition circuit(s) 110 or the virtual detector generator 112 can inform the vehicle(s) 104 about the selected or identified parameters, and the vehicle 104 can send recorded parameter values for the identified or selected subset of parameters to the machine learning model circuit 216. The data acquisition circuit(s) 110 or the machine learning model circuit 216 can specify the identified or selected plurality of parameters in a request for recorded parameter values to detect fuel contamination. The parameter values are recorded by sensors 230 or other components of the vehicle 104 during a trip of the vehicle 104.

[0114] In some implementations, the control system 116 of the vehicle 106 can send the parameter values as discussed in FIG. 4. The control system 116 can collect parameter values of the plurality of parameters per trip and send the parameter values to the health state / RUL detection system 102. In some implementations, the controller 116 can collect data sets of parameter values for vehicle trips made every day and send the collected data sets for a single day together at once.

[0115] The method 800 can include the processor(s) 204 or the machine learning model circuit 216 determining a health state and / or RUL of the vehicle component using the machine learning model and the parameter values of the plurality of parameters of the vehicle 104 (STEP 804). The machine learning model can be the model provided by the model validation circuit 214, and is configured or structured to provide or output an estimate of the health state and / or RUL of the vehicle component when fed the parameter values for the plurality of parameters as input.

[0116] The method 800 can include the processor(s) 204 or the notification circuit(s) 218 providing or sending an indication of the health state and / or RUL of the vehicle component detected or estimated by the machine learning model circuit 216 to a remote computer system for display on a user interface. The remote computer system can include a computer system of the vehicle 104 or the computer system 106. The indication when sent to the vehicle 104 can cause the control system 116 to display a signal indicative of the detected health state and / or RUL on a dashboard of the vehicle 104. When sent to the computer system 106, the indication can be displayed on a user interface and / or can trigger scheduling of a maintenanceof the vehicle if the indication indicates an unhealthy state or a relatively small RUL (e.g., smaller than a predefined threshold).

[0117] In some implementations, the method 800 can include the processor(s) 204 or the notification circuit(s) 218 providing or sending instructions to the vehicle 104 or the control system 116 to automatically trigger an event responsive to detection of an unhealthy state of the vehicle component. For example, the instructions (or commands) can cause the control system 116 to turn on or power a system or device, such as a heater, of the vehicle 104, trigger an error code related to the vehicle component detected to be unhealthy, restrict the vehicle to one or more driving or operational modes, impose a driving speed limit and / or other events (e.g., a derate condition). Such events or actions can be triggered until the vehicle 104 or piece of equipment is serviced and the unhealthy component is replaced or repaired. In some implementations, the instructions (or commands) can cause the control system 116 to display a warning signal on a dashboard of the vehicle 104 until the vehicle 104 or piece of equipment is serviced and the unhealthy component is replaced or repaired.

[0118] Referring to FIGS. 3 and 8, the controller 200 (and / or one or more components thereof, such as the processing circuit 202) is configured to use at least one first machine learning model to identify the subset of parameters that are relevant to the health state and / or RUL of a component (e.g., the DOC, the SCR system, the DPF, etc.), as described herein with respect to the method 300, and at least one second machine learning model to determine the health state and / or RUL of the component, as described herein with respect to the method 800. The first machine learning model may be a classifier model (e.g., XGBoost, etc.) that is used to classify the parameters and identify the most relevant parameters (e.g., based on SHAP values, etc.), as described herein with respect to the method 300. The controller 200 then generates the second machine learning model based on the relevant parameters.), as described herein with respect to the method 300. The second machine learning model may be a random forest model, a statistical learning model, a neural network (e.g., a deep neural network) and / or a machine learning model of another type, such as a regression model. The controller 200 uses the second machine learning model to determine the health state and / or RUL of the component, as described herein with respect to the method 800. Accordingly, the controller 200 uses at least two different machine learning models to determine the health state and / or RUL of the component.

[0119] As utilized herein, the terms “approximately,” “about,” “substantially”, and similar terms are intended to have a broad meaning in harmony with the common and accepted usage by those of ordinary skill in the art to which the subject matter of this disclosure pertains. It should be understood by those of skill in the art who review this disclosure that these terms are intended to allow a description of certain features described and claimed without restricting the scope of these features to the precise numerical ranges provided. Accordingly, these terms should be interpreted as indicating that insubstantial or inconsequential modifications or alterations of the subject matter described and claimed are considered to be within the scope of the disclosure as recited in the appended claims.

[0120] It should be noted that the term “exemplary” and variations thereof, as used herein to describe various embodiments, are intended to indicate that such embodiments are possible examples, representations, or illustrations of possible embodiments (and such terms are not intended to connote that such embodiments are necessarily extraordinary or superlative examples).

[0121] The term “coupled” and variations thereof, as used herein, means the joining of two members directly or indirectly to one another. Such joining may be stationary (e.g., permanent or fixed) or moveable (e.g., removable or releasable). Such joining may be achieved with the two members coupled directly to each other, with the two members coupled to each other using one or more separate intervening members, or with the two members coupled to each other using an intervening member that is integrally formed as a single unitary body with one of the two members. If “coupled” or variations thereof are modified by an additional term (e.g., directly coupled), the generic definition of “coupled” provided above is modified by the plain language meaning of the additional term (e.g., “directly coupled” means the joining of two members without any separate intervening member), resulting in a narrower definition than the generic definition of “coupled” provided above. Such coupling may be mechanical, electrical, or fluidic. For example, circuit A communicab ly “coupled” to circuit B may signify that the circuit A communicates directly with circuit B (i.e., no intermediary) or communicates indirectly with circuit B (e.g., through one or more intermediaries).

[0122] References herein to the positions of elements (e.g., “top,” “bottom,” “above,” “below”) are merely used to describe the orientation of various elements in the FIGURES. It should be noted that the orientation of various elements may differ according to otherexemplary embodiments, and that such variations are intended to be encompassed by the present disclosure.

[0123] While various circuits with particular functionality are shown in FIG. 2, it should be understood that the controller 140 may include any number of circuits for completing the functions described herein. For example, the activities and functionalities of offset circuit(s) 110, gain circuit(s) 210, reductant delivery circuit(s) 212, response lag circuit(s) 214, degradation level circuit(s) 216, diagnosis threshold circuit(s) 218 and / or diagnosis decision circuit(s) 220may be combined in multiple circuits or as a single circuit. Additional circuits with additional functionality may also be included. Further, the controller 140 may further control other activity beyond the scope of the present disclosure.

[0124] As mentioned above and in one configuration, the “circuits” may be implemented in machine-readable medium for execution by various types of processors, such as the processor 204 of FIG. 2. Executable code may, for instance, comprise one or more physical or logical blocks of computer instructions, which may, for instance, be organized as an object, procedure, or function. Nevertheless, the executables need not be physically located together, but may comprise disparate instructions stored in different locations which, when joined logically together, comprise the circuit and achieve the stated purpose for the circuit. Indeed, a circuit of computer readable program code may be a single instruction, or many instructions, and may even be distributed over several different code segments, among different programs, and across several memory devices. Similarly, operational data may be identified and illustrated herein within circuits, and may be embodied in any suitable form and organized within any suitable type of data structure. The operational data may be collected as a single data set, or may be distributed over different locations including over different storage devices, and may exist, at least partially, merely as electronic signals on a system or network.

[0125] While the term “processor” is briefly defined above, the term “processor” and “processing circuit” are meant to be broadly interpreted. In this regard and as mentioned above, the “processor” may be implemented as one or more processors, application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), digital signal processors (DSPs), or other suitable electronic data processing components structured to execute instructions provided by memory. The one or more processors may take the form of a single core processor, multi-core processor (e.g., a dual core processor, triple core processor, quad core processor, etc.), microprocessor, etc. In some embodiments, the one or moreprocessors may be external to the apparatus, for example the one or more processors may be a remote processor (e.g., a cloud-based processor). Alternatively or additionally, the one or more processors may be internal and / or local to the apparatus. In this regard, a given circuit or components thereof may be disposed locally (e.g., as part of a local server, a local computing system, etc.) or remotely (e.g., as part of a remote server such as a cloud-based server). To that end, a “circuit” as described herein may include components that are distributed across one or more locations.

[0126] Although the figures and description may illustrate a specific order of method steps, the order of such steps may differ from what is depicted and described, unless specified differently above. Also, two or more steps may be performed concurrently or with partial concurrence, unless specified differently above. Such variation may depend, for example, on the software and hardware systems chosen and on designer choice. All such variations are within the scope of the disclosure. Likewise, software implementations of the described methods could be accomplished with standard programming techniques with rule-based logic and other logic to accomplish the various connection steps, processing steps, comparison steps, and decision steps.

Claims

WHAT IS CLAIMED IS:

1. A system for training machine learning models for detecting health states of vehicle components, the system comprising: one or more processors; and a memory storing executable instructions, which when executed by the one or more processors, cause the one or more processors to: obtain a plurality of data sets of one or more vehicles, each data set recorded from a corresponding vehicle of the one or more vehicles at a corresponding remaining useful life (RUL) of a vehicle component and including parameter values of a predefined set of parameters of the vehicle; assign to each data set of the plurality of data sets a corresponding health state of the vehicle component based on the corresponding RUL of the vehicle component; select, using the plurality of data sets, a subset of parameters of the predefined set of parameters based on variations of the predefined set of parameters in response to a change in the health state of the vehicle component; train, using parameter values of the subset of parameters in the plurality of the data sets, a machine learning model for detection of the health state of the vehicle component; and provide the machine learning model for use to detect the health state of the vehicle component.

2. The system of claim 1, wherein the vehicle component is a diesel oxidation catalyst (DOC).

3. The system of claim 2, wherein the subset of parameters includes at least one of: a minimum model base soot load in aftertreatment (MBSLE) parameter; a maximum combine soot load in aftertreatment (CSLE) parameter; a maximum MBSLE parameter; a parameter representing average delta pressure soot load estimate (DPSLE) offset adjustment ; or a maximum urea pump pressure parameter.

4. The system of claim 1, wherein the vehicle component is a selective catalytic reduction (SCR) system.

5. The system of claim 4, wherein the subset of parameters includes at least one of: a maximum combine soot load in aftertreatment (CSLE) parameter; a maximum model base soot load in aftertreatment (MBSLE) parameter; a parameter representing average urea pump command while pressure is in closed loop but no dosing; a maximum diesel particulate filter (DPF) delta pressure parameter; or a parameter representing average compressor inlet pressure while equipment is working.

6. The system of claim 1, wherein the vehicle component is a particulate filter.

7. The system of claim 6, wherein the subset of parameters includes at least one of: a minimum model base soot load (MBSLEMIN) parameter; a maximum particulate filter pressure change (DPFDPMAX) parameter; a maximum pressure change soot load (DPSLEMAX) parameter; a mean model base soot load (MBSLEMEAN) parameter; or a maximum combine soot load (CSLEMAX) parameter.

8. The system of claim 1, wherein: the machine learning model is a second machine learning model; and selecting the subset of parameters of the predefined set of parameters is based on using a first machine learning model, different than the second machine learning model, to identify the variations of the predefined set of parameters in response to a change in the health state of the vehicle component.

9. The system of claim 8, wherein the first machine learning model is a classifier model.

10. The system of claim 1, wherein the machine learning model includes at least one of: a random forest model; a statistical learning model; ora neural network.

11. The system of claim 1, wherein the health state of the vehicle component includes an unhealthy state based on the RUL of the vehicle component being less than a predefined threshold and a healthy state based on the RUL of the vehicle component being greater than the predefined threshold.

12. The system of claim 11, wherein the predefined threshold is defined as an additional driving distance to be driven by a vehicle corresponding to the vehicle component before performance of the vehicle component reaches a failure state.

13. A method for training machine learning models for detecting health states of vehicle components, the method comprising: obtaining, by a computer system, a plurality of data sets of one or more vehicles, each data set recorded from a corresponding vehicle of the one or more vehicles at a corresponding remaining useful life (RUL) of a vehicle component and including parameter values of a predefined set of parameters of the vehicle; assigning, by the computer system, to each data set of the plurality of data sets a corresponding health state of the vehicle component based on the corresponding RUL of the vehicle component; selecting, by the computer system, using the plurality of data sets, a subset of parameters of the predefined set of parameters based on variations of the predefined set of parameters in response to a change in the health state of the vehicle component; training, by the computer system, using parameter values of the subset of parameters in the plurality of the data sets, a machine learning model for detection of the health state of the vehicle component; and providing, by the computer system, the machine learning model for use to detect the health state of the vehicle component.

14. The method of claim 13, wherein the vehicle component is at least one of a diesel oxidation catalyst (DOC), a selective catalytic reduction (SCR) system, or a diesel particulate filter (DPF).

15. The method of claim 14, wherein the subset of parameters includes at least one of:a minimum model base soot load in aftertreatment (MBSLE) parameter; a maximum combine soot load in aftertreatment (CSLE) parameter; a maximum MBSLE parameter; a parameter representing average delta pressure soot load estimate (DPSLE) offset adjustment; a maximum urea pump pressure parameter a maximum model base soot load in aftertreatment (MBSLE) parameter; a parameter representing average urea pump command while pressure is in closed loop but no dosing; a maximum diesel particulate filter (DPF) delta pressure parameter; a parameter representing average compressor inlet pressure while equipment is working a maximum pressure change soot load (DPSLEMAX) parameter; a mean model base soot load (MBSLEMEAN) parameter; or a maximum combine soot load (CSLEMAX) parameter.

16. The method of claim 13, wherein the machine learning model is a second machine learning model; and selecting the subset of parameters of the predefined set of parameters is based on using a first machine learning model, different than the second machine learning model, to identify the variations of the predefined set of parameters in response to a change in the health state of the vehicle component.

17. The method of claim 16, wherein the first machine learning model is a classifier model .

18. The method of claim 11, wherein the machine learning model includes at least one of: a random forest model; a statistical learning model; or a neural network.

19. The method of claim 11, wherein the vehicle component is defined to be in the unhealthy state based on the RUL of the vehicle component being less than a predefined threshold.

20. A non-transitory computer-readable medium storing instructions that, when executed by one or more processors of at least one processing circuit, cause the at least one processing circuit to perform operation comprising: obtaining a data set regarding a vehicle at a corresponding remaining useful life (RUL) of a vehicle component and including parameter values of a predefined set of parameters of the vehicle; assigning, to the data set, a corresponding health state of the vehicle component based on the corresponding RUL of the vehicle component; selecting, using the data set, a subset of parameters of the predefined set of parameters based on variations of the predefined set of parameters in response to a change in the health state of the vehicle component; training using parameter values of the subset of parameters in the plurality of the data sets, a machine learning model for detection of the health state of the vehicle component; using the machine learning model to determine the health state of the vehicle component, wherein the health state of the vehicle component is indicative of the RUL of the vehicle component.

Images