SYSTEM AND METHOD FOR IDENTIFYING ROOT CAUSE OF MACHINE ANOMALIES - Patent application

By converting physics-based digital twins into reduced models and using fault models, the system efficiently identifies and diagnoses machine anomalies and their root causes, addressing the limitations of existing condition-based maintenance systems.

JP2024543904APending Publication Date: 2024-11-26CATERPILLAR INC
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
JP2024531159
Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
Priority Date
2021-11-30
Filing Date
2022-11-16
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

Existing condition-based maintenance systems rely heavily on labeled data for anomaly detection and fail to identify the root cause of machine failures, and high-fidelity physics-based digital twin simulations are costly and time-consuming.

Method used

A method and system that converts high-fidelity physics-based digital twins into reduced models (ROMs) using deep learning, combined with fault models to identify operational anomalies and determine their root causes by comparing simulated and actual machine data.

Benefits of technology

Efficiently identifies machine anomalies and their root causes, reducing downtime and costs by leveraging scalable and cost-effective digital twin-based anomaly detection and analysis.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 2024543904000001_ABST
    Figure 2024543904000001_ABST
Patent Text Reader

Abstract

A method for identifying a cause of a machine operation anomaly includes creating a reduced-size model (ROM) for a digital twin model of a selected machine type (502) and feeding current data (504) from a deployed machine (502) to the ROM. The method may include comparing a current output from the selected ROM to a measured output (506) from the current data (504), and determining that an operation anomaly exists when a difference between the current output (508) and the measured output (506) exceeds a selected anomaly threshold. A cause of the operational anomaly may be identified by providing current data (504) to a plurality of fault models (602), each fault model (602) including a particular component fault, comparing (606) fault model outputs (604) from each of the plurality of fault models (602) to measured outputs (506) from the current data (504), selecting the fault model (602) having the fault model outputs (604) that most closely match the measured outputs (506), and indicating (610) the identified component failures associated with the selected fault model as causes of the operational anomaly.
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] This patent application is directed to machine condition monitoring and fault analysis, and more specifically, to identifying anomalies and the causes of those anomalies. [Background technology]

[0002] Equipment maintenance has historically been scheduled using time-based or interval-based criteria, i.e., hours of operation, miles traveled, etc. Advances in in-vehicle sensor technology and telematics have enabled a condition-based maintenance (CBM) approach based on the actual driving environment and the load on the system during the driving period. Most CBM is based on simple rule-based models that combine telematics data with expert knowledge of the monitored system. More recently, advanced data science has been employed to develop machine learning models for predictive maintenance that identify anomalies, indicate impending failures, and quantify the remaining useful life (RUL). However, these models rely heavily on labeled ground truth data, i.e., a clear definition of failure and multiple samples to provide a clear statistical picture of the inputs that predict failure. This is required for each anomaly or failure mode for which a model is developed, and is often difficult to collect. Furthermore, data-based anomaly detection typically does not identify the root cause of the failure, but only identifies that a problem has occurred. Further insight is required to diagnose the root cause and take appropriate corrective action.

[0003] U.S. Patent No. 10,752,269 to Govindappa et al. (hereinafter "Govindappa") describes a system including one or more processors configured to obtain operational parameters of a subsystem of a vehicle configured to travel along a route during a trip. The processor is configured to generate a digital twin of the subsystem based on the operational parameters. The processor is configured to receive simulated performance data generated by execution of the digital twin with a specified model of the subsystem. The processor is further configured to obtain field performance data of the subsystem monitored by one or more sensors mounted on the vehicle during the trip, and determine a performance composite index of the subsystem of the vehicle based on a variance between the simulated performance data and the field performance data. The system may also be configured to determine that the performance composite index of the subsystem is outside a satisfactory health range in response to the variance exceeding a specified variance threshold.

[0004] While Govindappa is able to identify variances between digital twin simulated performance data and field performance data, he is unable to identify the root causes of the variances. Furthermore, high-fidelity physics-based digital twin simulations can be expensive and time-consuming to run at scale. Thus, opportunities remain for improved anomaly identification and root cause analysis. The example systems and methods described herein are directed to overcoming one or more of the above-mentioned deficiencies and / or other problems in the prior art. Summary of the Invention

[0005] In some embodiments, a method for identifying a cause of a machine operation anomaly may include creating a reduced model (ROM) for a digital twin model of a selected machine type and feeding current data from a deployed machine to the ROM. The method may also include comparing a current output from the selected ROM to a measured output from the current data, and determining that an operation anomaly exists when a difference between the current output and the measured output exceeds a selected anomaly threshold. The cause of the operation anomaly may be identified by feeding the current data to a plurality of fault models, each fault model including a particular component failure. The fault model output from each of the plurality of fault models may be compared to the measured output from the current data. The fault model having an output closest to the measured output is selected, and the component failure associated with the selected fault model is displayed and identified as the cause of the operation anomaly.

[0006] According to some aspects, each fault model can be a ROM based on a corresponding digital twin model of the machine that includes a particular component failure. In some aspects, creating a ROM for the digital twin model includes creating a plurality of reduced models for the digital twin model of the selected machine type, feeding historical data from a plurality of machines of the selected machine type to the plurality of ROMs and the digital twin model, comparing an output from the digital twin model to an output from each ROM, and selecting a ROM having an output closest to an output of the digital twin model. In further aspects, the plurality of reduced models can include one or more of a gradient boosting model, a regression model, a recurrent neural network model, or a long-short-term memory model. According to some aspects, the method can further include receiving current data from a plurality of sensors on the deployed machine. In some aspects, the method can further include identifying a machine subsystem associated with the operational anomaly. In some aspects, the plurality of fault models is based on a subsystem model associated with the identified machine subsystem.

[0007] In some embodiments, a system for identifying a cause of a machine operation anomaly may include one or more processors and one or more memory devices having instructions stored thereon. When the instructions are executed, the processor creates a reduced model (ROM) for a digital twin model of a selected machine type. The instructions may also cause the processor to receive current data from a plurality of sensors on the deployed machine and provide the current data to the ROM. A current output from the selected ROM is compared to a measured output from the current data, and an operation anomaly is determined to exist when a difference between the current output and the measured output exceeds a selected anomaly threshold. A cause of the operation anomaly may be identified by providing the current data to a plurality of fault models, each fault model including a failure of a particular component. A fault model output from each of the plurality of fault models may be compared to the measured output from the current data. A fault model having an output closest to the measured output is selected, and a component failure associated with the selected fault model is identified as a cause of the operation anomaly.

[0008] According to some aspects, the system may further include a plurality of sensors on the deployed machine. hi some aspects, the current data from the plurality of sensors is received via a satellite network.

[0009] In some embodiments, one or more non-transitory computer-readable media may store computer-executable instructions that, when executed by one or more processors, cause the one or more processors to perform operations. The operations may include creating a reduced model (ROM) for the digital twin model of the selected machine type and feeding current data from the deployed machine to the ROM. The operations may also include comparing a current output from the selected ROM with a measured output from the current data, and determining that an operational anomaly exists when a difference between the current output and the measured output exceeds a selected anomaly threshold. A cause of the operational anomaly may be identified by feeding the current data to a plurality of fault models, each fault model including a failure of a particular component. A fault model output from each of the plurality of fault models may be compared to the measured output from the current data. A fault model having an output closest to the measured output is selected, and a component failure associated with the selected fault model is identified as a cause of the operational anomaly. [Brief description of the drawings]

[0010] The systems and methods described herein can be better understood by reference to the following detailed description in conjunction with the accompanying drawings, where like reference numbers indicate identical or functionally similar elements.

[0011] [Figure 1] FIG. 1 illustrates an overview of an environment in which some embodiments of the disclosed technology can operate. [Diagram 2] FIG. 2 is a block diagram illustrating an overview of a machine anomaly identification and cause analysis system in accordance with some embodiments of the disclosed technology. [Diagram 3] FIG. 3 is a diagram illustrating a digital twin anomaly detection system according to an embodiment of the disclosed technology. [Figure 4] FIG. 4 illustrates a system for developing a reduced model for detecting machine anomalies in accordance with some embodiments of the disclosed technology. [Diagram 5]FIG. 5 is a diagram illustrating a reduced model anomaly detection system in accordance with some embodiments of the disclosed technology. [Figure 6] FIG. 6 is a diagram illustrating a system for determining a cause of a machine operation anomaly in accordance with some embodiments of the disclosed technology. [Figure 7A] FIG. 7A is a graph of actual and simulated machine coolant temperatures during typical nominal operating conditions. [Figure 7B] FIG. 7B is a graph of actual and simulated machine coolant temperatures illustrating an operating anomaly. [Figure 7C] FIG. 7C is a graph of actual machine coolant temperatures and simulated temperatures with specific component failures. [Figure 8] FIG. 8 is a flow diagram illustrating a method for detecting machine anomalies using reduced models, in accordance with some embodiments of the disclosed technology. [Figure 9] FIG. 9 is a flow diagram illustrating a method for developing a reduced model of a digital twin for detecting and analyzing machine anomalies according to some embodiments of the disclosed technology. [Figure 10] FIG. 10 is a flow diagram illustrating a method for determining a cause of a machine operation anomaly in accordance with some embodiments of the disclosed technology. [Figure 11] FIG. 11 is a block diagram illustrating an overview of an apparatus in which some implementations may operate. [Figure 12] FIG. 12 is a block diagram illustrating an overview of an environment in which some implementations may operate. [Figure 13] FIG. 13 is a block diagram illustrating components that may be used in a system employing the disclosed technology in some implementations.

[0012] Headings provided herein are merely for convenience and do not necessarily affect the scope of the embodiments. Moreover, the drawings are not necessarily to scale. For example, the dimensions of some of the elements in the figures may be enlarged or reduced to help improve understanding of the embodiments. Moreover, while the disclosed technology is susceptible to various modifications and alternative forms, specific embodiments are illustrated in the drawings and described in detail below. However, it is not intended to unnecessarily limit the described embodiments. Rather, the embodiments are intended to cover all suitable modifications, combinations, equivalents, and alternatives falling within the scope of the present disclosure. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0013] Various embodiments of the above-mentioned systems and methods will now be described in more detail. The following description provides specific details for a thorough understanding and enabling description of these embodiments. However, those skilled in the art will understand that the techniques and technologies discussed herein may be practiced without many of these details. Similarly, those skilled in the art will understand that the technology may include many other features not described in detail herein. Furthermore, some known structures or functions may not be shown or described in detail below to avoid unnecessarily obscuring the relevant description.

[0014] The terms used below, while being used in conjunction with the detailed description of some specific examples of embodiments, are to be interpreted in their broadest reasonable sense. Indeed, although some terms may be emphasized below, any term that is intended to be interpreted in any limited manner is expressly and specifically defined as such in this section.

[0015] A method and system for machine anomaly identification and causal analysis is disclosed. Physics-based machine and system simulations created for product development can be leveraged for condition monitoring application improvement. The physics-based simulation model is calibrated to closely match the normal behavior of the physical system and is defined as a digital twin of the physical system. The digital twin calibrated for normal behavior can be converted to a lower order model, e.g., a reduced order model (ROM), based on deep learning, to improve the efficiency and cost of baseline comparison. Telematics data from the target system is used as input for the digital twin-based ROM. If the difference in output between the ROM and the physical system is within a defined threshold (e.g., within 10 percent of the ROM), the system continues to be monitored. Otherwise, if the difference in output exceeds the defined threshold (either below or above), further simulations are performed using a physics-based digital twin-based ROM seeded with models representing different failures, i.e., failures due to different root causes. These "seeded failure" ROMs can be multiple and represent different levels of failure for each root cause, e.g., the magnitude of a water leak can be varied as a root cause. The output from the seeded fault digital twin based ROM is compared to the telematics output from the physical system. The system can identify the seeded fault digital twin that most closely matches the physical output and identify the root cause of the failure. In some implementations, a single digital twin can identify a wide range of failures and root causes. For example, a digital twin of an engine system can identify failures in the air system, fuel system, cooling system, etc. and diagnose the root cause.

[0016] 1 illustrates an environment 10 in which some implementations of a machine anomaly identification and causal analysis system 100 may operate, according to an embodiment of the disclosed technology. The system environment 10 may include multiple machines, such as excavators 20(1) and 20(2), a satellite 12, a telematics database 102, a physics-based digital twin model system 104, a ROM system 106, a fault model system 108, and a network 110. The machine anomaly identification and causal analysis system 100 may be connected to the telematics database 102, the physics-based digital twin model system 104, the ROM system 106, and the fault model system 108 via the network 110. The telematics database 102 and the machine anomaly identification system 100 may receive telematics data from the excavators 20(1) and 20(2) via the satellite 12. The telematics data may include sensor data from the excavators, such as pressure sensors 22, vibration sensors 24, and temperature sensors 26, to name a few.

[0017] As shown in FIG. 2, the machine anomaly identification and causal analysis system 100 may include a ROM development module 120, an anomaly detection module 130, and a causal analysis module 140. In some embodiments, the ROM development module 120 is configured to convert a high-order accuracy model, such as a physics-based digital twin (e.g., digital twin model 104), into a scalable ROM (e.g., ROM 106). The module 120 can also statistically compare the ROM and the digital twin output and select the best ROM for each digital twin model. The ROM development module 120 can use various techniques for ROM development, such as gradient boosting, regression, recurrent neural networks, long-short-term memory models, and optimization, to name a few.

[0018] In some embodiments, the anomaly detection module 130 may be configured to provide current data, either received directly from a deployed machine (e.g., excavator 20(1)) or stored in the telematics database 102, to a selected ROM. The module 130 may also compare an output from the ROM to the current data and determine that an operating anomaly exists when the difference between the current output and the measured output exceeds a selected anomaly threshold.

[0019] In some embodiments, the causal analysis module 140 may be configured to identify a cause of the operational anomaly by feeding current data from the deployed machine to a number of fault models (e.g., fault model 108). Each fault model may include a particular fault, such as a component failure. The module 140 may compare the fault model output from each of the multiple fault models to a measured output from the current data. The fault model having an output closest to the measured output is selected, and a component failure associated with the selected fault model is identified as the cause of the operational anomaly.

[0020] FIG. 3 illustrates a digital twin anomaly detection system 300 according to an embodiment of the disclosed technology. In the anomaly detection system 300, a simulation engineer can calibrate a physics-based simulation model (i.e., digital twin 306) to closely match the normal behavior of a physical system. Telematics data 308 from the target system (i.e., machine 304) is used as input for the digital twin 306. If the difference in output (e.g., comparison 310) between the digital twin and the physical system is less than a defined threshold, the system is operating normally 312 and monitoring of the system can continue. Otherwise, if the difference in output exceeds (either below or above) a defined threshold, an anomaly 314 is identified. High-order accuracy physics-based digital twins can be costly and time-consuming to implement at scale. Furthermore, the anomaly detection system 300 only identifies that a problem has occurred, rather than identifying the root cause of the failure. For example, the embodiments described below with respect to FIGS. 4-6 improve the digital twin anomaly detection system 300 by converting the digital twin into a ROM and analyzing the anomaly to identify its cause.

[0021] FIG. 4 illustrates a system 400 for creating a ROM for a digital twin model of a selected machine type 402 (e.g., a dump truck) to detect machine operation anomalies, according to some embodiments of the disclosed technology. Creating a ROM for the digital twin model may include creating multiple candidate reduced models 410(1) and 410(2) for a digital twin model 406 of the selected machine type 402. Historical data 404 is provided from multiple machines of the selected machine type to multiple ROMs 410 and the digital twin model 406. Outputs 408 from the digital twin model are compared 414 to outputs 412(1) and 412(2) from ROMs 410(1) and 410(2), respectively. The model outputs may be statistically compared using a variety of suitable methods, including distance measurements. The ROM having an output that most closely matches the output of the digital twin model may be selected as the optimal ROM model 416 for use in anomaly detection. In some embodiments, the multiple reduced models 410 may include one or more of a gradient boosting model, a regression model, a recurrent neural network model, or a long short-term memory model.

[0022] FIG. 5 illustrates a reduced model anomaly detection system 500 according to some embodiments of the disclosed technology. The anomaly detection system 500 can detect an operational anomaly of a deployed machine 502 by feeding current data 504 (e.g., live or real-time data) from the deployed machine to a ROM 416 selected by the system 400 as described above. A current simulated output 508 from the selected ROM 416 is compared 510 to a measured output 506 (e.g., coolant temperature, oil pressure, etc.) from the current data 504. If the difference between the current output 508 and the measured output 506 exceeds a selected anomaly threshold, an operational anomaly 514 is determined to exist. In some embodiments, the threshold is exceeded when the mean absolute percentage error (MAPE) between the actual (measured output) and the simulated (current output) is greater than 0.10 (10%). If an anomaly is detected, the cause of the anomaly can be analyzed as described below with respect to FIG. 6. Otherwise, if the system is operating normally, the anomaly detection system 500 can continue to monitor the current data 504 for any anomalies. In some embodiments, a machine subsystem associated with an operational anomaly may be identified, for example, if there is an anomaly in engine coolant temperature, the cooling subsystem of the machine may be identified as needing analysis.

[0023] FIG. 6 illustrates a system 600 for determining the cause of a machine operation anomaly, according to some embodiments of the disclosed technology. As described above, when the anomaly detection system 500 detects an anomaly, the causal analysis system 600 can help identify the anomaly cause. The causal analysis system 600 can identify the cause of the operation anomaly by feeding the current data 504 (e.g., live or real-time data) to a number of fault models 602(1), 602(2), 602(3). In some embodiments, the multiple fault models are based on subsystem models associated with the machine subsystems identified by the anomaly detection system 500, as described with respect to FIG. 5. Each fault model 602 includes a specific component or system failure or degradation. Taking the use of an engine cooling system as an example, potential faults may include a thermostat stuck (e.g., open or closed), a radiator leak (e.g., large or small leak), a water pump failure, etc. The system 600 compares 606 the output 604 from each of the multiple fault models 602 to the measured output 506 from the current data 504. The fault model 602 having the fault model 604 closest to the measurement output 506 is selected, and a failure of the component associated with the selected fault model is identified as the cause 608 of the operational anomaly. The cause of the anomaly can then be communicated (e.g., displayed) to appropriate personnel 610, e.g., a machine operator, a maintenance technician, and / or an equipment dealer.

[0024] 7A-C are exemplary graphs of actual machine coolant temperature 702 and simulated machine coolant temperature 704 illustrating normal operation (FIG. 7A), operation anomaly (FIG. 7B), and root cause analysis comparison (FIG. 7) of engine coolant temperature (ECT). In FIG. 7A, the simulated or predicted output 704 and the measured output 702 of the ECT are very similar and do not exceed the anomaly threshold. Thus, the cooling system is operating normally. However, in FIG. 7B, there is a large difference between the simulated output 704 and the measured output 702. In this case, the measured output 702 is below the temperature expected based on the simulated output 704. The difference exceeds the anomaly threshold, indicating an anomaly. In FIG. 7C, a fault model simulation output 706 of a failed thermostat (e.g., stuck open) is shown, which matches well with the measured output 702, thereby indicating that the detected anomaly is likely due to a thermostat stuck in the open position.

[0025] FIG. 8 is a flow diagram illustrating a method 800 for detecting machine anomalies using reduced models, according to some embodiments of the disclosed technology. The method 800 may include, at step 802, creating a reduced model (ROM) for a digital twin model of a selected machine type. At step 804, current data is received from sensors on the deployed machine, and at step 806, the current data is provided to the ROM. At step 808, the current output from the ROM is compared to the measured output from the current data. At step 810, the method determines that an operational anomaly exists if the difference between the current output and the measured output exceeds a selected anomaly threshold, and at step 812, a cause of the operational anomaly is identified. At step 810, if the difference between the current output and the measured output does not exceed the anomaly threshold, the method loops to step 804 and continues monitoring the machine.

[0026] 9 is a flow diagram illustrating a method 802 for creating a reduced model of a digital twin for detecting machine anomalies, according to some embodiments of the disclosed technology. The method 802 may include creating multiple reduced models for a digital twin model of a selected machine type, at step 902, and providing historical data from multiple machines of the selected machine type to multiple ROMs and the digital twin model, at step 904. The method may compare an output from the digital twin model with an output from each ROM, at step 906, and select a ROM having an output closest to the output of the digital twin model, at step 908.

[0027] FIG. 10 is a flow diagram illustrating a method 812 for determining a cause of a machine operation anomaly according to some embodiments of the disclosed technology. The method 812 may include creating a plurality of fault models at step 1002. Each fault model includes a specific component failure. The method 812 may also include feeding the current data to the plurality of fault models at step 1004, as well as comparing the fault model output from each of the plurality of fault models with the measured output from the current data at step 1006. At step 1008, the method selects the fault model having the closest fault model to the measured output, and at step 1010, a component failure associated with the selected fault model is identified as the cause of the operation anomaly. In some embodiments, each fault model is a ROM based on a corresponding digital twin model of the machine or a subsystem of the machine including the specific component failure.

[0028] The right system The techniques disclosed herein may be embodied as dedicated hardware (e.g., circuits), as programmable circuits appropriately programmed with software and / or firmware, or as a combination of dedicated and programmable circuits. Thus, embodiments may include a machine-readable medium having stored instructions that can be used to cause a computer, microprocessor, processor, and / or microcontroller (or other electronic device) to perform a process. Machine-readable media may include, but are not limited to, optical disks, compact disk read-only memories (CD-ROMs), magneto-optical disks, ROMs, random access memories (RAMs), erasable programmable read-only memories (EPROMs), electrically erasable programmable read-only memories (EEPROMs), magnetic or optical cards, flash memories, or other types of media / machine-readable media suitable for storing electronic instructions.

[0029] Some implementations are discussed in more detail below with reference to the figures. FIG. 11 is a block diagram showing an overview of an apparatus in which some implementations of the disclosed technology may operate. The apparatus may comprise, for example, hardware components of an apparatus 1100 for performing machine anomaly detection and analysis. The apparatus 1100 may include one or more input devices 1120 that provide input to a CPU (processor) 1110 to inform actions. The actions are typically mediated by a hardware controller that interprets signals received from the input devices and communicates the information to the CPU 1110 using a communication protocol. The input devices 1120 include, for example, a sensor, a mouse, a keyboard, a touch screen, an infrared sensor, a touch pad, a wearable input device, a camera or image-based input device, a microphone, or other user input device.

[0030] The CPU 1110 may be a single processing unit or multiple processing units in a device, or may be distributed across multiple devices. The CPU 1110 may be coupled to other hardware devices using a bus, such as a PCI bus or a SCSI bus. The CPU 1110 may communicate with a hardware controller for the device, such as a display 1130. The display 1130 may be used to display text and graphics. In some embodiments, the display 1130 provides graphical and textual visual feedback to the user. In some implementations, the display 1130 includes an input device as part of the display, such as when the input device is a touch screen or equipped with a line-of-sight monitoring system. In some implementations, the display is separate from the input device. Examples of display devices include LCD display screens, LED display screens, projected displays, holographic displays, or augmented reality displays (such as head-up displays or head-mounted devices). Other I / O devices 1140 may also be coupled to the processor, such as a network card, a video card, an audio card, USB, FireWire or other external devices, sensors, cameras, printers, speakers, CD-ROM drives, DVD drives, disk drives, or Blu-ray devices.

[0031] In some implementations, the device 1100 also includes a communication device capable of communicating with a network node on a wireless or wired basis. The communication device can communicate with another device or a server over a network, for example, using the TCP / IP protocol. The device 1100 can utilize the communication device to distribute operations across multiple network devices.

[0032] The CPU 1110 may have access to memory 1150. Memory may include one or more of a variety of hardware devices for volatile and non-volatile storage, and may include both read-only and writeable memory. For example, memory may include random access memory (RAM), CPU registers, read-only memory (ROM), and writeable non-volatile memory, such as flash memory, hard drives, floppy disks, CDs, DVDs, magnetic storage devices, tape drives, device buffers, and the like. Memory is not a propagating signal separate from the underlying hardware, and thus memory is non-transient. Memory 1150 may include program memory 1160 that stores programs and software, such as an operating system 1162, a machine anomaly analysis platform 1164, and other application programs 1166. Memory 1150 may also include data memory 1170, which may include database information, and the like, that may be provided to the program memory 1160 or any element of the device 1100.

[0033] Some implementations may be operational with numerous other general purpose or special purpose computing system environments or configurations. Examples of known computing systems, environments, and / or configurations that may be suitable for use with the present technology include, but are not limited to, personal computers, server computers, handheld or laptop computers, cellular or mobile telephones, wearable electronic devices, game consoles, tablet devices, microprocessor systems, microprocessor-based systems, programmable consumer electronics, network PCs, microcomputers, mainframe computers, distributed computing environments including any of the above mentioned systems or devices, and the like.

[0034] 12 is a block diagram illustrating an overview of an environment 1200 in which some implementations of the disclosed technology may operate. The environment 1200 may include one or more client computing devices 1205A-D, examples of which may include device 1100. The client computing devices 1205 may operate in a networked environment using logical connections to one or more remote computers, such as a server computing device 1210, via a network 1230.

[0035] In some implementations, server computing device 1210 may be an edge server that receives client requests and coordinates fulfillment of those requests via other servers, such as servers 1220A-C. Server computing devices 1210 and 1220 may include computing systems, such as device 1100. Although each server computing device 1210 and 1220 is logically represented as a single server, each server computing device may be a distributed computing environment that encompasses multiple computing devices located at the same or different geographic physical locations. In some implementations, each server computing device 1220 corresponds to a group of servers.

[0036] Each of the client computing device 1205 and the server computing devices 1210 and 1220 may act as a server or a client to other server / client devices. The server 1210 may be connected to a database 1215. Each of the servers 1220A-C may be connected to a corresponding database 1225A-C. As described above, each of the servers 1220 may correspond to a group of servers, each of which may share a database or have its own database. The databases 1215 and 1225 may contain (e.g., store) information. Although the databases 1215 and 1225 are logically depicted as a single unit, each of the databases 1215 and 1225 may be a distributed computing environment that encompasses multiple computing devices and may be located within their corresponding servers or may be located in the same or different geographic physical locations.

[0037] The network 1230 may be a local area network (LAN) or a wide area network (WAN), but may also be other wired or wireless networks. The network 1230 may be the Internet, or some other public or private network. The client computing devices 1205 may be connected to the network 1230 through a network interface, such as by wired or wireless communication. Although the connections between the servers 1210 and 1220 are shown as separate connections, these connections may be any type of local, wide area, wired, or wireless network, including the network 1230 or separate public or private networks.

[0038] 13 is a block diagram illustrating components 1300 that may be used in a system employing the disclosed technology in some implementations. The components 1300 include hardware 1302, general-purpose software 1320, and specialized components 1340. As mentioned above, a system implementing the disclosed technology may use a variety of hardware, including a processing unit 1304 (e.g., CPU, GPU, APU, etc.), working memory 1306, storage memory 1308, and input and output devices 1310. The components 1300 may be implemented on a client computing device, such as client computing device 1205, or a server computing device, such as server computing device 1210 or 1220.

[0039] The general-purpose software 1320 may include various applications, including an operating system 1322, local programs 1324, and a basic input output system (BIOS) 1326. The specialized components 1340 may be subcomponents of the general-purpose software applications 1320, such as the local programs 1324. The specialized components 1340 may include components that can be used to control data transfer and specialized components, such as a ROM development module 1344, an anomaly detection module 1346, a cause analysis module 1348, a telematics module 1350, and an interface 1342. In some implementations, the components 1300 may be in a computing system that is distributed across multiple computing devices or may be an interface to a server-based application that executes one or more of the specialized components 1340.

[0040] Those skilled in the art will appreciate that the components illustrated in Figures 11-13 above, and each of the flow diagrams described above, may be modified in various ways. For example, the order of logic may be rearranged, sub-steps may be performed in parallel, illustrated logic may be omitted, or other logic may be included. In some implementations, one or more of the above-described components may perform one or more of the processes described herein. [Industrial Applicability]

[0041] In some embodiments, the machine anomaly identification and cause analysis system may include a ROM development module 1344, an anomaly detection module 1346, and a cause analysis module 1348 (FIG. 13). In operation, the ROM development module 1344 may select a reduced model (ROM) for the digital twin model of the selected machine type running on the digital twin model system 104. The module may create multiple reduced models running on the ROM system 106 for the digital twin model and feed historical data stored in the telematics database 102 from multiple machines of the selected machine type to the multiple ROMs and the digital twin model. The module may compare the output from the digital twin model with the output from each ROM and select the ROM with the output closest to the output of the digital twin model. The selected ROM may be used in place of the digital twin for anomaly detection and failure analysis.

[0042] The anomaly detection module 1346 can receive current data from the sensors 22, 24, and 26 on the deployed machine 20(1) and / or from the telematics database 102 and feed the data into a ROM created / selected by the ROM development module 1344. The module compares the current output from the ROM received from the ROM system 106 with the measured output from the current data. The module determines that an operating anomaly exists if the difference between the current output and the measured output exceeds a selected anomaly threshold.

[0043] The cause analysis module 1348 can identify the cause of an operational anomaly by feeding current data received from the sensors 22, 24, and 26 and / or from the telematics database 102 to a plurality of fault models on the fault model system 108. Each fault model can include a specific component failure. The module compares the fault model output from each of the plurality of fault models received from the fault model system 108 to the measurement output from the current data. The module selects the fault model having the fault model output closest to the measurement output and identifies the component failure associated with the selected fault model as the cause of the operational anomaly. The detected anomaly and the identified cause can be reported, for example, to a machine operator, a maintenance technician, and / or a dealer. Thus, the disclosed machine anomaly identification and cause analysis system can help to quickly diagnose and resolve problems on a machine, thereby reducing downtime and maximizing machine profitability. Additionally, the disclosed system and method can identify anomalies and alert personnel before more costly problems occur. For example, the system can identify anomalies (e.g., overheating) in the engine cooling system before damage to the engine occurs.

[0044] remarks The above description and drawings are illustrative and should not be construed as limiting. Numerous specific details are described to provide a thorough understanding of the present disclosure. However, in some instances, well-known details are not described to avoid obscuring the description. Furthermore, various modifications may be made without departing from the scope of the embodiments.

[0045] References herein to "one embodiment" or "embodiment" mean that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the present disclosure. The appearances of the phrase "in one embodiment" in various places in this specification do not necessarily all refer to the same embodiment, nor are they separate or alternative embodiments that exclude other embodiments from one another. Furthermore, various features are described that are shown in some embodiments but not in other embodiments. Similarly, various requirements are described that may be required in some embodiments but not in other embodiments.

[0046] The terms used herein generally have their ordinary meanings in the art, in the context of the disclosure, and in the specific context in which each term is used. It will be understood that the same thing can be said in more than one way. Thus, for any one or more of the terms discussed herein, alternative phrases and synonyms can be used, and there is no particular importance to whether the term is detailed or discussed in the present specification. Synonyms are provided for some terms. The description of one or more synonyms does not exclude the use of other synonyms. The use of examples anywhere in the present specification, including examples of any term discussed herein, is merely illustrative and is not intended to further limit the scope and meaning of the present disclosure or the scope and meaning of any exemplary term. Similarly, the present disclosure is not limited to the various embodiments provided herein. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present disclosure pertains. In case of conflict, the present specification, including definitions, will control.

Claims

1. A method (800) for identifying a cause of a machine operation anomaly, the method comprising: Creating (802) a reduced-size model (ROM) (416) for a digital twin model (406) of a selected machine type (402); Providing (806) current data (504) from the deployed machine (502) to said ROM (416); comparing (808) the current output (508) from the ROM (416) with the measured output (506) from the current data (504); determining (810) that an operational anomaly (514) exists if the difference between the current output and the measured output exceeds a selected anomaly threshold; and Identifying (812) a cause of the operational anomaly (514), providing (1004) the current data (504) to a plurality of fault models (602), each fault model (602) including a particular component fault; comparing (1006) fault model outputs (604) from each of the plurality of fault models (602) to the measured outputs (506) from the current data (504); selecting (1008) the fault model (602) having the fault model output (604) that most closely matches the measured output (506); and identifying (812), including indicating (1010) the component fault (608) associated with the selected fault model as a cause of the operational anomaly (514).

2. The method of claim 1 , wherein each fault model is a ROM based on a corresponding digital twin model of the machine that includes the particular component fault.

3. Creating (802) the ROM (416) for the digital twin model (406), creating (902) a plurality of reduced-size models (ROM) (410) for the digital twin model (406) of the selected machine type (402); Providing (904) historical data (404) from a plurality of machines of the selected machine type (402) to the plurality of ROMs (410) and to the digital twin model (406); comparing (906) an output (408) from the digital twin model (406) with an output (412) from each of the ROMs (410); and 2. The method (800) of claim 1, comprising selecting (908) the ROM (416) having the output (412) that most closely matches the output (408) of the digital twin model (406).

4. 4. The method of claim 3, wherein the plurality of reduced models comprises one or more of a gradient boosting model, a regression model, a recurrent neural network model, or a long short-term memory model.

5. The method (800) of any preceding claim, further comprising receiving (804) the current data (504) from a plurality of sensors (22, 24, 26) on the deployed machine (502).

6. A system (100) for identifying a cause of a machine operation anomaly, comprising: one or more processors (1110); and One or more memory devices (1150), which when executed by one or more processors (1110), cause the one or more processors (1110) to: Creating (802) a reduced-size model (ROM) (416) for a digital twin model (406) of a selected machine type (402); receiving (804) current data (504) from a plurality of sensors (22, 24, 26) on the deployed machine (502); providing (806) said current data (504) to said ROM (416); comparing (808) the current output (508) from the ROM (416) with the measured output (506) from the current data (504); determining (810) that an operational anomaly (514) exists if a difference between the current output (508) and the measured output (506) exceeds a selected anomaly threshold; and Identifying (812) a cause of the operational anomaly (514), providing (1004) the current data (504) to a plurality of fault models (602), each fault model (602) including a particular component fault; comparing (1006) fault model outputs (604) from each of the plurality of fault models (602) to the measured outputs (506) from the current data (504); selecting (1008) the fault model (602) having the fault model output (604) that most closely matches the measured output (506); and The system (100) comprises one or more memory devices (1150) having instructions stored thereon to cause the identifying (812), including indicating (1010), the component failure associated with the selected fault model as a cause of the operational anomaly (514).

7. The system of claim 6, wherein each fault model is a ROM based on a corresponding digital twin model of the machine that includes the particular component fault.

8. Creating (802) the ROM (416) for the digital twin model (406), creating (902) a plurality of reduced-size models (ROM) (410) for the digital twin model (406) of the selected machine type (402); Providing (904) historical data (404) from a plurality of machines of the selected machine type (402) to the plurality of ROMs (410) and to the digital twin model (406); comparing (906) an output (408) from the digital twin model (406) with an output (412) from each of the ROMs (410); and 7. The system of claim 6, further comprising selecting (908) the ROM (416) having the output (412) that most closely matches the output (408) of the digital twin model (406).

9. 10. The system of claim 8, wherein the plurality of reduced models comprises one or more of a gradient boosting model, a regression model, a recurrent neural network model, or a long short-term memory model.

10. 7. The system (100) of claim 6, further comprising instructions for identifying a machine subsystem associated with the operation anomaly (514), wherein the plurality of fault models (602) is based on a subsystem model associated with the identified machine subsystem.

Citation Information

Patent Citations

  • Construction machine

    JP2020166745A

  • System and methods for generating fault indications for an additive manufacturing process based on a probabilistic comparison of the outputs of multiple process models to measured sensor data

    US20200218628A1