Automotive physical layer (PHY) fault diagnosis

Abstract

To provide a method and a system for using an echo measurement value for detecting a cable failure in a cable of a vehicle communication network.SOLUTION: A method includes the steps of: obtaining at least one return echo signal corresponding to a signal communicated through a cable in a vehicle; applying, to the at least one return echo signal, a model trained by a computer for distinguishing a plurality of predetermined fault types in the cable from among at least two different predetermined fault types in the cable in order to identify the predetermined fault types; and initiating operation in response to the identified predetermined fault type, based on characteristics of the return echo signal learned from computer analysis of many previous echo signals.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] [CROSS-REFERENCE TO RELATED APPLICATIONS] This application claims the benefit of commonly owned U.S. Provisional Application No. 63 / 118,484, filed November 25, 2020, the disclosure of which is incorporated herein by reference.

[0002] The present disclosure relates generally to in-vehicle networks, and more particularly to methods and systems for diagnosing faults in cables in in-vehicle networks. [Background technology]

[0003] Modern vehicles, including autonomous vehicles in particular, operate generating extremely large amounts of data that is sensed or otherwise detected, analyzed, and transmitted through in-vehicle networks via cables. Failures in these cables can result in the in-vehicle network functioning at a reduced speed, functioning incorrectly, or not functioning at all. Failures in cables are typically due to aging and physical damage. If cables fail or function improperly, the vehicle may become unsafe or inoperable, and repairs to the vehicle and in-vehicle network are costly.

[0004] The foregoing description is provided as a general overview of the related art in the field and is not to be construed as an admission that any of the information it contains constitutes prior art against the present patent application. Summary of the Invention

[0005] The disclosed subject matter relates to methods and systems for using echo measurements to detect cable faults in cables of a vehicular communication network. The above-described systems and methods detect multiple cable fault types, including short-to-ground and short-to-power cable faults, that were previously undetectable due to system design constraints. Additionally, the methods and systems are easily integrated into existing vehicular communication networks, as no special hardware is required when using the methods or systems with existing vehicular communication networks.

[0006] An embodiment of the disclosed subject matter relates to a method for fault diagnosis in one or more cables of a vehicle communication network, the method comprising: obtaining at least one return echo signal corresponding to the signal communicated over the cable at the vehicle; applying a computer-trained model that distinguishes among a plurality of predetermined fault types in a cable to the at least one returning echo signal to identify a predetermined fault type from among at least two different predetermined fault types in the cable based on characteristics of the returning echo signal learned from computer analysis of multiple previous echo signals; initiating an action in response to the identified predetermined fault type; Includes.

[0007] In some embodiments, the method for fault diagnosis is such that the step of applying the computer-trained model comprises applying an artificial intelligence (AI) model that is pre-trained to distinguish the predetermined fault type in the cable from among the at least two predetermined fault types.

[0008] In some embodiments, the method for fault diagnosis is a method in which the AI ​​model includes multiple feature extraction layers configured to determine that the cable is faulty and identify the predetermined fault type. In some embodiments, the method for fault diagnosis is a method in which the acquiring at least one return echo signal includes receiving the at least one return echo signal by an Ethernet physical layer (PHY) device, the Ethernet PHY device applying echo cancellation in response to the receiving. In some embodiments, the method for fault diagnosis is a method in which the characteristics of the return signal include a shape of the return echo signal, and the applying the computer-trained model includes analyzing the shape of the return echo signal against at least two predetermined shapes of return echo signals, each of the predetermined shapes being associated with a predetermined fault type from among at least two different predetermined fault types.

[0009] In some embodiments, the method for fault diagnosis is such that applying the computer-trained model to the at least one returned echo signal comprises determining a probability that the one or more cables will exhibit the identified predetermined cable fault type selected from among at least two different predetermined cable fault types. In some embodiments, the method for fault diagnosis is such that the AI ​​model comprises one or more of a trained neural network and / or a machine learning model. In some embodiments, the method for fault diagnosis is such that the plurality of predetermined fault types in a cable are determined to be Short circuit (short circuit), Open circuit (open), open and grounded, Normal and grounded, One wire shorted to ground, One wire shorted to the power supply, One wire is open, Two wires shorted to ground, Two wires shorted to the power supply, and two wires open The method includes two or more of the following:

[0010] In some embodiments, the method for fault diagnosis is such that the at least one return echo signal includes a plurality of return echo signals, each return echo signal obtained from a corresponding cable, and applying the computer-trained model includes analyzing each return echo signal to identify a predetermined fault type in each cable from among at least two predetermined fault types. In some embodiments, the method for fault diagnosis is such that the initiating action includes one or more of logging the fault, issuing a warning, notifying a driver or other user, and switching to a backup cable.

[0011] An embodiment of the disclosed subject matter relates to a system for fault diagnosis in one or more cables of a vehicle communication network. The fault diagnosis system includes one or more processors and a program memory storing executable instructions that, when executed by the one or more processors, cause the system to: obtaining at least one return echo signal corresponding to the signal communicated over the cable at the vehicle; applying a computer-trained model that distinguishes between a plurality of predetermined fault types in a cable to the at least one returning echo signal to identify a predetermined fault type from among at least two different predetermined fault types in the cable based on characteristics of the returning echo signal learned from computer analysis of multiple previous echo signals; Initiating an action in response to the identified fault type. is an executable instruction that causes

[0012] In some embodiments, the system for fault diagnosis is a system where the executable instructions further cause the system to apply the computer-trained model to analyze the characteristics of the returning echo signal, including a shape of the returning echo signal, to determine the probability of the identified predetermined fault type from among at least two different predetermined fault types in the cable. In some embodiments, the system for fault diagnosis is a system where the computer-trained model is stored in the program memory and includes an artificial intelligence (AI) model that determines the probability of the identified predetermined fault type from among at least two different predetermined fault types in the cable. In some embodiments, the system for fault diagnosis is a system where the AI ​​model includes one or more of a trained neural network and / or a machine learning model.

[0013] An embodiment of the disclosed subject matter relates to an Ethernet physical layer (PHY) device for communication over a cable in an automotive or industrial network, the Ethernet PHY device comprising: a transmitter for transmitting an outbound Ethernet communication signal onto the cable; a receiver for receiving an inbound Ethernet communication signal from the cable, the inbound Ethernet communication signal including the return echo signal caused by the transmitted outbound Ethernet communication signal; an echo canceller for estimating the returning echo signal and for canceling the returning echo signal in the received inbound Ethernet communication signal; a storage medium in communication with the echo canceller for storing the estimated return echo signal, the stored estimated return echo signal being available for retrieval by a device external to the Ethernet PHY device; and In some embodiments, the Ethernet PHY device is such that the returning echo signal is of a predetermined size.

[0014] The present disclosure will be more fully understood from the following detailed description of the embodiments thereof when taken in conjunction with the drawings, in which: [Brief explanation of the drawings]

[0015] [Figure 1A-1] FIG. 1 is a block diagram of an exemplary in-vehicle network (IVN) in a vehicle that operates in accordance with an embodiment described herein.

[0016] [Figure 1A-2] FIG. 1A-1 is an inset of FIG.

[0017] [Figure 1B] 1 is a diagram of a system outside a vehicle and an in-vehicle network (IVN) operating in accordance with an embodiment described herein.

[0018] [Figure 2] 2 is a diagram of an exemplary physical layer (PHY) transceiver used to obtain return echo data measurements in accordance with one embodiment described herein, among other things.

[0019] [Figure 3] FIG. 2 is a flow diagram of an exemplary process according to one embodiment described herein.

[0020] [Figure 4]FIG. 1A is a block diagram that schematically illustrates a system architecture for determining cable faults and fault types in the in-vehicle network of FIG. 1A-1, according to one embodiment described herein.

[0021] [Figure 5A] FIG. 2 is a block diagram of an example neural network structure for performing cable fault detection and identification, according to one embodiment described herein.

[0022] [Figure 5B] FIG. 5B is a diagram of example layers of the neural network structure of FIG. 5A according to one embodiment described herein.

[0023] [Figure 6] FIG. 1 is a diagram of cable fault types and corresponding echoes.

[0024] [Figure 7] FIG. 1 is a diagram of training speed for a neural network model, according to one embodiment described herein.

[0025] [Figure 8A] 5 is a flow diagram of steps performed by the system of FIG. 4 according to one embodiment described herein. [Figure 8B] 5 is a flow diagram of steps performed by the system of FIG. 4 according to one embodiment described herein. DETAILED DESCRIPTION OF THE INVENTION

[0026] FIG. 1A-1 illustrates a vehicle 100, e.g., an automobile, having an in-vehicle network (IVN) 101, also known as a vehicle communication network; the terms vehicle communication network and in-vehicle network are used interchangeably herein. The vehicle 100 may be a standard, non-autonomous vehicle or an autonomous vehicle, classified according to one of the Society of Automotive Engineers (SAE) levels of automation: 0 (no automation), 1 (driver assistance), 2 (partial automation), 3 (conditional automation), 4 (high automation), or 5 (full automation). The vehicle communication network may include, for example, integrated circuit (IC) chips for performing various functions. There may also be multiple cables connecting the various components of the vehicle communication network. With this arrangement of the IVN, the disclosed subject matter operates at the link level and the network level.

[0027] Vehicle communication network 101 includes processor 110, e.g., a central processing unit (CPU), a graphics processing unit (GPU), or other suitable computer processing device; processor 110 represents an on-board computer, also known herein as a local computer. Processor 110 communicates with controller 112 and switch / gateway 114, directly or indirectly, through Ethernet link 118a via physical layer device (PHY) 116. Switch / gateway 114 communicates with each other, directly or indirectly, through Ethernet link 118a via PHY 116. Switch / gateway 114 communicates with PHY 116, directly or indirectly, through Ethernet link 118b.

[0028] Links 118a, 118b include cables, also known as Ethernet cables, for transmitting data and / or power. Links 118a, 118b are hereinafter also referred to as "cables" or "Ethernet cables." Cable types may be shielded or unshielded, single-core or multi-core, and multi-core cables may be, for example, twisted pair cables. Data transmission capabilities may not be the same for all cable types. For example, in one embodiment, Ethernet link 118a supports data rates of 10 / 25 Gbps, while Ethernet link 118b supports data rates of 2.5 / 5 / 10 Gbps.

[0029] Each PHY 116 links a link partner, such as between an electronics unit and a local computer or processor 110, typically via a suitable media access control (MAC) device. The electronics unit may include, for example, a camera 120, a radar / lidar / sonar 122, or other suitable sensors 124, such as temperature sensors, magnetic field sensors, etc. The PHYs 116 communicate with the local computer or processor 110, typically over a cable via a switch 114 and a controller 112.

[0030] Each link 118a, 118b has at least one end connected to switch 114. Typically, an echo signal acquired at one end of a link is sufficient to diagnose an Ethernet cable. Thus, in one embodiment, switch 114 reads the echo signal, for example, from memory 220 (FIG. 2) of PHY 116 to which switch 114 is locally connected, and relays the echo signal to processor 110 in the form of an echo measurement representing the returned echo signal. A trained computer model 400, such as an artificial intelligence (AI) model, downloaded by and executed on processor 110 receives the echo measurement as an input.

[0031] In one embodiment, PHY 116 receives echo signals, e.g., return echo signals, from cables 118a, 118b, estimates corresponding return echo signals, and transmits the estimated return echo signals back over Ethernet cables 118a, 118b, e.g., as echo measurements, via switch 114. An exemplary PHY 116 for receiving return echo signals and estimating echo cancellation signals as echo measurements is shown, for example, in FIG. 2 and described in more detail below.

[0032] 1A-2, for processor 110, processor 110 downloads model 400, e.g., a rule-based (non-AI) model or an AI model. Processor 110 is also associated with at least one transceiver 132 to receive returning echo signals and / or data associated with the returning echo signals, input the signals and / or data into model 400, and transmit outputs generated by the model as metrics to other system computers, processors, components, and the like, as described below.

[0033] 1B shows a system 140 external to the vehicle 100 and the IVN 101 for detecting cable faults in the Ethernet cables 118a, 118b connected to the PHY 116 of the vehicle 100. The system 140 includes a computer 142 that has downloaded an artificial intelligence (AI) model 400, including non-AI models (such as rule-based models) or other downloadable models, to receive echo responses, e.g., echo cancellation signals and / or data associated with the echo cancellation signals, from the PHY 116 to determine whether there is a fault in the Ethernet cables 118a, 118b and, if there is a fault, the type of cable fault.

[0034] 1A-1, the configuration of IVN 101 and the composition and / or deployment of network components, such as processor 110, controller 112, switch 114, PHY 116, and various other system elements, such as camera 120, radar / lidar / sonar 122, or other suitable sensors 124, such as temperature and magnetic field sensors, are exemplary configurations shown for clarity only. In alternative embodiments, any other suitable configurations may be used.

[0035] Various elements of the IVN 101, such as the processor 110, controller 112, switch 114, PHY 116, and other system elements, such as the camera 120, radar / lidar / sonar 122, or other suitable sensors 124, such as temperature and magnetic field sensors, may be implemented using dedicated hardware or firmware, such as in one or more application-specific integrated circuits (ASICs) and / or one or more field-programmable gate arrays (FPGAs), using, for example, hardwired or programmable logic. Additionally or alternatively, the functionality of some of the above-described components of the IVN may be implemented in software and / or using a combination of hardware and software elements. Elements not essential to an understanding of the disclosed technology have been omitted from the figures for clarity.

[0036] In some embodiments, the processor of the internal computer 110 and / or the processor of the external computer 142 (which may be one or more computers) include one or more programmable processors that are programmed in software to perform the functions described herein. The software may be downloaded to any of these processors in electronic form, for example, over a network, or alternatively or additionally may be provided and / or stored on a non-transitory tangible medium, such as magnetic, optical, or electronic memory.

[0037] 2 illustrates an exemplary PHY 116, represented as circuit 200, according to one embodiment described herein. The PHY 116 of FIG. 2 performs echo cancellation for internal use (to improve receive performance).

[0038] In this example, PHY 116 includes a data modulator 202 that outputs an Ethernet signal containing data symbols. A transmitter (TX) 204 transmits the Ethernet signal over cable 118a / 118b via a hybrid junction 206 (a hybrid junction for splitting and combining signals to / from different channels). The transmitted data is directed to a peer PHY 116 (not shown) at the other end of the cable.

[0039] Upon reception, the Ethernet signal from the peer PHY 116 is received from the cable 118a, 118b via the hybrid junction 206 and provided to the filter 208. The filter 208 filters the received signal to remove undesired frequency components, for example, using a low-pass or band-pass filter. An analog-to-digital converter (ADC) 210 digitizes (samples) the signal. For example, the ADC 210 may digitize the received signal in batches of 16 phases with 48 samples per phase for a total of 768 samples.

[0040] The digitized signal is filtered by a feed-forward equalizer (FFE) 212. A decision circuit 214 (e.g., a slicing circuit) decodes bits from the equalized signal. The decoded bits are passed through a physical coding sublayer (PCS, not shown) and then provided as an output to, for example, a medium access control (MAC) device (not shown). For example, the MAC device may be in the switch 114. In this example, the signal is also equalized by a decision feedback equalizer (DFE) 216, whose input is taken from the output of the decision circuit 214.

[0041] PHY 116 further includes an echo canceller (EC) 218, which cancels echoes of the transmitted Ethernet signal that may leak into and interfere with the received Ethernet signal. Typically, EC 218 estimates the time waveform (shape in the time domain) of the echo signal based on the transmitted signal and cancels the echo signal from the received signal. Generally speaking, EC 218 generates a replica of the transmitted Ethernet signal with a magnitude and phase that matches the estimated echo signal and then subtracts this replica from the received Ethernet signal. The cancellation operation is indicated by adder 219 in the figure.

[0042] In one embodiment, PHY 116 further comprises a storage medium, e.g., memory 220. EC 218 stores the estimated echo signals as vectors of digital values ​​in memory 220. Memory 220 is such that the echo signals are accessible for readout by any suitable external processor or system (e.g., by processor 110 or by computer 142) for use in detecting and classifying possible faults in cables 118a, 118b. In an exemplary embodiment, although not necessarily, the estimated echo signals include 768 digital values ​​sampled with a sampling rate of 66.66 MHz using 16 phases, resulting in a resolution of 0.9376 nS per sample.

[0043] As can be seen in the figure, FFE 212, DFE 216, and EC 218 are controlled by suitable adaptation algorithms running on a suitable processor in the PHY device (not shown). In this example, the input to these algorithms is an error signal (shown in the figure as err) indicating the difference between the received signal and the respective bit decision.

[0044] A local or internal computer, i.e., either processor 110 or computer 142 of external cable fault detection system 140 (also referred to herein as “external computer 142”), performs an exemplary process of cable fault detection and identification of the detected cable fault type based on the echo signals reported by PHY 116. The exemplary process is generally illustrated in FIG. 3, which now receives attention.

[0045] In FIG. 3 , the process begins at block 300. At this starting point, in one embodiment, PHY 116 or IVN 101 is in a diagnostic mode, ready to perform cable diagnostics. The process proceeds to block 302, where a model, e.g., a computer-trained model, is obtained for detecting cable faults and identifying the cable fault type for a detected cable fault in the cable. The model 400 serves to distinguish between multiple predetermined fault types in the cable (cable fault types) and identifies the predetermined fault types based on characteristics, e.g., shape, of the returning echo signals (and / or data associated with the returning echo signals, known as “signal data”), where the predetermined fault types are learned from computer analysis of multiple previous echo signals, e.g., from among at least two different predetermined fault types in the cable.

[0046] For example, model 400 may be a trained neural network model. This is typically accomplished by local computer 100 or external computer 142 downloading the trained neural network model from a location along the network and / or from the cloud. The process proceeds to block 304, where signal data is obtained from PHY 116 corresponding to return echo signals received at PHY 116 from one or more cables 118a, 118b. For example, PHY 116 also uses this signal data, e.g., by estimating, to form echo cancellation signals used internally by PHY 116.

[0047] Proceeding to block 306, the acquired signal data is applied to the trained model. The trained model analyzes the signal data to determine whether the corresponding cable has a fault and, if a fault is detected, identifies the fault, including classifying the type of cable fault (cable fault type) among known types of cable faults. In an optional process at block 306, the trained model may also determine a probability of the type of cable fault from among at least two predetermined fault types or from among a set of at least two predetermined fault types.

[0048] For a cable analyzed by the model, the detected fault type, the predetermined fault identified for the detected fault type, and optionally, the probability of the detected fault type from among at least two predetermined fault types are metrics output by the model in block 308.

[0049] The process proceeds to block 310, where, based on and in response to the output metrics, one or more actions are initiated and taken based on the identified fault type. For example, the initiated actions may be one or more of logging the fault, issuing a warning, notifying a driver or other user associated with vehicle 100, and switching to a backup cable.

[0050] The process then proceeds to block 312, where the signal data described above is optionally uploaded based on the detected cable fault and the identity of the detected cable fault to retrain or enhance the training of the trained model. The process ends at block 314, and may be repeated for as long as desired.

[0051] FIG. 4 is an architecture block diagram of a trained model 400, such as a neural network model (e.g., an artificial intelligence (AI) model or a machine learning (ML) model), downloaded to a vehicle or local computer, i.e., typically one of the processor 110 or external computer 142, according to an embodiment of the present disclosure. The trained model 400 (e.g., a computer-trained model) receives input, which is return echo signal data 410, such as data associated with return echo signals from cables 118a, 118b, through which the PHY 116 receives the return echo signals via the Ethernet cables 118a, 118b. The echo cancellation signal is expressed as a value or data representing the signal shape for the Ethernet cable 118b of the IVN 101. The echo cancellation signal and corresponding signal data are typically generated at predetermined time intervals, typically when the vehicle 100 is started. The above example is for signal shape in the time domain. Alternatively, the described techniques can also be adapted to detect signal shape anomalies in the frequency domain.

[0052] In alternative embodiments, the signal data described above is received, for example, by on-board 110 of vehicle 100 or by external computer 142 in response to periodic queries, or is received automatically, for example, periodically.

[0053] In one embodiment, the trained model 400 (e.g., a computer-trained model) may be downloaded from a remote location over an external network such as the Internet, and portions of the trained model 400 may be stored in a cloud 420. There is also an associated module for runtime recording 422. The trained model 400, running on the processor 110 or computer 142, outputs a set of metrics in response to input information, i.e., signal data representing return echo signals for the Ethernet cables 118a, 118b received by the PHY 116.

[0054] The set of metrics is an indication, for example, including the detection of a cable fault 430, an identification of the cable fault type of the detected cable fault 432, and a probability of the cable fault type of the detected cable fault from among two or more types (e.g., a set of cable fault types) of cable fault 434. The results of these metrics are analyzed, for example, by the processor 110 or the external computer 142 to initiate an action to be taken in the vehicle based on the identified cable fault type 436. The initiated actions include, for example, logging the cable fault (and type of cable fault), issuing a warning, notifying the vehicle driver or other party associated with the vehicle, and switching to a backup cable.

[0055] The trained model may be a rule-based model. The rule-based model operates, for example, by analyzing the signal's major peaks. As shown in Figure 6, in box 602, a positive major peak indicates an open fault, a negative major peak indicates a short fault, while if there is no significant major peak, the cable is healthy.

[0056] 5A, which is a diagram of a network architecture for an exemplary trained model 400, which is a neural network model, according to one embodiment of the present disclosure. The neural network model 400 is a seven-layer neural network (e.g., a seven-layer convolved feature extractor 502) with a bidirectional network structure (an output layer with a customized loss function, described in detail below) for fault detection 504a and fault type identification, including classification 504b. The mean squared error (MSE) combined loss functions for the detection 504a and the cross-entropy for the classification 504b are mathematically expressed as follows:

[0057] Input, i.e., signal data, is fed into layer 1 of the neural network model 400, and output is received from the detector 504a and the classifier 504b. The number of layers, the type of layers, including hidden layers, and / or the size of the layers depend, for example, on the number of transformations to be performed and the complexity of the transformations.

[0058] In one embodiment, each layer (Conv1D to Conv7D) in the feature extraction unit 502 is, for example, a fully convolutional layer 510, for example, indicated by layer 1 (Conv1D) in FIG. 5B. For example, as shown in FIG. 5A, the size of the first layer (CONV1D_1) is 394×64, the size of the second layer (CONV1D_2) is 192×16, the size of the third layer (CONV1D_3) is 96×6, the size of the fourth layer (CONV1D_4) is 48×16, the size of the fifth layer (CONV1D_5) is 24×8, the size of the sixth layer (CONV1D_6) is 12×8, and the size of the seventh layer (CONV1D_7) is 6×64. The neural network model 400 includes node activation functions, for example, sigmoid or rectified linear units (ReLU), to perform the desired transformation.

[0059] This particular neural network architecture is provided herein merely by way of example, and the principles of the present disclosure may alternatively be implemented in other types of neural networks having a greater or fewer number of layers and a greater or fewer number of inputs, outputs, and nodes within each layer. The depicted neural network model 400 is applicable to Ethernet cable fault determination and Ethernet cable fault identification, including classifying a determined cable fault and determining a probability of a fault type for the determined fault from among two or more (e.g., a set) predetermined cable fault types. Typically, processor 110, or a processor of external computer 142, executes multiple models of this type to process cable fault information for multiple Ethernet cables associated with communication links and other components.

[0060] The neural network model 400, downloaded to the vehicle processor 110 or external computer 142, for example from the cloud or other centralized processor serving multiple vehicles, is trained according to an objective function for the classifier 504b. Because training typically consumes many computer resources and a large amount of power, training may preferably be performed in the cloud. In the case of discrete outputs, for example, the objective function for the classifier 504b is based on cross-entropy (CE), expressed as follows:

number

[0061] For example, if the fault type is "open", y1 to y4 are classified into [p o ,p s ,p n&g ,p o&g ] may be [0.9,0,0,0.1], which corresponds to

[0062] For continuous outputs such as fault determination and fault identification, which include fault classification and determining the probability of a fault type from among two or more predetermined cable fault types, a loss function is determined for the Ethernet cable. The loss function is based on mean square error (MSE) for the detector 504a and cross entropy (CE) for the classifier 504b. The final or customized loss function (e.g., total loss (TL)) is expressed as a weighted sum with weights w1 and w2: TL = w1 × MSE + w2 × CE.

[0063] Any suitable algorithm may be used to train the neural network model 400 based on this customized loss function. For example, the algorithm may be an Adam optimization algorithm, and random initial weights may be used. The learning rate for the customized loss function may be controlled by adjusting the learning rate during training of the neural network model.

[0064] The trained neural network model 400 is trained, for example, on a remote server along a network or in the cloud. The neural network model is trained by receiving inputs, e.g., signal shapes indicative of detected cable faults, normal cables, or various cable fault types, from a large number of vehicles (IVNs) different from the vehicle 100, which are uploaded to the neural network model 400. Training the neural network model 400 using inputs from many vehicles is, for example, typically centralized so that the trained model benefits from a large amount of input. Thus, the trained model is useful for a given vehicle (e.g., typically an individual vehicle), such as the vehicle 100.

[0065] For example, model 400 is trained based on known or predetermined signal shapes for various types of cable faults obtained from multiple IVNs in multiple vehicles. The above-mentioned predetermined echo signal shapes for normal cables and various cable fault types are shown, for example, in FIG. 6. For an Ethernet cable length of approximately 6 meters, 1) box 602 shows the return echo signal shape for normal (fault-free cable), open fault, or short fault in the Ethernet cable. 2) box 604 shows the return echo signal shape for open fault, or open fault and ground fault in the Ethernet cable. 3) box 606 shows the return echo signal shape for normal (fault-free cable) and normal and ground fault in the Ethernet cable. The model 400 analyzes the shape of the returning echo signal from the signal data against the shape of the signal for each cable fault type to determine whether there is a cable fault, and if there is a cable fault, identifies the type of cable fault, or if not, classifies it, or for a detected cable fault, determines the probability of the type of cable fault from at least two different types of cable fault based on the signal shape.

[0066] For example, trained neural network model 400 is trained based on signal data collected over a period of time from a component, i.e., PHY 116 of IVN 101, in a set of vehicles or multiple vehicles different from vehicle 100 (e.g., a given vehicle), which neural network model 400 downloads to a local computer (e.g., processor 110) or external computer 142 of the vehicle. For example, the model may determine whether there is a predetermined fault in the cable, distinguish among multiple predetermined fault types in the cable in the vehicular network, and for the determined predetermined fault, identify the predetermined fault type in the cable based on characteristics of the return echo signals in the cable, e.g., the predetermined fault type is learned from computer analysis of multiple previous echo signals from among at least two different predetermined fault types (or from a set of at least two different predetermined fault types), and / or, optionally, determine a probability for the identified predetermined fault type in the cable from among at least two different predetermined fault types (or from a set of at least two different predetermined fault types). For example, predetermined fault types in the cable include short circuit (short), open circuit (open), open and ground, normal and ground, one wire shorted to ground, one wire shorted to power, one wire open, two wires shorted to ground, two wires shorted to power, and two wires open.

[0067] The collected signal data is, for example, periodically uploaded to a centralized processor to train the trained neural network model 300, where the trained neural network model 400 resides, and may be located along a network remote from the vehicle 100, such as the local computer 110 of the vehicle 100, an external computer 142, or the cloud.

[0068] The neural network model 400 is trained to generate a set of metrics based on the collected signal data, for example, as shown and described for Fig. 4. The set of metrics indicates a detected fault in the cable 430, a cable fault type (identified type) for the detected cable fault 432, and a probability of the fault type for the detected cable fault from among two or more predetermined fault types 434, along with an action to be initiated in response to the detected and identified cable fault type 436.

[0069] The trained neural network model is deployed, for example, as a software package (which may be downloadable, for example, when the vehicle is parked and connected to an external wired or wireless network connection) for installation on a computer, e.g., processor 110, of the local vehicle 100, or on an external computer 142 (e.g., processor). The trained neural network model is typically associated with an application programming interface (API) for updating the trained neural network model 400.

[0070] 7 is a diagram of a training setup for a neural network model 400. The training rate is based on an optimization algorithm, such as a loss function, combined mean squared error and cross entropy, root mean square propagation (RMSProp). The diagram shows an adjustable learning rate for the model 400, such as the learning rate (LR) (y-axis), contrasted with the reference time (x-axis), which is a time-based value. Initially, training is performed at a rapid rate and slows down over time.

[0071] Attention is now directed to Figures 8A and 8B, which illustrate flow diagrams detailing computer-implemented steps in accordance with embodiments of the disclosed subject matter. The above-described processes, including sub-processes, may be performed automatically and / or in real time, for example.

[0072] The process begins at start block 802 where a local computer 110 of the IVN of the vehicle 100, i.e., a local computer associated with a vehicle communication network, such as a given vehicle or external computer 142, is activated. The process proceeds to block 804 where a model, e.g., a computer-trained model, is obtained.

[0073] The computer-trained model distinguishes among multiple predetermined fault types in cables in the vehicle communication network and identifies a predetermined fault type in the cable from among at least two different predetermined fault types (or from a set of at least two different predetermined fault types) based on characteristics of return echo signals in the cable learned from computer analysis of multiple previous echo signals, and / or optionally determines a probability for the identified predetermined fault type in the cable from among at least two different predetermined fault types (or from a set of at least two different predetermined fault types). For example, predetermined fault types in the cable include short circuit (short), open circuit (open), open and ground, normal and ground, one wire shorted to ground, one wire shorted to power, one wire open, two wires shorted to ground, two wires shorted to power, and two wires open.

[0074] The characteristics of the return echo signal include, for example, a shape of the return echo signal. Applying the computer-trained model may include, for example, analyzing the shape of the return echo signal against at least two predetermined shapes of the return echo signal, each of the predetermined echo signal shapes being associated with a predetermined cable fault type from among at least two different predetermined fault types.

[0075] Additionally, the at least one acquired return echo signal may be, for example, a plurality of return echo signals, each of the return echo signals acquired from a corresponding cable of the IVN 101. Applying the computer-trained model includes analyzing each return echo signal to identify a predetermined cable fault type in each cable from among at least two predetermined fault types.

[0076] The computer-trained model may be, for example, an artificial intelligence (AI) model that is pre-trained to distinguish a predetermined fault type in a cable from among at least two predetermined fault types (or from a set of at least two different predetermined fault types). The AI ​​model may include, for example, multiple feature extraction layers configured to determine that the cable has a fault and to identify the fault type such that a detected fault is identified as one of the predetermined fault types. The AI ​​model may include, for example, one or more trained neural networks and / or machine learning models.

[0077] The process proceeds to block 806, where the computer-trained model is downloaded to a local computer associated with the vehicle communication network, e.g., local computer 110 or external computer 142 associated with the vehicle communication network of the given vehicle. The computer-trained model is then deployed in block 808 by a local computer associated with the vehicle communication network, e.g., local computer 110 or external computer 142 of the given vehicle. The process proceeds to block 810, where at least one return echo signal (and / or data associated with the at least one return echo signal) corresponding to a signal communicated over a cable at the given vehicle is obtained.

[0078] The process proceeds to block 812, where at least one return echo signal (and / or return echo signal data) from a cable for a given vehicle is input, or if not input, applied, to a computer-trained model. Next, at block 814, output from the model is received, where the output is based on operations performed by the computer-trained model corresponding to the input at least one return echo signal (and / or return echo signal data) of the computer-trained model for the given vehicle.

[0079] The process proceeds to block 816 where a response is made to the received output by initiating the following actions depending on the identified predetermined fault type, the actions including one or more of logging the identified predetermined cable fault, issuing a warning, notifying the driver or other user associated with the given vehicle, and switching to a backup cable.

[0080] Proceeding to block 818, return echo signals having characteristics of known predetermined cable fault types are provided (e.g., periodically) to update and / or add to the at least two different fault types (e.g., a set of at least two different predetermined fault types). Next, at block 820, characteristics of return echo signals from the given vehicle and / or one or more of the other vehicles are provided (e.g., periodically) to the computer-trained model to further train the model.

[0081] The process proceeds to block 822 and ends. The process may be repeated for as long as there is demand.

[0082] The processor 110 (of the local computer in the vehicle) or the external computer 142 may include a general-purpose computer programmed with software, including, for example, a trained neural network model, to perform the functions described herein. The software may be downloaded to the computer in electronic form, for example, over a network, or alternatively or additionally, may be provided and / or stored on a non-transitory tangible medium, such as magnetic, optical, or electronic memory.

[0083] Although the embodiments described herein primarily deal with automotive applications (e.g., cars or lorries), the methods and systems described herein may also be used in other types of vehicular networks, such as in aircraft and boats.

[0084] It should be noted that the above-described embodiments are cited as examples, and that the present invention is not limited to what has been particularly shown and described above. Rather, the scope of the present invention includes both combinations and subcombinations of the various features described above, as well as variations and modifications thereof that would occur to one skilled in the art upon reading the foregoing description and that are not disclosed in the prior art. Documents incorporated by reference into this patent application are to be considered integral parts of this application. However, to the extent that any term is defined in these incorporated documents in a manner that contradicts a definition expressly or impliedly made herein, only the definition in this specification shall be considered.

Claims

1. 1. A method for fault diagnosis in one or more cables of a vehicle communication network, said method comprising: obtaining at least one return echo signal corresponding to the signal communicated over the cable at the vehicle; applying a computer-trained model that distinguishes among a plurality of predetermined fault types in a cable to the at least one returning echo signal to identify a predetermined fault type from among at least two different predetermined fault types in the cable based on characteristics of the returning echo signal learned from computer analysis of multiple previous echo signals; initiating an action in response to the identified predetermined fault type; Including, the characteristics of the return signal include a shape of the return echo signal; applying the computer-trained model includes analyzing the shape of the returned echo signal against at least two predetermined shapes of the returned echo signal, each of the predetermined shapes being associated with a predetermined fault type from among at least two different predetermined fault types. method.

2. 2. The method for fault diagnosis of claim 1, wherein applying the computer-trained model comprises applying an artificial intelligence (AI) model that is pre-trained to distinguish the predetermined fault type in the cable from among the at least two predetermined fault types.

3. 3. The method for fault diagnosis of claim 2, wherein the AI ​​model includes a plurality of feature extraction layers configured to determine that the cable has a fault and identify the predetermined fault type.

4. 4. The method for fault diagnosis of claim 1, wherein the step of obtaining the at least one return echo signal comprises receiving the at least one return echo signal by an Ethernet physical layer (PHY) device, the Ethernet PHY device applying echo cancellation to the at least one return echo signal in response to the receiving step.

5. 5. A method for fault diagnosis according to claim 1, wherein applying the computer-trained model to the at least one returned echo signal comprises determining a probability that the one or more cables will exhibit the identified predetermined cable fault type selected from among at least two different predetermined cable fault types.

6. 4. The method for fault diagnosis according to claim 2 or 3, wherein the AI ​​model comprises one or more of a trained neural network and / or a machine learning model.

7. The plurality of predetermined fault types in the cable include: One wire shorted to ground, One wire shorted to the power supply, One wire is open, Two wires shorted to ground, Two wires shorted to the power supply, and Two wires open 7. The method for fault diagnosis according to claim 1, comprising two or more of:

8. 8. The method for fault diagnosis of claim 1, wherein the at least one return echo signal comprises a plurality of return echo signals, each return echo signal being obtained from a corresponding cable, and wherein applying the computer-trained model comprises analyzing each return echo signal to identify a predetermined fault type in each cable from among at least two predetermined fault types.

9. 9. A method for fault diagnosis as claimed in any one of claims 1 to 8, wherein the initiating action comprises one or more of the following steps: logging the fault, issuing a warning, notifying a driver or other user, and switching to a backup cable.

10. 1. A system for diagnosing faults in one or more cables of a vehicle communication network, the system comprising: one or more processors; When executed by the one or more processors, the system: obtaining at least one return echo signal corresponding to the signal communicated over the cable at the vehicle; applying a computer-trained model that distinguishes between a plurality of predetermined fault types in a cable to the at least one returning echo signal to identify a predetermined fault type from among at least two different predetermined fault types in the cable based on characteristics of the returning echo signal learned from computer analysis of multiple previous echo signals; applying the computer-trained model to analyze the characteristics of the return echo signal, including its shape, to determine a probability of the identified predetermined fault type from among at least two different predetermined fault types in the cable; Initiating an action in response to the identified fault type. a program memory storing executable instructions for causing the A system comprising:

11. 11. The system for diagnosing faults of claim 10, wherein the computer-trained model is stored in the program memory and includes an artificial intelligence (AI) model that determines the probability of the identified predetermined fault type from among at least two different predetermined fault types in the cable.

12. 12. The system for diagnosing faults of claim 11, wherein the AI ​​model comprises one or more of a trained neural network and / or a machine learning model.

13. 1. An Ethernet physical layer (PHY) device for communication over a cable in an automotive or industrial network, the Ethernet PHY device comprising: a transmitter for transmitting outbound Ethernet communication signals onto the cable; a receiver for receiving an inbound Ethernet communication signal from the cable, the inbound Ethernet communication signal including a return echo signal caused by the transmitted outbound Ethernet communication signal; an echo canceller for estimating the returning echo signal and for canceling the returning echo signal in the received inbound Ethernet communication signal; a storage medium in communication with the echo canceller for storing the estimated returned echo signal, the stored estimated returned echo signal being available for retrieval by a device external to the Ethernet PHY device; and a processor configured to determine a probability of a predetermined fault type identified from among at least two different predetermined fault types in the cable based on the shape of the return echo signal; 1. An Ethernet PHY device comprising:

14. 14. The Ethernet PHY device of claim 13, wherein the returning echo signal is of a predetermined size.

Images