Vehicle fault diagnosis method and device, equipment and storage medium

Abstract

The invention discloses a vehicle fault diagnosis method and device, equipment and a storage medium, and relates to the technical field of vehicle detection. The vehicle fault diagnosis method comprises the steps of obtaining a fault diagnosis model; wherein the fault diagnosis model is obtained based on pre-constructed double-auto-encoder training; acquiring a current diagnosis parameter of the target vehicle; wherein the current diagnosis parameters are used for representing key parameters of the target vehicle thermal management system; determining a fault score of the target vehicle according to the current diagnosis parameter, the fault diagnosis model and a pre-constructed fault score calculation model; wherein the fault score is used for diagnosing the thermal management system of the target vehicle. According to the vehicle fault diagnosis method provided by the invention, the fault diagnosis model is obtained based on double auto-encoder training, and the accuracy of the fault diagnosis model is improved, so that the accuracy of target vehicle fault score calculation is improved according to the fault diagnosis model, the diagnosis parameters and the fault score calculation model; and thus, the fault diagnosis accuracy of the vehicle thermal management system is improved.

Description

Technical Field

[0001] The present application relates to the field of vehicle detection technology, and in particular to a vehicle fault diagnosis method, apparatus, device and storage medium. Background Art

[0002] With the rapid development of the automotive industry and the increasing level of intelligence, modern vehicle systems are becoming increasingly complex, integrating a large number of sensors and actuators to monitor and control various vehicle operating conditions. During operation, these sensors and actuators generate a large amount of time series data, containing rich vehicle status information such as engine speed, torque, oil temperature, and water temperature. This data is crucial for timely detection and diagnosis of vehicle faults.

[0003] However, processing and analyzing this multivariate time series data and applying it to fault diagnosis in automotive thermal management systems still faces numerous challenges. Existing automotive thermal management system monitoring and fault diagnosis algorithms mostly rely on traditional fault diagnosis methods, such as rule-based, statistical, and machine learning algorithms. However, these methods often suffer from high false positive rates when processing high-dimensional, nonlinear, and noisy multivariate time series data, resulting in low fault diagnosis accuracy. Summary of the Invention

[0004] Purpose of the invention: The embodiments of the present application provide a vehicle fault diagnosis method, apparatus, device and storage medium to improve the accuracy of vehicle thermal management system fault diagnosis.

[0005] Technical Solution: A vehicle fault diagnosis method according to an embodiment of the present application is applied to diagnose faults in a vehicle thermal management system; the method comprises:

[0006] Obtaining a fault diagnosis model; wherein the fault diagnosis model is obtained based on pre-built dual autoencoder training;

[0007] Acquiring current diagnostic parameters of the target vehicle; wherein the current diagnostic parameters are used to characterize key parameters of the thermal management system of the target vehicle;

[0008] A fault score of the target vehicle is determined based on the current diagnostic parameters, the fault diagnosis model, and a pre-built fault score calculation model; wherein the fault score is used to diagnose a thermal management system of the target vehicle.

[0009] In some embodiments, the dual autoencoder includes a first autoencoder and a second autoencoder; the first autoencoder includes a first encoder and a first decoder, and the second autoencoder includes the first encoder and a second decoder; the first decoder and the second decoder have the same structure.

[0010] In some embodiments, determining the fault score of the target vehicle based on the current diagnostic parameters, the fault diagnosis model, and a pre-built fault score calculation model includes:

[0011] Determine a first target output and a second target output according to the fault diagnosis model; wherein the first target output is used to represent the mean square error of the first decoder, and the second target output is used to represent the mean square error of the second decoder;

[0012] A fault score of the target vehicle is determined based on the current diagnostic parameters, the first target output, the second target output, and the fault score calculation model.

[0013] In some embodiments, determining the fault score of the target vehicle according to the current diagnostic parameters, the first target output, the second target output, and the fault score calculation model includes:

[0014] determining a first item according to the current diagnostic parameter, the first target output, and a first preset weight;

[0015] determining a second item according to the current diagnostic parameter, the second target output, and a second preset weight; wherein the first item and the second item are used to represent two items of the fault score calculation model;

[0016] The fault score is determined according to the sum of the first term and the second term.

[0017] In some embodiments, a method for obtaining a fault diagnosis model includes:

[0018] Acquiring historical diagnostic parameters of a test vehicle; wherein the historical diagnostic parameters are used to characterize key parameters of a thermal management system of the test vehicle;

[0019] The fault diagnosis model is determined according to the historical diagnosis parameters, the first autoencoder, and the second autoencoder.

[0020] In some embodiments, determining the fault diagnosis model based on the historical diagnostic parameters, the first autoencoder, and the second autoencoder includes:

[0021] Pre-training the first autoencoder and the second autoencoder respectively according to the historical diagnostic parameters;

[0022] Performing adversarial training on the first autoencoder and the second autoencoder respectively according to the historical diagnostic parameters;

[0023] The pre-training and adversarial training operations are repeatedly performed until the fault diagnosis model is determined when a preset iterative training condition is met.

[0024] In some embodiments, the network structure of the first encoder includes a convolutional layer, a first long short-term memory network layer, and a second long short-term memory network layer; the network structure of the first decoder includes a third long short-term memory network layer and a deconvolution layer.

[0025] Accordingly, an embodiment of the present application further provides a vehicle fault diagnosis device for diagnosing faults in a vehicle thermal management system; the device comprises:

[0026] A first acquisition module is used to acquire a fault diagnosis model; wherein the fault diagnosis model is obtained based on pre-built dual autoencoder training;

[0027] A second acquisition module is used to acquire current diagnostic parameters of the target vehicle; wherein the current diagnostic parameters are used to characterize key parameters of the thermal management system of the target vehicle;

[0028] A determination module is configured to determine a fault score of the target vehicle based on the current diagnostic parameters, the fault diagnosis model, and a pre-built fault score calculation model; wherein the fault score is used to diagnose a thermal management system of the target vehicle.

[0029] Correspondingly, an embodiment of the present application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the vehicle fault diagnosis method as described above when executing the computer program.

[0030] Accordingly, an embodiment of the present application further provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the vehicle fault diagnosis method as described above is implemented.

[0031] Beneficial Effects: Compared to the prior art, the vehicle fault diagnosis method, apparatus, device, and storage medium of the embodiments of the present application include: obtaining a fault diagnosis model; wherein the fault diagnosis model is obtained based on pre-built dual autoencoder training; obtaining current diagnostic parameters of the target vehicle; wherein the current diagnostic parameters are used to characterize key parameters of the target vehicle's thermal management system; determining the target vehicle's fault score based on the current diagnostic parameters, the fault diagnosis model, and a pre-built fault score calculation model; wherein the fault score is used to diagnose the target vehicle's thermal management system. The vehicle fault diagnosis method provided in the present application improves the accuracy of the fault diagnosis model by obtaining the fault diagnosis model based on dual autoencoder training, thereby improving the accuracy of the target vehicle's fault score calculation based on the fault diagnosis model, diagnostic parameters, and fault score calculation model, and further improving the accuracy of vehicle thermal management system fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0033] Figure 1 is a flow chart of a vehicle fault diagnosis method provided in an embodiment of the present application;

[0034] Figure 2 Schematic diagram of the principle structure of the fault diagnosis model provided in the embodiment of the present application;

[0035] Figure 3 Schematic diagram of the principle structure of the encoder provided in the embodiment of the present application;

[0036] Figure 4 Schematic diagram of the principle structure of the decoder provided in the embodiment of the present application;

[0037] Figure 5 This is a schematic diagram of the overall process of a vehicle fault diagnosis method provided in an embodiment of the present application;

[0038] Figure 6 Schematic diagram of the distribution difference of fault scores of normal vehicles and faulty vehicles provided in the embodiment of the present application;

[0039] Figure 7 is a fault score curve of a normal vehicle provided in an embodiment of the present application;

[0040] Figure 8 is a failure score curve of a faulty vehicle provided in an embodiment of the present application;

[0041] Figure 9 This is a principle structural block diagram of a vehicle fault diagnosis device provided in an embodiment of the present application;

[0042] Figure 10 It is a structural diagram of an electronic device provided in an embodiment of the present application.

[0043] Reference numerals:

[0044] 101 - first acquisition module; 102 - second acquisition module; 103 - determination module; 100 - vehicle fault diagnosis device. DETAILED DESCRIPTION

[0045] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the embodiments described are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0046] It should be understood that although the terms first, second, etc. may be used herein to describe various components, these components should not be limited by these terms. These terms are used to distinguish one component from another. Thus, the first component discussed below could be referred to as the second component without departing from the teachings of the present invention. As used herein, the term "and / or" includes any one and all combinations of one or more of the associated listed items.

[0047] Those skilled in the art will appreciate that the drawings are merely schematic diagrams of exemplary embodiments and may not be to scale. The modules or processes in the drawings are not necessarily required to implement the present application and therefore cannot be used to limit the scope of protection of the present application.

[0048] The applicant's research has revealed that with the rapid development of the automotive industry and the increasing level of intelligence, modern automotive systems are becoming increasingly complex, integrating a large number of sensors and actuators to monitor and control the various operating states of the vehicle. These sensors and actuators generate a large amount of time-series data during operation, containing a wealth of vehicle status information, such as engine speed, torque, oil temperature, and water temperature. This data is crucial for the timely detection and diagnosis of vehicle faults.

[0049] However, processing and analyzing this multivariate time series data and applying it to fault diagnosis in automotive thermal management systems still faces numerous challenges. First, related automotive thermal management system monitoring and fault diagnosis algorithms mostly rely on traditional fault diagnosis methods, such as rule-based methods, statistical methods, and machine learning algorithms. However, these methods often suffer from high computational complexity and high false positive rates when processing high-dimensional, nonlinear, and noisy multivariate time series data. Second, due to the real-time and continuous requirements of vehicle driving data, improving the computational efficiency of fault diagnosis algorithms while ensuring fault diagnosis accuracy is also a current research focus. With the continuous development of big data and artificial intelligence technologies, there is an urgent need for fault diagnosis algorithms that can efficiently process complex time series data to improve the level of intelligent vehicle fault diagnosis.

[0050] In summary, developing an efficient and accurate fault diagnosis algorithm for multivariate time series data in the process of vehicle operation has important practical significance and technical value.

[0051] In related technologies, autoencoders are used for fault diagnosis. An autoencoder is a type of neural network that uses the input data X0 itself as supervision to guide the neural network in its attempt to learn a mapping relationship, thereby obtaining a reconstructed output X0. An autoencoder generally consists of two sub-networks: an encoder and a decoder. The encoder compresses the input data into a low-dimensional representation, and the decoder uses this low-dimensional representation to reconstruct the input data. The model parameters are trained by minimizing the reconstruction loss to achieve the goal of reconstructing the input data.

[0052] In fault diagnosis, auto-encoder (AE)-based methods primarily utilize the AE's ability to reconstruct input data. When the input data is normal, the AE reconstructs it well. However, when the input data is faulty, the reconstruction error increases due to differences in the characteristics of the faulty data and normal data.

[0053] Furthermore, when faced with massive amounts of high-dimensional data for fault diagnosis, traditional autoencoder-based algorithms can suffer from poor reconstruction performance. This is because high-dimensional data contains a large amount of complex information, and the autoencoder may not be able to fully capture the subtle differences between normal and faulty data when processing this information. When these differences are not effectively captured by the model, the difference in the mean squared error (MSE) between the normal and faulty datasets becomes insignificant, affecting the accuracy of fault diagnosis.

[0054] On the other hand, while related technologies using generative adversarial networks (GANs) have powerful generative capabilities, their training process is often plagued by issues such as pattern collapse and non-convergence. These issues lead to a lack of stability in GAN models, making them difficult to implement and deploy in practical applications.

[0055] In view of this, the embodiments of the present application provide a vehicle fault diagnosis method, apparatus, device and storage medium. The present application obtains a fault diagnosis model based on dual autoencoder training to improve the accuracy of the fault diagnosis model, thereby improving the accuracy of the target vehicle fault score calculation based on the fault diagnosis model, diagnostic parameters and fault score calculation model, thereby improving the accuracy of the vehicle thermal management system fault diagnosis.

[0056] Figure 1This is a flow chart of a vehicle fault diagnosis method provided in an embodiment of the present application. This method can be applied to a vehicle thermal management system to accurately diagnose faults in the vehicle thermal management system. This method can be executed by a vehicle fault diagnosis device, which can be implemented in software and / or hardware and can be configured in a processor or controller of the vehicle thermal management system. Figure 1 , the method comprises the following steps:

[0057] Step 110: Obtain a fault diagnosis model.

[0058] Among them, the fault diagnosis model is trained based on the pre-built dual autoencoder.

[0059] In some embodiments, the dual autoencoder includes a first autoencoder and a second autoencoder; the first autoencoder includes a first encoder and a first decoder, and the second autoencoder includes a first encoder and a second decoder; the first decoder and the second decoder have the same structure.

[0060] An autoencoder is a type of neural network that uses the input data X as supervision to guide the neural network in its attempt to learn a mapping relationship, thereby obtaining a reconstructed output X. An autoencoder consists of two sub-networks: an encoder and a decoder. The encoder compresses the input data into a low-dimensional representation, while the decoder uses this low-dimensional representation to reconstruct the input data. The model parameters are trained by minimizing the reconstruction loss to achieve the goal of reconstructing the input data.

[0061] The dual autoencoder of the embodiment of the present application includes a first autoencoder and a second autoencoder. The first autoencoder and the second autoencoder share the same encoder, namely the first encoder. The first autoencoder and the second autoencoder use decoders with the same structure, namely the first decoder and the second decoder, respectively.

[0062] Figure 2 : It is a schematic diagram of the principle structure of the fault diagnosis model provided in the embodiment of the present application. Exemplarily, the fault diagnosis model provided in the embodiment of the present application is a deep learning model of a dual autoencoder structure based on the idea of ​​generating confrontation, that is, a dual autoencoder generative adversarial network (GAN). The GAN network consists of two neural networks: a generator and a discriminator. The working principle of the GAN network can be summarized as a "game" process. During the training process, the generator and the discriminator compete with each other and continuously adjust their parameters to achieve the optimal state.

[0063] For example, see Figure 2 , the fault diagnosis model consists of an encoder and two decoders with the same structure (i.e. Figure 2 The dual autoencoder operates as follows: input data X is fed into the encoder, which then reduces the dimensionality of the output to obtain low-dimensional feature data Z. The low-dimensional feature data Z is then fed into decoders 1 and 2, respectively. Decoders 1 and 2 reconstruct the input data (i.e., low-dimensional feature data Z) using adversarial methods, resulting in their respective decoder outputs: decoder 1 outputs AE1(X), and decoder 2 outputs AE2(X).

[0064] In some embodiments, the network structure of the first encoder includes a convolutional layer, a first long short-term memory network layer, and a second long short-term memory network layer; the network structure of the first decoder includes a third long short-term memory network layer and a deconvolution layer.

[0065] The first decoder and the second decoder have the same network structure, and the convolution layer is a one-dimensional convolution layer.

[0066] Among them, the Long Short-Term Memory (LSTM) network is a special variant of the Recurrent Neural Network (RNN) designed specifically for processing and predicting long-term dependencies in time series data. LSTM is sensitive to time and can learn patterns and features in time series data.

[0067] Figure 3 Schematic diagram of the principle structure of the encoder provided in the embodiment of the present application. For example, the network structure of the encoder in the embodiment of the present application is as follows Figure 3 As shown in the figure, it includes a one-dimensional convolution layer and two LSTM layers. Among them, LSTM has a very powerful ability to process time series data, and the convolution network layer has a strong ability to extract data features. Among them, the advantage of using two LSTM network layers is that the second LSTM network layer can make the model have better reconstruction capabilities. The working principle of the encoder is: First, for the input data X (where X is multivariate time series data, such as X 11 、X 12 、X 13 ,…X 1n , X 21 、X 22 、X 23 ,…X 2nThe algorithm uses a one-dimensional convolutional neural network (CNN) to perform sliding convolution on the input data X along the time series dimension, capturing short-range sequence fluctuations and compressing the input data into a compact vector while preserving its structure. To learn about time series fluctuations, the input data X is further reduced in dimensionality through two LSTM layers, resulting in a more compact latent representation (i.e., low-dimensional representation data) as output. A more compact latent representation represents high-dimensional input data using lower-dimensional feature data, condensing and refining the input data. For example, 8x64 data can be converted into a 16x1 latent feature representation.

[0068] Figure 4 Schematic diagram of the principle structure of the decoder provided in the embodiment of the present application. For example, the structures of the two decoders in the embodiment of the present application are the same, and the network structure of the decoder is as follows Figure 4 As shown in the figure, the decoder's function is to restore the input feature data. The decoder consists of an LSTM network layer and a deconvolutional network layer. The decoder operates by receiving the output of the encoder network as input (i.e., the input low-dimensional representation data). Then, through the LSTM network layer and the deconvolutional network layer, the low-dimensional latent representation (i.e., the low-dimensional representation data) is reconstructed into data close to the original input. This output is compared with the original input data, and the loss function is determined by calculating the mean squared error. The backpropagation algorithm is then used to optimize the parameters of the entire network.

[0069] It should be noted that the network structure of the encoder and decoder can also be other types of neural network models, such as the Gated Recurrent Unit (GRU), the Transformer model (Transformer) and the Multi-Layer Perceptron (MLP), etc., which can be set according to actual conditions and are not specifically limited here.

[0070] In some embodiments, the method for obtaining a fault diagnosis model includes the following steps:

[0071] Step 1: Obtain historical diagnostic parameters of the test vehicle. The historical diagnostic parameters are used to characterize key parameters of the thermal management system of the test vehicle.

[0072] The test vehicle is a vehicle used to provide diagnostic parameters during the fault diagnosis model training phase, and the vehicle is a car, electric car, or other vehicle.

[0073] Among them, historical diagnostic parameters are key parameters of the test vehicle in historical driving segments.

[0074] In the vehicle thermal management system, the collection and preprocessing of multivariate time series data of vehicle driving segments (such as multivariate time series data of vehicle drive motors) is the basis for building an efficient fault diagnosis model. The multivariate time series data of the drive motor includes multiple dimensions such as electrical, mechanical, and temperature, such as speed, torque, rear motor D-axis current feedback, rear motor Q-axis current feedback, rear motor D-axis voltage feedback, rear motor Q-axis voltage feedback, water inlet temperature, controller temperature, oil pump current and oil pump temperature and other key parameters. These parameters together reflect the motor heating condition and motor cooling condition, and are key parameters for fault diagnosis. For example, in an embodiment of the present application, the multivariate time series data of the drive motor of the test vehicle including speed, torque, rear motor D-axis current feedback, rear motor Q-axis current feedback, rear motor D-axis voltage feedback, rear motor Q-axis voltage feedback, water inlet temperature, and controller temperature are used as an example for explanation. The same applies below and will not be repeated.

[0075] It should be noted that after acquiring multivariate time series data, the data is also preprocessed. Preprocessing includes data cleaning, normalization, and sliding window processing. Exemplarily, the specific process of data preprocessing is as follows: the raw data is cleaned to remove missing values ​​and outliers due to sensor communication anomalies, and continuous driving segments are divided according to the speed and torque not being completely zero, thereby ensuring that the data within each segment is continuous, thereby improving the availability of data for subsequent fault diagnosis model training. The cleaned data is then normalized so that all data can be compared and analyzed on the same scale. Finally, the data is subjected to sliding window processing to reflect the data's time series. Exemplarily, the window length of the sliding window processing is 64 sampling points, and the step size is 32 sampling points (wherein the window length and step size can also be other values, which can be set according to actual conditions and are not specifically limited here). As a result, the sliding window processing allows the subsequent fault diagnosis model training process to utilize time series data, thereby further improving the accuracy of the fault diagnosis model and, in turn, the accuracy of vehicle thermal management system fault diagnosis.

[0076] Step 2: Determine a fault diagnosis model based on historical diagnosis parameters, the first autoencoder and the second autoencoder.

[0077] Specifically, historical diagnostic parameters of the test vehicle are obtained, and the historical diagnostic parameters are input into the first autoencoder and the second autoencoder for adversarial training to obtain a fault diagnosis module, thereby improving the accuracy of the fault diagnosis model, thereby improving the accuracy of the target vehicle's fault score calculation based on the fault diagnosis model, diagnostic parameters and fault score calculation model, and further improving the accuracy of vehicle thermal management system fault diagnosis.

[0078] In some embodiments, determining a fault diagnosis model based on historical diagnostic parameters, a first autoencoder, and a second autoencoder includes: pre-training the first autoencoder and the second autoencoder according to the historical diagnostic parameters, respectively; performing adversarial training on the first autoencoder and the second autoencoder according to the historical diagnostic parameters, respectively; and repeating the pre-training and adversarial training operations until the fault diagnosis model is determined when preset iterative training conditions are met.

[0079] Among them, the preset iterative training conditions include meeting the preset number of iterations, etc., which can be set according to actual conditions and are not specifically limited here.

[0080] Specifically, the training process of the fault diagnosis model includes two stages, namely the pre-training stage and the adversarial training stage. The specific implementation process of the training process is as follows: First, two autoencoders are trained to reconstruct the original input data X to perform model pre-training. Secondly, the two autoencoders are trained in an adversarial manner. The first autoencoder AE1 attempts to deceive the second autoencoder AE2 to generate reconstructed data AE1(X) based on the real data X (i.e., the original input data). The goal of the second autoencoder AE2 is to determine whether the data is the real X or the reconstructed data AE1(X) generated by the first autoencoder AE1. This strengthens the fault diagnosis model's ability to understand the data, grasps the subtle differences between normal data and fault data in certain aspects, and thus performs more efficient and accurate fault detection and diagnosis.

[0081] In the pre-training stage, two autoencoders are trained to learn to reconstruct the input data X. The input data X is compressed by the encoder to the latent space Z (i.e., low-dimensional feature data), and then reconstructed by each decoder. The loss function of the two autoencoders at this stage is:

[0082] L1=‖X-AE1(x)‖2;

[0083] L2=‖X-AE2(x)‖2;

[0084] In the adversarial training phase, the second autoencoder AE2 is trained to distinguish whether the input data is real data or data generated from the first autoencoder AE1, and at the same time, the first autoencoder AE1 is trained to deceive the second autoencoder AE2. The data from the first autoencoder AE1 is reconstructed by the second autoencoder AE2. Using the adversarial training setting, the goal of the first autoencoder AE1 is to minimize the difference between the input data X and the output of the second autoencoder AE2 (considered as a generator). The goal of the second autoencoder AE2 is to maximize this difference (considered as a discriminator). The first autoencoder AE1 is trained based on deceiving the second autoencoder AE2, and the second autoencoder AE2 distinguishes the candidate data reconstructed by the first autoencoder AE1 from the real data. The training objectives are:

[0085] L1=+‖X-AE2(AE1(x))‖2;

[0086] L2=-||X-AE2(AE1(x))||2;

[0087] The two autoencoders have a dual purpose: the first autoencoder AE1 minimizes the reconstruction error of the input data X (i.e., the pre-training phase) and minimizes the error between the input data X and the reconstructed output of the second autoencoder AE2 (i.e., the pre-training phase). Like the first autoencoder AE1, the second autoencoder AE2 minimizes the reconstruction error of the input data X (i.e., the pre-training phase) and then maximizes the reconstruction error of the input data reconstructed by the first autoencoder AE1 (i.e., the adversarial training phase). The dual-purpose training objective of each autoencoder is expressed as a combination of the above pre-training and adversarial training in an evolutionary scheme, where the proportions of each part evolve over time:

[0088]

[0089] Among them, X-AE1(x) is the first target output, which is used to characterize the mean square error of the first decoder; X-AE2(AE1(x)) is the second target output, which is used to characterize the mean square error of the second decoder.

[0090] Therefore, by designing a loss function, the fault diagnosis model can be trained in stages, and the first autoencoder and the second autoencoder are pre-trained and iteratively trained multiple times through historical diagnostic parameters, wherein the pre-training stage is carried out in the early stage of training to train the fault diagnosis model to complete the reconstruction operation, and the adversarial training stage is carried out in the later stage of training to train the two autoencoders for adversarial learning until the iteration training condition is reached and the fault diagnosis model is obtained when the iteration is stopped. The fault diagnosis model obtained at this time is the optimal model. Since the fault diagnosis model integrates the ideas of autoencoders and adversarial learning, the fault diagnosis model can learn the subtle differences between the data more deeply, which is conducive to improving the accuracy of subsequent vehicle thermal management system fault diagnosis. And the model training method provided in this application can automatically change the loss function by the growth of training rounds without manual setting. In addition, since the LSTM layer and one-dimensional convolution layer are used in each autoencoder of this application, it can capture the time series dependency and feature correlation of the input data, grasp the changing trend and local features of the data in the time dimension, and extract the key features of the time series, which is conducive to improving the accuracy of the fault diagnosis model.

[0091] It should be noted that the calculation of the loss function provided in the embodiment of the present application can also use cross entropy loss and absolute value error, etc., which can be set specifically according to actual conditions and are not specifically limited here.

[0092] Cross-entropy loss is a concept in information theory that measures the difference between two probability distributions. In machine learning, especially in classification problems, cross-entropy loss is often used to measure the difference between the probability distribution predicted by the model and the probability distribution of the true label.

[0093] Among them, the absolute value error, also known as the mean absolute error (MAE), is often used in regression problems to measure the average absolute difference between the model's predicted value and the true value.

[0094] Step 120: Obtain current diagnostic parameters of the target vehicle.

[0095] Among them, the current diagnostic parameters are used to characterize the key parameters of the target vehicle thermal management system.

[0096] Among them, the target vehicle is a vehicle that needs to perform thermal management system fault diagnosis.

[0097] It should be noted that in the technical solution of the embodiment of the present application, the test vehicle and the target vehicle may be the same or different, and may be set according to actual conditions, and no specific limitation is made here.

[0098] Specifically, historical diagnostic parameters of the test vehicle are obtained; these historical diagnostic parameters are used to characterize key parameters of the test vehicle's thermal management system. A fault diagnosis model is determined based on the historical diagnostic parameters, a first autoencoder, and a second autoencoder. Current diagnostic parameters of the target vehicle are then obtained and used to subsequently determine a fault score for the target vehicle based on the current diagnostic parameters.

[0099] Step 130: Determine the fault score of the target vehicle based on the current diagnostic parameters, the fault diagnosis model, and the pre-built fault score calculation model.

[0100] Among them, the fault score is used to diagnose the thermal management system of the target vehicle.

[0101] Specifically, the system obtains historical diagnostic parameters of the test vehicle, which characterize key parameters of the test vehicle's thermal management system. A fault diagnosis model is determined based on the historical diagnostic parameters, a first autoencoder, and a second autoencoder. Then, the system obtains current diagnostic parameters of the target vehicle. Finally, the system determines the target vehicle's fault score based on the current diagnostic parameters, the fault diagnosis model, and a pre-built fault score calculation model. This determines whether the target vehicle has experienced a fault based on the fault score.

[0102] In some embodiments, the fault score of the target vehicle is determined based on current diagnostic parameters, a fault diagnosis model, and a pre-built fault score calculation model, including: determining a first target output and a second target output based on the fault diagnosis model; wherein the first target output is used to characterize the mean square error of the first decoder, and the second target output is used to characterize the mean square error of the second decoder; and determining the fault score of the target vehicle based on the current diagnostic parameters, the first target output, the second target output, and the fault score calculation model.

[0103] Among them, the specific process of determining the first target output and the second target output according to the fault diagnosis model is: pre-training the first autoencoder and the second autoencoder according to the historical diagnosis parameters respectively; performing adversarial training on the first autoencoder and the second autoencoder according to the historical diagnosis parameters respectively; repeating the pre-training and adversarial training operations until the fault diagnosis model is determined when the preset iterative training conditions are met, and the mean square error of the first decoder and the mean square error of the second decoder obtained when the preset iterative training conditions are met are used as the first target output and the second target output respectively.

[0104] In some embodiments, a fault score of a target vehicle is determined based on current diagnostic parameters, a first target output, a second target output, and a fault score calculation model, including: determining a first item based on the current diagnostic parameters, the first target output, and a first preset weight; determining a second item based on the current diagnostic parameters, the second target output, and a second preset weight; wherein the first item and the second item are used to represent two items of the fault score calculation model; and determining the fault score based on the sum of the first item and the second item.

[0105] The second preset weight is one minus the first preset weight.

[0106] The formula for determining the fault score is:

[0107] score=α‖X-AE1(x)‖2+(1-α)‖X-AE2(AE1(x))‖2;

[0108] Where score is the fault score; α is the first preset weight; 1-α is the second preset weight; X-AE1(x) is the first target output; X-AE2(AE1(x)) is the second target output.

[0109] Exemplarily, the first preset weight α ranges from 0 to 1. The first preset weight α is selected to maximize the difference between the scores of normal vehicles and faulty vehicles in the test set. Exemplarily, in the technical solution of the embodiment of the present application, the first preset weight α is 0.35.

[0110] It can be understood that the vehicle fault diagnosis method provided in the embodiment of the present application improves the accuracy of the fault diagnosis model by obtaining a fault diagnosis model based on dual autoencoder training, thereby improving the accuracy of the target vehicle fault score calculation based on the fault diagnosis model, diagnostic parameters and fault score calculation model, and further improving the accuracy of vehicle thermal management system fault diagnosis.

[0111] Figure 5 This is a schematic diagram of the overall process of a vehicle fault diagnosis method provided in an embodiment of the present application. Figure 5 The overall workflow of the vehicle fault diagnosis method provided by the embodiment of the present application is as follows: First, feature selection, specifically selecting the eight features of speed, torque, rear motor D-axis current feedback, rear motor Q-axis current feedback, rear motor D-axis voltage feedback, rear motor Q-axis voltage feedback, water inlet temperature, and controller temperature. Then, data collection and data preprocessing, collecting the multivariate time series of the vehicle driving segment, including the eight feature data of speed, torque, rear motor D-axis current feedback, rear motor Q-axis current feedback, rear motor D-axis voltage feedback, rear motor Q-axis voltage feedback, water inlet temperature, and controller temperature, and preprocessing these eight feature data. Secondly, training the fault diagnosis model, including two stages of pre-training and adversarial training, and obtaining the optimal fault diagnosis model through training. According to the optimized fault diagnosis model, the fault score and / or fault score line of the target vehicle are obtained. Finally, according to the fault score and / or fault score line of the target vehicle, it is judged whether the thermal management system of the target vehicle has an oil leakage or water leakage fault.

[0112] Figure 6 It is a schematic diagram of the distribution difference of the fault scores of normal cars and faulty cars provided in the embodiment of the present application. Under normal circumstances, the fault score of a normal car should be relatively close to the fault score of the normal car test set (i.e., the test vehicle), and should be more clearly distinguished from the fault score of the faulty car. Therefore, based on the difference in the fault scores of the normal car and the faulty car, the critical score line (i.e., the fault score line) is determined, and it is used to judge whether the car has a fault. For example, the fault score distribution of a normal car and a faulty car is as follows Figure 6 As shown, curve S1 represents the fault score distribution of the normal car training set, curve S2 represents the fault score distribution of the normal car test set, curve S3 represents the fault score distribution of the faulty car, and the middle vertical bar L0 represents the fault score line. Figure 6 It can be seen that the fault score distribution of the normal vehicle training set is similar to the fault score distribution of the normal vehicle test set, while the fault score distribution of the faulty vehicle is significantly different from that of the normal vehicle. This shows that the vehicle fault diagnosis method provided in the embodiment of the present application has a good fault diagnosis effect.

[0113] Furthermore, in order to verify the effect of the fault diagnosis model provided by the embodiment of the present application on automobile fault diagnosis, experiments were conducted through local experiments and online verification. The data used in the experiment are two types of data for new energy vehicles: normal vehicles and faulty vehicles, thereby ensuring the diversity and representativeness of the data. After the model training is completed, the fault diagnosis model will draw a fault score change curve for each vehicle. If the proportion of the fault score change curve of the vehicle that is greater than the fault score line on that day exceeds the fault proportion line, the vehicle is judged to be a faulty vehicle. Among them, the fault score change curve of each vehicle is used to represent the real-time fault score change of each vehicle. Among them, the proportion greater than the fault score line refers to the ratio of the number of fault scores greater than the fault score line in the fault score change curve to the sum of all fault scores in the fault score change curve.

[0114] For example, the fault score line of the embodiment of the present application can be 0.022, and the fault ratio line can be 20%. In addition, the fault score line and the fault ratio line can also be set to other values, which can be set according to actual conditions and are not specifically limited here.

[0115] Figure 7 is the fault score curve of a normal vehicle provided in the embodiment of this application, Figure 8 This is the fault score curve of the faulty vehicle provided in the embodiment of the present application. If the fault score of a normal vehicle is below the fault score line for most of the day, it means that it is normal, for example Figure 7 The curve L1 shown is the change of the fault score of a normal car. However, the faulty car will be above the fault score line for most of the day, indicating a fault, e.g. Figure 8 The curve L2 shown is the change in the fault score of the faulty vehicle.

[0116] Finally, the present embodiment was tested on an offline dataset. The test results showed that in a normal vehicle dataset, 130 normal vehicles were detected out of 138 vehicles, with an accuracy rate of 97.4%. In a faulty vehicle dataset, 30 out of 31 oil-leaking vehicles were detected, with an accuracy rate of 96.8%. And 18 out of 19 water-leaking vehicles were detected, with an accuracy rate of 94.7%. This shows that the fault diagnosis model provided by the vehicle fault diagnosis method provided by the present embodiment has efficient and accurate fault detection capabilities.

[0117] In summary, the vehicle fault diagnosis method provided by the embodiment of the present application can achieve: early warning and precise positioning of vehicle faults, improve the accuracy and reliability of vehicle fault diagnosis, and provide strong guarantees for the safe and stable operation of the vehicle. Moreover, the fault diagnosis algorithm proposed in the embodiment of the present application cleverly combines the advantages of the autoencoder structure and the generative adversarial network. Specifically, the algorithm amplifies the reconstruction error of abnormal data by training two autoencoder structures for adversarial learning. This adversarial learning mechanism can prompt the fault diagnosis model to pay more attention to the difference between normal data and fault data, thereby improving the accuracy of fault diagnosis.

[0118] In addition, the embodiment of the present application builds an incomplete autoencoder and adopts two autoencoder structures for confrontation. The generator and the discriminator compete with each other and optimize each other, and introduce a variable loss function loss term to comprehensively optimize the two adversarial autoencoders and extract key features in the time series data, thereby achieving accurate identification of vehicle faults. The two models learn from each other in the confrontation, and their capabilities become stronger with the training process. They can capture the mutual connection between multiple variables in normal data and reconstruct them better. When the input data comes from a faulty vehicle, the model can capture the slight differences between the data and identify the fault. As a result, the algorithm shows higher accuracy. Compared with GAN, this algorithm avoids problems such as pattern collapse and non-convergence, making the model easier to implement and deploy.

[0119] In summary, the vehicle fault diagnosis method proposed in the embodiment of the present application has significant advantages in processing massive high-dimensional data. It can not only improve the accuracy of fault diagnosis, but also ensure the stability of the model and the training efficiency.

[0120] Figure 9 This is a schematic diagram of the principle structure of a vehicle fault diagnosis device provided in the embodiment of the present application. Correspondingly, the embodiment of the present application also provides a vehicle fault diagnosis device, please refer to Figure 9 The vehicle fault diagnosis device 100 includes: a first acquisition module 101, used to obtain a fault diagnosis model; wherein the fault diagnosis model is obtained based on pre-built dual autoencoder training; a second acquisition module 102, used to obtain current diagnostic parameters of the target vehicle; wherein the current diagnostic parameters are used to characterize key parameters of the thermal management system of the target vehicle; a determination module 103, used to determine the fault score of the target vehicle based on the current diagnostic parameters, the fault diagnosis model and a pre-built fault score calculation model; wherein the fault score is used to diagnose the thermal management system of the target vehicle.

[0121] The technical solution of the embodiment of the present application provides a vehicle fault diagnosis device, which obtains a fault diagnosis model through dual autoencoder training to improve the accuracy of the fault diagnosis model, thereby improving the accuracy of the target vehicle fault score calculation based on the fault diagnosis model, diagnostic parameters and fault score calculation model, and further improving the accuracy of vehicle thermal management system fault diagnosis.

[0122] In some embodiments, the dual autoencoder includes a first autoencoder and a second autoencoder; the first autoencoder includes a first encoder and a first decoder, and the second autoencoder includes a first encoder and a second decoder; the first decoder and the second decoder have the same structure.

[0123] In some embodiments, the determination module 103 is further configured to: determine a first target output and a second target output according to the fault diagnosis model; wherein the first target output is used to represent the mean square error of the first decoder, and the second target output is used to represent the mean square error of the second decoder;

[0124] A fault score of the target vehicle is determined based on current diagnostic parameters, a first target output, a second target output, and a fault score calculation model.

[0125] In some embodiments, the determination module 103 is further configured to: determine a first item according to the current diagnostic parameter, the first target output, and the first preset weight;

[0126] determining a second item according to the current diagnostic parameter, the second target output, and the second preset weight; wherein the first item and the second item are used to represent two items of the fault score calculation model;

[0127] The fault score is determined based on the sum of the first and second terms.

[0128] In some embodiments, the first acquisition module 101 is further used to: acquire historical diagnostic parameters of the test vehicle; wherein the historical diagnostic parameters are used to characterize key parameters of the thermal management system of the test vehicle;

[0129] A fault diagnosis model is determined according to historical diagnosis parameters, the first autoencoder and the second autoencoder.

[0130] In some embodiments, the first acquisition module 101 is further configured to: pre-train the first autoencoder and the second autoencoder respectively according to the historical diagnostic parameters;

[0131] Performing adversarial training on the first autoencoder and the second autoencoder respectively according to the historical diagnostic parameters;

[0132] The pre-training and adversarial training operations are repeatedly performed until the fault diagnosis model is determined when a preset iterative training condition is met.

[0133] In some embodiments, the network structure of the first encoder includes a convolutional layer, a first long short-term memory network layer, and a second long short-term memory network layer; the network structure of the first decoder includes a third long short-term memory network layer and a deconvolution layer.

[0134] Figure 10 Schematic diagram of the structure of an electronic device provided in the embodiment of the present application. Correspondingly, the embodiment of the present application also provides an electronic device, please refer to Figure 10 The electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the vehicle fault diagnosis method described above are implemented. Since the vehicle fault diagnosis method has been described in detail above, it will not be repeated here.

[0135] Accordingly, an embodiment of the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-mentioned vehicle fault diagnosis method. Since the vehicle fault diagnosis method has been described in detail above, it will not be repeated here.

[0136] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0137] The above is a detailed introduction to the vehicle fault diagnosis method, device, equipment and storage medium provided in the embodiments of the present application, and specific examples are used to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the technical solution and core idea of ​​the present application; ordinary technicians in this field should understand that they can still modify the technical solutions recorded in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein; and these modifications or replacements do not cause the essence of the corresponding technical solution to deviate from the scope of the technical solution of the embodiments of the present application.

Claims

1. A vehicle fault diagnosis method, characterized in that: The method is applied to diagnosing a fault in a vehicle thermal management system; the method comprises: Obtaining a fault diagnosis model; wherein the fault diagnosis model is obtained based on pre-built dual autoencoder training; Acquiring current diagnostic parameters of the target vehicle; wherein the current diagnostic parameters are used to characterize key parameters of the thermal management system of the target vehicle; A fault score of the target vehicle is determined based on the current diagnostic parameters, the fault diagnosis model, and a pre-built fault score calculation model; wherein the fault score is used to diagnose a thermal management system of the target vehicle.

2. The vehicle fault diagnosis method according to claim 1, characterized in that: The dual autoencoder includes a first autoencoder and a second autoencoder; the first autoencoder includes a first encoder and a first decoder, and the second autoencoder includes the first encoder and a second decoder; the first decoder and the second decoder have the same structure.

3. The vehicle fault diagnosis method according to claim 2, characterized in that: Determining the fault score of the target vehicle according to the current diagnostic parameters, the fault diagnosis model, and a pre-built fault score calculation model includes: Determine a first target output and a second target output according to the fault diagnosis model; wherein the first target output is used to represent the mean square error of the first decoder, and the second target output is used to represent the mean square error of the second decoder; A fault score of the target vehicle is determined based on the current diagnostic parameters, the first target output, the second target output, and the fault score calculation model.

4. The vehicle fault diagnosis method according to claim 3, characterized in that: Determining the fault score of the target vehicle according to the current diagnostic parameter, the first target output, the second target output, and the fault score calculation model includes: determining a first item according to the current diagnostic parameter, the first target output, and a first preset weight; determining a second item according to the current diagnostic parameter, the second target output, and a second preset weight; wherein the first item and the second item are used to represent two items of the fault score calculation model; The fault score is determined according to the sum of the first term and the second term.

5. The vehicle fault diagnosis method according to claim 2, characterized in that: The method for obtaining the fault diagnosis model includes: Acquiring historical diagnostic parameters of a test vehicle; wherein the historical diagnostic parameters are used to characterize key parameters of a thermal management system of the test vehicle; The fault diagnosis model is determined according to the historical diagnosis parameters, the first autoencoder, and the second autoencoder.

6. The vehicle fault diagnosis method according to claim 5, characterized in that: The determining the fault diagnosis model according to the historical diagnostic parameters, the first autoencoder, and the second autoencoder includes: Pre-training the first autoencoder and the second autoencoder respectively according to the historical diagnostic parameters; Performing adversarial training on the first autoencoder and the second autoencoder respectively according to the historical diagnostic parameters; The pre-training and adversarial training operations are repeatedly performed until the fault diagnosis model is determined when a preset iterative training condition is met.

7. The vehicle fault diagnosis method according to claim 2, characterized in that: The network structure of the first encoder includes a convolution layer, a first long short-term memory network layer and a second long short-term memory network layer; the network structure of the first decoder includes a third long short-term memory network layer and a deconvolution layer.

8. A vehicle fault diagnosis device, characterized in that: Applicable to diagnosing faults in a vehicle thermal management system; the device comprises: A first acquisition module is used to acquire a fault diagnosis model; wherein the fault diagnosis model is obtained based on pre-built dual autoencoder training; A second acquisition module is used to acquire current diagnostic parameters of the target vehicle; wherein the current diagnostic parameters are used to characterize key parameters of the thermal management system of the target vehicle; A determination module is configured to determine a fault score of the target vehicle based on the current diagnostic parameters, the fault diagnosis model, and a pre-built fault score calculation model; wherein the fault score is used to diagnose a thermal management system of the target vehicle.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the vehicle fault diagnosis method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the vehicle fault diagnosis method according to any one of claims 1 to 7 is implemented.

Images