Real-time fault diagnosis method and device for steering gear based on unsupervised domain adaptation

Abstract

The present invention relates to the field of fault diagnosis technology, and discloses a real-time fault diagnosis method and device for a steering gear based on unsupervised domain adaptation, the method comprising: using a pre-built steering gear intelligent fault diagnosis framework to extract data features of labeled source domain steering gear data and unlabeled target domain steering gear data; performing feature alignment and feature discrimination on the data features to obtain discriminated source domain features and discriminated target domain features; calculating the fault diagnosis loss value between the discriminated source domain features and the discriminated target domain features and the source domain steering gear data and the target domain steering gear data; adjusting the parameters of the steering gear intelligent fault diagnosis framework according to the fault diagnosis loss value to obtain a steering gear fault detection model; and when receiving steering gear motion signal data, using the steering gear fault detection model to calculate the steering gear fault type corresponding to the target steering gear. Through the implementation of the present invention, the steering gear fault detection model is used to perform steering gear fault diagnosis, and the steering gear fault type can be accurately calculated in real time.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and in particular to a method and device for real-time fault diagnosis of a steering gear based on unsupervised domain adaptation. Background Art

[0002] The servo is one of the commonly used mechanical equipment in the fields of aerospace, shipbuilding and automobiles. As a high-precision position servo drive, the servo operates the rudder to achieve the swing of the rudder, thereby maintaining the smooth operation of the equipment. However, the servo is prone to failure when it runs for a long time or is affected by uncontrollable factors. If it is not detected and measures are not taken in time, the stability of the applied system will be reduced, thereby affecting the working state of the equipment, causing serious consequences and inestimable losses. In order to ensure the safe and stable operation of the servo, it is necessary to pay attention to the working state of the system at all times, test and collect parameters of the servo, and analyze the alignment performance so that potential problems can be discovered in time to reduce losses. Therefore, the fault detection of the servo is receiving more and more attention.

[0003] At present, the main steering gear diagnosis methods mainly include the fault diagnosis system of hydraulic steering gear based on regression support vector machine (SVR), the measurable parameters of hydraulic steering gear in normal state are used to train the SVR model, and the working state of hydraulic steering gear is dynamically monitored. A multi-parameter model fault diagnosis method that designs corresponding fault state observers and parameter estimators for different fault types can accurately describe the dynamic characteristics of faulty steering gear and detect the fault of steering gear in time. Although the model-based fault diagnosis method has some advantages, with the development of computing technology, control technology and artificial intelligence theory, the system has become more and more complex, which may involve the interaction of multiple subsystems. It becomes complicated and time-consuming to build an accurate model. The system may be disturbed by uncertain factors such as environmental changes and noise, resulting in the steering gear fault data in complex dynamic environments showing small samples, incomplete labels, and large differences in data distribution. At the same time, the mapping relationship between the characteristic data and the working state of different steering gears under different load torques in complex dynamic environments is difficult to quantitatively express, resulting in low accuracy and large error in online fault diagnosis of steering gears. Therefore, how to perform data-driven real-time and accurate fault diagnosis of steering gears based on models and limited data has become an urgent problem to be solved. Summary of the invention

[0004] The present invention provides a method and device for real-time fault diagnosis of a steering gear based on unsupervised domain adaptation, the main purpose of which is to solve the problems of poor efficiency and low accuracy in steering gear fault diagnosis.

[0005] To achieve the above-mentioned purpose, the present invention provides a real-time fault diagnosis method for a steering gear based on unsupervised domain adaptation, comprising: obtaining labeled source domain steering gear data and unlabeled target domain steering gear data, and using a pre-built steering gear intelligent fault diagnosis framework to respectively extract data features of the source domain steering gear data and the target domain steering gear data; using the steering gear intelligent fault diagnosis framework to perform feature alignment and feature discrimination on the data features to obtain discriminated source domain features and discriminated target domain features; calculating the fault diagnosis loss value between the discriminated source domain features and the discriminated target domain features and the source domain steering gear data and the target domain steering gear data; adjusting the parameters of the steering gear intelligent fault diagnosis framework according to the fault diagnosis loss value until the fault diagnosis loss value is less than a preset loss value threshold, and obtaining a steering gear fault detection model corresponding to the steering gear intelligent fault diagnosis framework; when receiving the steering gear motion signal data collected by the edge intelligent chip carried on the target steering gear, using the steering gear fault detection model to calculate the steering gear fault type corresponding to the target steering gear.

[0006] The present invention also provides a real-time fault diagnosis device for a steering gear based on unsupervised domain adaptation, comprising: a data feature extraction module, used to obtain labeled source domain steering gear data and unlabeled target domain steering gear data, and use a pre-built steering gear intelligent fault diagnosis framework to respectively extract data features of the source domain steering gear data and the target domain steering gear data; a feature alignment and feature discrimination module, used to use the steering gear intelligent fault diagnosis framework to perform feature alignment and feature discrimination on data features, and obtain discriminated source domain features and discriminated target domain features; a fault diagnosis loss value calculation module, used to calculate the fault diagnosis loss value between the discriminated source domain features and the discriminated target domain features and the source domain steering gear data and the target domain steering gear data; a steering gear fault detection model determination module, used to adjust the parameters of the steering gear intelligent fault diagnosis framework according to the fault diagnosis loss value, until the fault diagnosis loss value is less than a preset loss value threshold, and obtain the steering gear fault detection model corresponding to the steering gear intelligent fault diagnosis framework; a steering gear fault type diagnosis module, used to calculate the steering gear fault type corresponding to the target steering gear using the steering gear fault detection model when receiving the steering gear motion signal data collected by the edge intelligent chip carried on the target steering gear.

[0007] The embodiment of the present invention can effectively filter a large amount of noise contained in the servo data by extracting data features of the source domain servo data and the target domain servo data, while retaining the information contained in the original servo signal, solving the gradient vanishing problem, and improving the accuracy of feature extraction; aligning and distinguishing the data features can continuously improve the quality of generated samples, thereby generating alignment features that are closer to the labeled source domain servo data, and improving the accuracy of the servo fault detection model training; calculating the fault diagnosis loss value between the distinguished source domain features and the distinguished target domain features and the source domain servo data and the target domain servo data, and adjusting the parameters of the servo intelligent fault diagnosis framework through the fault diagnosis loss value, so as to feedback the feature alignment network to generate target domain servo data that is closer to the source domain, thereby increasing the robustness of the servo fault detection model training and improving the fault detection accuracy of the servo fault detection model; when receiving the servo motion signal data collected by the edge intelligent chip carried by the target servo, the servo fault detection model can be used to accurately calculate the servo fault type corresponding to the target servo in real time, thereby improving the efficiency and accuracy of the servo fault detection. Therefore, the present invention proposes a method and device for real-time fault diagnosis of a steering gear based on unsupervised domain adaptation, which can solve the problems of poor efficiency and low accuracy of steering gear fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 A flowchart of a method for real-time fault diagnosis of a steering gear based on unsupervised domain adaptation provided by an embodiment of the present invention;

[0009] Figure 2 A functional module diagram of a steering gear intelligent fault diagnosis framework provided by an embodiment of the present invention;

[0010] Figure 3 Functional module diagram of a novel residual deformable convolution provided by an embodiment of the present invention

[0011] Figure 4 A functional module diagram of a residual adaptive feature extraction module provided by an embodiment of the present invention;

[0012] Figure 5 A functional module diagram of a novel deep deformable self-attention module provided by an embodiment of the present invention;

[0013] Figure 6 A functional module diagram of a novel multi-branch feature domain discriminator provided by an embodiment of the present invention;

[0014] Figure 7 A functional module diagram of a novel progressive robust fault diagnosis head provided by an embodiment of the present invention;

[0015] Figure 8A functional module diagram of a real-time fault diagnosis device for a steering gear based on unsupervised domain adaptation provided in accordance with an embodiment of the invention.

[0016] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0017] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0018] The embodiment of the present application provides a real-time fault diagnosis method for a servo based on unsupervised domain adaptation. The execution subject of a real-time fault diagnosis method for a servo based on unsupervised domain adaptation includes but is not limited to at least one of the electronic devices such as a server, a terminal, etc. that can be configured to execute the method provided in the embodiment of the present application. In other words, a real-time fault diagnosis method for a servo based on unsupervised domain adaptation can be executed by software or hardware installed on a terminal device or a server device. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc. The server can be an independent server, or it can be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0019] Reference Figure 1 As shown, it is a flow chart of a method for real-time fault diagnosis of a steering gear based on unsupervised domain adaptation provided by an embodiment of the present invention. The specific flow of the method is as follows:

[0020] S1. Obtain labeled source domain servo data and unlabeled target domain servo data, and use the pre-built servo intelligent fault diagnosis framework to extract data features of the source domain servo data and the target domain servo data respectively.

[0021] In one embodiment, the labeled source domain servo data is servo fault signal data with servo fault categories labeled, and the unlabeled target domain servo data is servo fault data of unknown fault categories in a complex dynamic environment. The unlabeled target domain servo data is trained with the labeled source domain servo data to generalize the training data, thereby improving the accuracy of model training.

[0022] In one embodiment, see Figure 2As shown in the figure, the intelligent fault diagnosis framework of the servo includes a new residual adaptive feature extractor (RAFE), a new global autocorrelation feature alignment network (GAFNet), a new multi-branch feature-domain discriminator (MFD), and a new progressive robust fault diagnosis head (PRFH).

[0023] Specifically, the new residual adaptive feature extractor adopts an adaptive soft threshold residual feature extraction method to effectively filter out a large amount of noise contained in the servo signal. At the same time, the residual connection method can retain the information contained in the original servo signal and solve the gradient vanishing problem, thereby efficiently extracting features from the servo input signal under a variety of complex dynamic working scenarios to better characterize the servo fault type. In addition, by reducing the size of the feature map and increasing the number of channels, high performance is maintained while reducing network redundancy and computational complexity.

[0024] Specifically, the new Global Autocorrelation Feature-Alignment Network (GAFNet) uses deep deformable self-attention to fully focus on the different dimensional features and contextual information of the input signal, and effectively models the long-distance dependency of the input signal features, so that the feature distribution of the target domain servo data in a complex dynamic environment gradually approaches the feature distribution of the labeled source domain servo data in a known environment, achieving robust feature alignment. It can effectively handle the problems of unknown distribution of servo fault data and complex mapping relationships in various complex dynamic working scenarios, so as to realize high-precision real-time intelligent diagnosis of servos.

[0025] Specifically, the new Multi-Branch Feature-Domain Discriminator (MFD) outputs labels through the data feature distribution of the source domain servo data and the target domain servo data after feature alignment, indicating whether the input data comes from the source domain servo data or the target domain servo data, thereby distinguishing the source domain servo data from the target domain servo data. It uses multiple pooling methods to model the signal feature distribution, so as to distinguish the source domain servo data from the target domain servo data as accurately as possible, and the goal of the feature alignment network is to generate target domain samples that are close enough to the source domain servo data distribution to deceive the discriminator. This game process drives and guides the feature alignment network to continuously improve the quality of generated samples, so as to learn from the discriminator feedback through back propagation, and gradually generate samples closer to the labeled source domain data distribution samples in a known environment from the unlabeled target domain data feature distribution samples in a complex dynamic unknown environment, so as to achieve high-precision real-time intelligent diagnosis training for servos.

[0026] Specifically, the new Progressive Robust Fault-Diagnosis Head (PRFH) efficiently enhances and integrates the servo input signals after feature alignment, gradually focuses on the fault type information contained in the servo signals through a multi-layer progressive network, and uses a skip connection structure to retain the original information contained in the input multi-level signal features, thereby effectively solving the problem of the difficulty in expressing the mapping relationship between the servo's own rotation signal and the working state. The integrated feature information is further mapped to the sample labeling space through a multi-layer perceptron, thereby efficiently and accurately outputting the servo fault type.

[0027] In one embodiment, the intelligent fault diagnosis framework of the servo includes a novel residual adaptive feature extractor (RAFE) for extracting data features of the labeled source domain servo data and the unlabeled target domain servo data, wherein the novel residual adaptive feature extractor is composed of a plurality of residual adaptive feature extraction modules (RAFM), each residual adaptive feature extraction module RAFM gradually focuses on the high-frequency information to the low-frequency information of the servo data, effectively focuses on the fault type contained in the servo signal, and provides robust servo signal features for the subsequent servo fault diagnosis process. At the same time, the residual adaptive feature extraction module RAFM gradually reduces the size of the feature map and gradually increases the number of channels, enriches the effective information contained in the data features, reduces the amount of calculation, and improves the processing efficiency.

[0028] In detail, the residual adaptive feature extraction module RAFM includes a new residual deformable convolution RDConv to integrate the characteristic information of the servo signal, and then uses the global maximum pooling (GAP), global average pooling (GAP) and activation function, fully connected layer (FClayer) to obtain the adaptive soft threshold (Adaptive Soft Threshold). The noise interference contained in the servo characteristics is filtered out by the threshold, and the residual structure is used to retain the characteristic information in the servo data. It can effectively solve the problem of gradient disappearance during the training process of the servo fault detection model and improve the accuracy of the servo fault detection model.

[0029] In one embodiment, a pre-built intelligent fault diagnosis framework for steering gear is used to extract data features of source domain steering gear data and target domain steering gear data, respectively, including: using the residual adaptive feature extraction module in the intelligent fault diagnosis framework for steering gear to perform residual deformable convolution on the source domain steering gear data and the target domain steering gear data to obtain residual features; performing global maximum pooling and global average pooling on the residual features to obtain maximum pooling and average pooling features; performing full connection and feature activation on the maximum pooling and average pooling features to obtain maximum pooling activation features and average pooling activation features. active features; calculate the adaptive soft thresholds corresponding to the source domain steering engine data and the target domain steering engine data according to the maximum pooling activation features and the average pooling activation features; perform feature screening on the source domain steering engine data and the target domain steering engine data according to the adaptive soft thresholds to obtain the source domain target features corresponding to the source domain steering engine data and the target domain target features corresponding to the target domain steering engine data; perform regularization operations on the source domain target features and the source domain steering engine data to obtain the data features of the source domain steering engine data, and perform regularization operations on the target domain target features and the target domain data to obtain the data features of the target domain steering engine data.

[0030] In one embodiment, the residual deformable convolution is a novel residual deformable convolution (RDConv) in the residual adaptive feature extraction module RAFM, ​​see Figure 3 As shown in the figure, RDConv introduces an offset for each point based on the ordinary one-dimensional convolution kernel, so that the convolution kernel is not a fixed shape, and the convolution area can always cover the important information of the servo signal features to improve the accuracy of feature extraction.

[0031] In one embodiment, the residual adaptive feature extraction module in the steering gear intelligent fault diagnosis framework is used to perform residual deformable convolution on the source domain steering gear data and the target domain steering gear data to obtain residual features, including:

[0032] The residual deformable convolution is performed on the source domain servo data and the target domain servo data using the following formula:

[0033]

[0034] in, represents the residual feature, Represents the source domain servo data and the target domain servo data, represents one-dimensional convolution, represents 1×1 convolution, Represents deformable convolution.

[0035] In one embodiment, the adaptive soft threshold is a parameter that is gradually updated during the training process of the servo fault detection model. It is a numerical value. Signal values ​​in the source domain servo data and the target domain servo data that are smaller than the adaptive soft threshold are regarded as noise and will be eliminated, and signal values ​​greater than the adaptive soft threshold are regarded as valid information and will be retained. Therefore, the noise in the source domain servo data and the target domain servo data can be removed.

[0036] In one embodiment, the data extraction process can be expressed by the following formula:

[0037]

[0038] in, Represents data characteristics, represents the residual adaptive feature extraction module, Indicates source domain servo data or target domain servo data.

[0039] In one embodiment, see Figure 4 As shown in the figure, the noise interference contained in the source domain servo data or the target domain servo data can be removed by the residual adaptive feature extraction module. At the same time, the original information in the source domain servo data or the target domain servo data is retained by using the residual structure, which can effectively solve the gradient disappearance problem in the training process of the servo fault detection model, thereby improving the accuracy of the servo fault detection model training. Figure 4 middle represents matrix multiplication, Represents element-by-element addition.

[0040] In one embodiment, the residual adaptive feature extraction module can be expressed as:

[0041]

[0042] in, Indicates source domain servo data or target domain servo data, represents residual deformable convolution, represents the global maximum pooling, express Activation function, represents the fully connected layer, express Activation function, represents global average pooling, express Activation function, represents the connection operation on the channel dimension, represents 1×1 convolution, Indicates feature screening operation based on adaptive soft threshold. represents the regularization operation, Represents data characteristics.

[0043] In one embodiment, the residual adaptive feature extraction module adopts an adaptive soft threshold residual feature extraction method, which can effectively filter out a large amount of noise contained in the servo data. At the same time, the residual connection method can retain the information contained in the original servo signal and solve the gradient disappearance problem, thereby efficiently extracting features from the servo input signal under various complex dynamic working scenarios to better characterize the servo fault type. In addition, by reducing the feature map size and increasing the number of channels, high performance is maintained while reducing network redundancy and computational complexity.

[0044] S2. Use the steering gear intelligent fault diagnosis framework to align and discriminate data features to obtain source domain features and target domain features.

[0045] In one embodiment, the distribution of data features varies greatly, and the mapping relationship between feature data and working states of different servos under different load torques is difficult to express quantitatively. Therefore, in order to make the data feature distribution of target domain servo data in a complex dynamic environment gradually close to the feature distribution of source domain servo data with annotations in a known environment and realize robust feature alignment, the data features are aligned through a new global autocorrelation feature alignment network (GAFNet) in the servo intelligent fault diagnosis framework.

[0046] Specifically, GAFNet is mainly composed of a novel residual deformable convolution RDConv, a deep deformable self-attention module (DDSA) and a one-dimensional convolution. It first enhances the servo signal features through the novel residual deformable convolution RDConv, then uses deep deformable self-attention DDSA to fully focus on the fault type information contained in the servo signal features, and finally uses one-dimensional convolution to integrate the robust servo signal features to generate data samples with the same data feature distribution as the labeled source domain servo data in the known environment.

[0047] In one embodiment, feature discrimination is to determine whether the data after feature alignment is derived from source domain servo data or target domain servo data. The present invention proposes a novel multi-branch feature domain discriminator MFD, which is used to guide the feature alignment network to continuously improve the quality of generated samples, thereby learning from the discriminator feedback through back propagation, and gradually generating distribution samples closer to the labeled source domain servo data in a known environment from the feature distribution of the target domain data in a complex dynamic environment, so as to realize high-precision real-time intelligent diagnosis training of servo faults.

[0048] In one embodiment, a steering gear intelligent fault diagnosis framework is used to perform feature alignment and feature discrimination on data features to obtain discrimination source domain features and discrimination target domain features, including: using a new global autocorrelation feature alignment network in the steering gear intelligent fault diagnosis framework to perform residual convolution and self-attention calculation on data features to obtain alignment features; using a new multi-branch feature domain discriminator in the steering gear intelligent fault diagnosis framework to perform feature discrimination on the alignment features to obtain discrimination source domain features and discrimination target domain features.

[0049] In one embodiment, the novel global autocorrelation feature alignment network in the steering gear intelligent fault diagnosis framework is used to perform residual convolution and self-attention calculation on the data features to obtain the alignment features. Including:

[0050] The following formula is used to perform residual convolution and self-attention calculation on the data features:

[0051]

[0052] in, represents the alignment feature, Represents data characteristics, represents residual deformable convolution, represents one-dimensional convolution, represents the preset new deep deformable self-attention module, Represents an alignment feature.

[0053] In one embodiment, see Figure 5 As shown in the figure, in order to pay targeted attention to the irregular fault type information contained in the data features of the servo data in a complex dynamic unknown environment, the new deep deformable self-attention module (Deep Deformable Self-Attention, DDSA) first sets the learnable offset through the new residual deformable convolution RDConv, and then uses the fully connected layer to obtain the attention matrix QKV required for attention calculation, and at the same time increases the context position encoding (Content-Position) by setting the length dimension learnable parameters and channel dimension learnable parameters The different information positions contained in the servo input features are located, and finally the data features of the robust servo data are effectively modeled through operations such as Softmax normalization and Feed Forward Network (FFN), which can solve the problems of non-Gaussian noise interference and difficulty in analyzing the servo fault categories. Figure 5 middle represents element-by-element addition, Represents matrix multiplication.

[0054] In one embodiment, the novel deep deformable self-attention module can be expressed as:

[0055]

[0056] in, represents the input of the novel deep deformable self-attention module, represents residual deformable convolution, represents the preset length dimension learnable parameter, Indicates the preset channel dimension learnable parameters, express Normalization operation of the activation function, represents the regularization operation, represents a pre-built feed-forward neural network, Represents the output of the novel deep deformable self-attention module.

[0057] In one embodiment, the feature alignment network uses deep deformable self-attention to fully focus on the different dimensional features and contextual information of the input signal, and effectively models the long-distance dependency of the input features, so that the feature distribution of the target domain data in a complex dynamic environment is gradually close to the feature distribution of the labeled source domain servo data in a known environment, thereby realizing robust feature alignment. Through the feature alignment network, the data feature distribution of the target domain servo data in a complex dynamic environment can be gradually close to the data feature distribution of the labeled source domain data in a known environment, thereby realizing robust feature alignment, thereby effectively handling the problems of unknown distribution of servo fault data and complex mapping relationships in various complex dynamic working scenarios, thereby improving the accuracy of servo fault diagnosis.

[0058] In one embodiment, a novel multi-branch feature domain discriminator MFD is used to guide the feature alignment network to continuously improve the quality of generated samples, thereby learning from the discriminator feedback through back propagation, and gradually making the data feature distribution of the target domain servo data in a complex dynamic environment closer to the data feature distribution samples of the labeled source domain servo data in a known environment, so as to realize high-precision real-time intelligent diagnosis training of servo, see Figure 6As shown in the figure, MFD adopts three pooling methods: maximum pooling, average pooling, and soft pooling. It uses a multi-branch structure to focus on different levels of information of the aligned features after feature alignment, so as to accurately determine whether the data source is a labeled source domain steering engine data distribution sample or an unlabeled target domain steering engine data distribution sample in a complex dynamic unknown environment. Figure 6 middle represents element-wise addition.

[0059] Specifically, the new multi-branch feature domain discriminator can be expressed as:

[0060]

[0061] in, represents the alignment feature, represents residual deformable convolution, represents a linear layer, represents one-dimensional maximum pooling, represents one-dimensional average pooling, represents one-dimensional soft pooling, represents the regularization operation, represents the fully connected layer, express Activation function, Represents discriminative source domain features or discriminative target domain features.

[0062] In one embodiment, the discriminant domain feature is that the new multi-branch feature domain discriminator discriminates the source of the alignment feature from the labeled source domain servo engine data, and the discriminant target domain feature is that the new multi-branch feature domain discriminator discriminates the source of the alignment feature from the unlabeled target domain servo engine data. The discriminator distinguishes the source domain servo engine data and the target domain servo engine data as accurately as possible, and the feedback feature alignment network is used to continuously improve the quality of the generated samples, thereby generating alignment features that are closer to the labeled source domain servo engine data, thereby improving the accuracy of the servo engine fault detection model training.

[0063] S3, calculating the fault diagnosis loss value between the source domain features and the target domain features and the source domain steering engine data and the target domain steering engine data.

[0064] In one embodiment, the fault diagnosis loss value includes the loss value of the source domain servo data in the servo intelligent fault diagnosis framework, the loss value of the new multi-branch feature domain discriminator, and the adversarial loss value of discriminating the source domain features and the target domain by calculating the new global autocorrelation feature alignment network, wherein the aligned features of the source domain are predicted by the new progressive robust fault diagnosis head to obtain the predicted fault type, and then the loss value between the predicted fault type and the annotation of the source domain servo data is calculated.

[0065] In one embodiment, before calculating the fault diagnosis loss value between distinguishing source domain features and distinguishing target domain features and source domain servo data and target domain servo data, the method also includes: obtaining alignment features of source domain servo data, using a novel progressive robust fault diagnosis head in a servo intelligent fault diagnosis framework to perform fault prediction on the alignment features to obtain a predicted servo fault category; and calculating the source domain loss value between the predicted servo fault and the labeled fault category of the source domain servo data.

[0066] In one embodiment, the novel progressive robust fault diagnosis head PRFH gradually models the fault type information contained in the steering gear input signal characteristics through a multi-layer feature output module, thereby realizing robust steering gear fault diagnosis. Figure 7 As shown in the figure, PRFH contains three layers of feature output modules, each of which is composed of a new residual deformable convolution RDConv, a PReLU activation function, and a fully connected layer. In addition, the original information contained in the input multi-level signal features is retained through a jump connection structure, thereby effectively solving the problem of the difficulty in expressing the mapping relationship between the steering gear's own rotation signal and the working state, and further mapping the integrated feature information to the sample label space through a multi-layer perceptron, thereby efficiently and accurately outputting the steering gear fault type.

[0067] Specifically, the new progressive robust fault diagnosis head can be expressed as:

[0068]

[0069] in, represents the alignment features of the source domain servo data, represents residual deformable convolution, express Activation function, represents the fully connected layer, represents the connection operation on the channel dimension, represents 1×1 convolution, represents a pre-built multi-layer perceptron, Predict the type of servo failure.

[0070] In one embodiment, for the source domain servo data, since the source domain servo data is labeled, the mean cross entropy loss function is used to calculate the difference between the final prediction result and the true value, and the accurate servo fault diagnosis result is obtained by minimizing the loss function. In addition, for the labeled source domain servo data and the unlabeled target domain servo data, the adversarial training process of the feature alignment network and the discriminator is adopted, that is, the feature alignment network maps each pair of source domain servo data and target domain servo data to a specific feature space to extract the specific features of the source domain and the target domain, and aligns the feature distribution of the specific domain by deep domain adaptation to learn the invariant representation of the source domain and the target domain. The feature alignment network continuously improves the quality of the generated samples under the guidance of the adaptive loss in the adversarial training field, while the discriminator continuously improves the ability to identify the source of the data, and finally obtains a feature alignment network that can robustly process the source domain and the target domain data.

[0071] In one embodiment, calculating the source domain loss value between the predicted steering engine fault and the labeled fault category of the source domain steering engine data includes:

[0072] The source domain loss value is calculated using the following formula:

[0073]

[0074] in, represents the source domain loss value, Indicates the total number of data in the source domain servo data, represents the cross entropy loss function, Indicates The predicted servo fault category corresponding to the aligned features of the source domain servo data, Indicates The labeled fault categories corresponding to the aligned features of the source domain servo data.

[0075] In one embodiment, in the adversarial learning between the feature alignment network and the discriminator, a least squares loss function is introduced to train the feature alignment network, aiming to generate source domain servo data class features from the target domain servo data to deceive the discriminator.

[0076] In one embodiment, the fault diagnosis loss value between discriminating source domain features and discriminating target domain features and source domain servo data and target domain servo data is calculated, including: extracting target domain discriminant features corresponding to target domain servo data from discriminating source domain features and discriminating target domain features; calculating the domain adaptive loss value corresponding to the target domain servo data using a preset first loss function according to the target domain discriminant features; calculating the discrimination loss value of the source domain servo data and the target domain servo data using a preset second loss function according to the discriminating source domain features and discriminating target domain features; obtaining the source domain loss value corresponding to the source domain servo data, and performing weighted calculation on the source domain loss value and the domain adaptive loss value using a preset third loss function to obtain an inference loss value; and aggregating the inference loss value and the discriminant loss value to obtain a fault diagnosis loss value.

[0077] In one embodiment, the loss value of the feature discrimination network is calculated by using the discriminant features of the target domain corresponding to the target domain servo data that are discriminated as source domain features, thereby improving the accuracy of the feature alignment network.

[0078] In one embodiment, the domain adaptation loss value corresponding to the target domain servo data is calculated using a preset first loss function according to the target domain discrimination feature, including:

[0079] The first loss function is expressed as:

[0080]

[0081] in, represents the domain adaptation loss value, represents the discriminative features of the target domain, Represents the discriminative source domain features in the target domain discriminative features.

[0082] In one embodiment, the target domain discriminant features include the discriminant source domain features that are discriminated as source domain servo data by the near discriminator and the discriminant target domain features that are discriminated as target domain, and then the discriminator can be trained through the domain adaptive loss value so that the feature distribution of the target domain servo data is closer to the source domain servo data.

[0083] In one embodiment, the discrimination loss values ​​of the source domain servo data and the target domain servo data are calculated by using a preset second loss function according to the discrimination source domain features and the discrimination target domain features, including:

[0084] The second loss function is expressed as:

[0085]

[0086] in, represents the discriminant loss value, Respectively represent the source domain servo data and the target domain servo data, Represents the discriminative source domain features and the discriminative target domain features, Indicates the feature label corresponding to the source domain servo data or the target domain servo data.

[0087] In one embodiment, the feature label indicates the data source of the source domain servo data or the target domain servo data. For example, if the feature label is source domain servo data, it means that the data is data in the source domain servo; if the feature label is target domain servo data, it means that the data is data in the target domain servo. The parameters in the discriminator are adjusted by discriminating the loss value so that the discriminator can more accurately identify the source of the input data, thereby feeding back the feature alignment network to generate target domain servo data that is closer to the source domain, thereby increasing the robustness of the servo fault detection model training and improving the fault detection accuracy of the servo fault detection model.

[0088] In one embodiment, in the process of fault detection of the actual steering gear fault detection model, only the feature alignment network is used, and the discriminator is only used in the model training stage. Therefore, it is necessary to perform weighted calculation on the source domain loss value and the domain adaptation loss value to obtain the inference loss value.

[0089] In one embodiment, a preset third loss function is used to perform weighted calculation on the source domain loss value and the domain adaptation loss value to obtain an inference loss value, including:

[0090] The third loss function is expressed as:

[0091]

[0092] in, represents the inference loss value, represents the source domain loss value, represents the domain adaptation loss value, , They represent the preset weight coefficients of the source domain loss value and the domain adaptation loss value respectively.

[0093] In one embodiment, the fault diagnosis loss value can be used to guide the intelligent fault diagnosis framework of the steering gear to adjust parameters, so as to obtain a steering gear fault detection model with more accurate fault detection.

[0094] S4. Adjust parameters of the steering gear intelligent fault diagnosis framework according to the fault diagnosis loss value until the fault diagnosis loss value is less than a preset loss value threshold, thereby obtaining a steering gear fault detection model corresponding to the steering gear intelligent fault diagnosis framework.

[0095] In one embodiment, the parameters of the new residual adaptive feature extractor and the new progressive robust fault diagnosis head in the steering gear intelligent fault diagnosis framework are modified by back-propagating the source domain loss value gradient. For example, the offset in the new residual deformable convolution, the length dimension learnable parameters and other parameters are adjusted until the fault source domain loss value is less than a preset loss value threshold.

[0096] In one embodiment, the feature alignment network and the discriminator in the steering gear intelligent fault diagnosis framework can be adversarially trained through the inference loss value in the fault diagnosis loss value, that is, the feature alignment network continuously improves the quality of generated samples under the guidance of the adaptive loss value in the adversarial training field, while the discriminator continuously improves the ability to identify the source of data, and finally a feature alignment network that can robustly process the source domain steering gear data and the target domain steering gear data is obtained.

[0097] In one embodiment, the gradient back propagation method can be used to adjust parameters until each loss value in the fault diagnosis loss value is less than a preset loss value threshold, thereby obtaining a trained steering gear intelligent fault diagnosis framework.

[0098] In one embodiment, the discriminator in the steering gear intelligent fault diagnosis framework is only used for training the steering gear fault detection model. Therefore, the steering gear fault detection model is obtained through the new residual adaptive feature extractor, the new global autocorrelation feature alignment network, and the new progressive robust fault diagnosis head after parameter update in the steering gear intelligent fault diagnosis framework.

[0099] S5. When receiving the servo motion signal data collected by the edge smart chip carried by the target servo, the servo fault detection model is used to calculate the servo fault type corresponding to the target servo.

[0100] In one embodiment, the edge smart chip can be an intelligent chip deployed on the industrial-grade servo end based on a high-speed, low-latency, and wide-connection embedded NPU (Neural-Network Processing Unit) platform. The edge smart chip deploys the above-mentioned trained servo fault detection model, and collects servo motion signal data under various complex dynamic scenarios through the edge smart chip mounted on the servo body, and independently implements edge intelligent processing in the chip, that is, based on the above-mentioned new servo fault diagnosis model, high-precision, strong robustness, and real-time servo fault diagnosis results are obtained at the servo motion signal data acquisition end.

[0101] In one embodiment, when receiving the servo motion signal data collected by the edge intelligent chip carried on the target servo, the servo fault type corresponding to the target servo is calculated using the servo fault detection model, including: using the new residual adaptive feature extractor in the servo fault detection model to extract the servo signal features in the servo motion signal data; using the new global autocorrelation feature alignment network in the servo fault detection model to align the servo signal features to obtain aligned signal features; using the new progressive robust fault diagnosis head in the servo fault detection model to classify the aligned signal features into fault categories to obtain the servo fault type corresponding to the target servo.

[0102] In one embodiment, the novel residual adaptive feature extractor, the novel global autocorrelation feature alignment network, and the novel progressive robust fault diagnosis head in the servo fault detection model are the servo intelligent fault diagnosis framework after the parameter update in the above S4 is completed, thereby obtaining a servo fault detection model with higher fault diagnosis accuracy.

[0103] In one embodiment, the diagnostic results obtained can be used to serve communication data optimization, real-time response operations, industrial fault warning, intelligent analysis applications, privacy protection and data security through the edge intelligent chip carried by the target servo. Ensure that the number and volume of communications with the cloud platform are reduced as much as possible, thereby reducing waiting time and computing costs. Since the intelligent diagnosis tasks of the servo in various complex dynamic scenarios are deployed to the servo motion signal data acquisition end, it can effectively reduce the congestion of the backbone network, alleviate bandwidth occupancy, achieve lower latency, improve processing efficiency, speed up response requests, and further improve the accuracy of intelligent fault diagnosis of servo in various complex dynamic scenarios. In particular, the speed after the deployment of the edge intelligent chip can reach the real-time processing speed, while improving the accuracy of servo fault diagnosis.

[0104] The embodiment of the present invention can effectively filter a large amount of noise contained in the servo data by extracting data features of the source domain servo data and the target domain servo data, while retaining the information contained in the original servo signal, solving the gradient vanishing problem, and improving the accuracy of feature extraction; aligning and distinguishing the data features can continuously improve the quality of generated samples, thereby generating alignment features that are closer to the labeled source domain servo data, and improving the accuracy of the servo fault detection model training; calculating the fault diagnosis loss value between the distinguished source domain features and the distinguished target domain features and the source domain servo data and the target domain servo data, and adjusting the parameters of the servo intelligent fault diagnosis framework through the fault diagnosis loss value, so as to feedback the feature alignment network to generate target domain servo data that is closer to the source domain, thereby increasing the robustness of the servo fault detection model training and improving the fault detection accuracy of the servo fault detection model; when receiving the servo motion signal data collected by the edge intelligent chip carried by the target servo, the servo fault detection model can be used to accurately calculate the servo fault type corresponding to the target servo in real time, thereby achieving the efficiency and accuracy of the servo fault detection. Therefore, the present invention proposes a method for real-time fault diagnosis of a steering gear based on unsupervised domain adaptation, which can solve the problems of poor efficiency and low accuracy of steering gear fault diagnosis.

[0105] like Figure 8 , is a functional module diagram of a steering gear real-time fault diagnosis device based on unsupervised domain adaptation provided by an embodiment of the present invention.

[0106] The present invention provides a real-time fault diagnosis device 800 for a steering gear based on unsupervised domain adaptation, which can be installed in an electronic device. According to the functions to be implemented, the real-time fault diagnosis device 800 for a steering gear based on unsupervised domain adaptation in this embodiment can include: a data feature extraction module 801, a feature alignment and feature discrimination module 802, a fault diagnosis loss value calculation module 803, a steering gear fault detection model determination module 804, and a steering gear fault type diagnosis module 805. The module of the present invention can also be referred to as a unit, which refers to a series of computer program segments that can be executed by a processor of an electronic device and can complete fixed functions, and is stored in a memory of an electronic device.

[0107] In this embodiment, the functions of each module / unit are as follows: a data feature extraction module 801 is used to obtain labeled source domain steering engine data and unlabeled target domain steering engine data, and use a pre-built steering engine intelligent fault diagnosis framework to extract data features of the source domain steering engine data and the target domain steering engine data respectively; a feature alignment and feature discrimination module 802 is used to use the steering engine intelligent fault diagnosis framework to perform feature alignment and feature discrimination on data features to obtain source domain features and target domain features; a fault diagnosis loss value calculation module 803 is used to calculate the source domain features and the target domain features. The fault diagnosis loss value between the target domain features and the source domain servo data and the target domain servo data is identified; the servo fault detection model determination module 804 is used to adjust the parameters of the servo intelligent fault diagnosis framework according to the fault diagnosis loss value, until the fault diagnosis loss value is less than the preset loss value threshold, and the servo fault detection model corresponding to the servo intelligent fault diagnosis framework is obtained; the servo fault type diagnosis module 805 is used to calculate the servo fault type corresponding to the target servo using the servo fault detection model when receiving the servo motion signal data collected by the edge intelligent chip carried by the target servo.

[0108] In detail, in one embodiment, each module in a real-time fault diagnosis device 700 for a steering gear based on unsupervised domain adaptation adopts the same technical means as the real-time fault diagnosis method for a steering gear based on unsupervised domain adaptation in the accompanying drawings when used, and can produce the same technical effects, which will not be repeated here.

[0109] For example, although not shown, the electronic device may also include a power source (such as a battery) for supplying power to various components. Preferably, the power source may be logically connected to at least one processor 401 through a power management device, so that the power management device can realize functions such as charging management, discharging management, and power consumption management. The power source may also include any components such as one or more DC or AC power sources, recharging devices, power failure detection circuits, power converters or inverters, and power status indicators. The electronic device may also include a variety of sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be described in detail here.

[0110] It should be understood that the embodiment is for illustration only and the scope of the patent application is not limited by this structure.

[0111] The present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it can implement a real-time fault diagnosis method for a steering engine based on unsupervised domain adaptation according to any of the above embodiments.

[0112] It should be noted that the computer-readable storage medium may be volatile or non-volatile. For example, the computer-readable medium may include: any entity or device capable of carrying computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM). In the several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses and methods may be implemented in other ways. For example, the device embodiments described above are merely schematic, for example, the division of modules is only a logical function division, and there may be other division methods in actual implementation.

[0113] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0114] In addition, each functional module in each embodiment of the present invention may be integrated into one processing unit, each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0115] It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0116] Therefore, no matter from which point of view, the embodiments should be regarded as illustrative and non-restrictive, and the scope of the present invention is limited by the appended claims rather than the above description, so it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention. Any attached figure mark in the claims should not be regarded as limiting the claims involved.

[0117] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0118] In addition, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices stated in the system claim can also be implemented by one unit or device through software or hardware. The words first, second, etc. are used to indicate names, and do not indicate any particular order.

[0119] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A real-time fault diagnosis method for a steering gear based on unsupervised domain adaptation, characterized in that: The method comprises: Obtain labeled source domain steering engine data and unlabeled target domain steering engine data, and use a pre-built steering engine intelligent fault diagnosis framework to extract data features of the source domain steering engine data and the target domain steering engine data respectively; Using the steering gear intelligent fault diagnosis framework to perform feature alignment and feature discrimination on the data features, and obtain discrimination source domain features and discrimination target domain features; Calculating the fault diagnosis loss value between the discrimination source domain feature and the discrimination target domain feature and the source domain steering engine data and the target domain steering engine data; According to the fault diagnosis loss value, the parameters of the steering gear intelligent fault diagnosis framework are adjusted until the fault diagnosis loss value is less than a preset loss value threshold, thereby obtaining a steering gear fault detection model corresponding to the steering gear intelligent fault diagnosis framework; When receiving the steering gear motion signal data collected by the edge intelligent chip carried by the target steering gear, the steering gear fault detection model is used to calculate the steering gear fault type corresponding to the target steering gear; The method of using the pre-built steering gear intelligent fault diagnosis framework to extract data features of the source domain steering gear data and the target domain steering gear data includes: Using the residual adaptive feature extraction module in the steering gear intelligent fault diagnosis framework, performing residual deformable convolution on the source domain steering gear data and the target domain steering gear data to obtain residual features; Performing global maximum pooling and global average pooling on the residual features respectively to obtain maximum pooling and average pooling features; Performing full connection and feature activation on the maximum pooling and the average pooling features to obtain maximum pooling activation features and average pooling activation features; Calculating the adaptive soft thresholds corresponding to the source domain steering engine data and the target domain steering engine data according to the maximum pooling activation feature and the average pooling activation feature; Performing feature screening on the source domain steering engine data and the target domain steering engine data according to the adaptive soft threshold to obtain source domain target features corresponding to the source domain steering engine data and target domain target features corresponding to the target domain steering engine data; Regularization operation is performed on the source domain target feature and the source domain steering engine data to obtain data features of the source domain steering engine data, and regularization operation is performed on the target domain target feature and the target domain steering engine data to obtain data features of the target domain steering engine data.

2. A method for real-time fault diagnosis of a steering gear based on unsupervised domain adaptation as claimed in claim 1, characterized in that: The method of using the residual adaptive feature extraction module in the steering gear intelligent fault diagnosis framework to perform residual deformable convolution on the source domain steering gear data and the target domain steering gear data to obtain residual features includes: The residual deformable convolution is performed on the source domain servo data and the target domain servo data using the following formula: in, represents the residual feature, represents the source domain steering engine data and the target domain steering engine data, represents one-dimensional convolution, represents 1×1 convolution, Represents deformable convolution.

3. A method for real-time fault diagnosis of a steering gear based on unsupervised domain adaptation as claimed in claim 1, characterized in that: The residual adaptive feature extraction module is expressed as: in, Indicates source domain servo data or target domain servo data, represents residual deformable convolution, represents the global maximum pooling, express Activation function, represents the fully connected layer, express Activation function, represents global average pooling, express Activation function, represents the connection operation on the channel dimension, represents 1×1 convolution, Indicates feature screening operation based on adaptive soft threshold. represents the regularization operation, Represents data characteristics.

4. A method for real-time fault diagnosis of a steering gear based on unsupervised domain adaptation as claimed in claim 1, characterized in that: The method of using the steering gear intelligent fault diagnosis framework to perform feature alignment and feature discrimination on the data features to obtain discrimination source domain features and discrimination target domain features includes: Using the global autocorrelation feature alignment network in the steering gear intelligent fault diagnosis framework to perform residual convolution and self-attention calculation on the data features, to obtain alignment features; The multi-branch feature domain discriminator in the steering gear intelligent fault diagnosis framework is used to perform feature discrimination on the alignment features to obtain discrimination source domain features and discrimination target domain features.

5. A method for real-time fault diagnosis of a steering gear based on unsupervised domain adaptation as claimed in claim 4, characterized in that: The method of using the global autocorrelation feature alignment network in the steering gear intelligent fault diagnosis framework to perform residual convolution and self-attention calculation on the data features to obtain alignment features includes: The following formula is used to perform residual convolution and self-attention calculation on the data features: in, represents the alignment feature, Represents data characteristics, represents residual deformable convolution, represents one-dimensional convolution, represents the preset deep deformable self-attention module, Represents an alignment feature.

6. A method for real-time fault diagnosis of a steering gear based on unsupervised domain adaptation as claimed in claim 4, characterized in that: The deep deformable self-attention module is expressed as: in, represents the input of the deep deformable self-attention module, represents residual deformable convolution, represents the preset length dimension learnable parameter, Indicates the preset channel dimension learnable parameters, express Normalization operation of the activation function, represents the regularization operation, represents a pre-built feed-forward neural network, represents the output of the deep deformable self-attention module.

7. A method for real-time fault diagnosis of a steering gear based on unsupervised domain adaptation as claimed in claim 4, characterized in that: The multi-branch feature domain discriminator is expressed as: in, represents the alignment feature, represents residual deformable convolution, represents a linear layer, represents one-dimensional maximum pooling, represents one-dimensional average pooling, represents one-dimensional soft pooling, represents the regularization operation, represents the fully connected layer, express Activation function, Represents discriminative source domain features or discriminative target domain features.

8. The method for real-time fault diagnosis of a steering gear based on unsupervised domain adaptation as claimed in claim 1, characterized in that: Before calculating the fault diagnosis loss value between the discrimination source domain feature and the discrimination target domain feature and the source domain steering engine data and the target domain steering engine data, the method further includes: Acquire the alignment features of the source domain steering gear data, and use the progressive robust fault diagnosis head in the steering gear intelligent fault diagnosis framework to perform fault prediction on the alignment features to obtain a predicted steering gear fault category; Calculating a source domain loss value between the predicted steering engine fault and the labeled fault category of the source domain steering engine data; The source domain loss value is calculated using the following formula: in, represents the source domain loss value, Indicates the total number of data in the source domain servo data, represents the cross entropy loss function, Indicates The predicted servo fault category corresponding to the aligned features of the source domain servo data, Indicates The labeled fault categories corresponding to the aligned features of the source domain servo data.

9. A method for real-time fault diagnosis of a steering gear based on unsupervised domain adaptation as claimed in claim 8, characterized in that: The progressive robust fault diagnosis head is expressed as: in, represents the alignment features of the source domain servo data, represents residual deformable convolution, express Activation function, represents the fully connected layer, represents the connection operation on the channel dimension, represents 1×1 convolution, represents a pre-built multi-layer perceptron, Predict the type of servo failure.

10. The method for real-time fault diagnosis of a steering gear based on unsupervised domain adaptation according to claim 1, characterized in that: The calculating of the fault diagnosis loss value between the distinguishing source domain feature and the distinguishing target domain feature and the source domain steering engine data and the target domain steering engine data includes: Extracting target domain discriminant features corresponding to the target domain servo data from the discriminant source domain features and the discriminant target domain features; Calculate the domain adaptation loss value corresponding to the target domain servo data using a preset first loss function according to the target domain discrimination feature; Calculate the discrimination loss value of the source domain servo data and the target domain servo data using a preset second loss function according to the discrimination source domain feature and the discrimination target domain feature; Obtaining a source domain loss value corresponding to the source domain servo data, and performing weighted calculation on the source domain loss value and the domain adaptive loss value using a preset third loss function to obtain an inference loss value; The inference loss value and the discrimination loss value are collected to obtain a fault diagnosis loss value.

11. A method for real-time fault diagnosis of a steering gear based on unsupervised domain adaptation as claimed in claim 10, characterized in that: The first loss function is expressed as: in, represents the domain adaptation loss value, represents the target domain discriminant feature, represents the discriminative source domain feature in the target domain discriminative feature; The second loss function is expressed as: in, represents the discriminant loss value, represent the source domain steering engine data and the target domain steering engine data respectively, represents the discriminative source domain feature and the discriminative target domain feature, Indicates a feature label corresponding to the source domain steering engine data or the target domain steering engine data; The third loss function is expressed as: in, represents the inference loss value, represents the source domain loss value, represents the domain adaptive loss value, , They respectively represent the preset weight coefficients of the source domain loss value and the domain adaptation loss value.

12. A method for real-time fault diagnosis of a steering gear based on unsupervised domain adaptation as claimed in claim 1, characterized in that: When receiving the steering gear motion signal data collected by the edge smart chip carried by the target steering gear, using the steering gear fault detection model to calculate the steering gear fault type corresponding to the target steering gear, including: Extracting the steering gear signal features in the steering gear motion signal data using the residual adaptive feature extractor in the steering gear fault detection model; Using the global autocorrelation feature alignment network in the steering gear fault detection model to perform feature alignment on the steering gear signal features to obtain aligned signal features; The progressive robust fault diagnosis head in the steering gear fault detection model is used to classify the alignment signal features into fault categories to obtain the steering gear fault type corresponding to the target steering gear.

13. A real-time fault diagnosis device for a steering gear based on unsupervised domain adaptation, characterized in that: The device comprises: The data feature extraction module is used to obtain labeled source domain steering gear data and unlabeled target domain steering gear data, and use the residual adaptive feature extraction module in the steering gear intelligent fault diagnosis framework to perform residual deformable convolution on the source domain steering gear data and the target domain steering gear data to obtain residual features; perform global maximum pooling and global average pooling on the residual features to obtain maximum pooling and average pooling features; perform full connection and feature activation on the maximum pooling and average pooling features to obtain maximum pooling activation features and average pooling activation features; and perform feature extraction based on the maximum pooling activation features and The average pooling activation feature calculates the adaptive soft threshold corresponding to the source domain steering engine data and the target domain steering engine data; the source domain steering engine data and the target domain steering engine data are feature screened according to the adaptive soft threshold to obtain the source domain target feature corresponding to the source domain steering engine data and the target domain target feature corresponding to the target domain steering engine data; the source domain target feature and the source domain steering engine data are regularized to obtain the data feature of the source domain steering engine data, and the target domain target feature and the target domain steering engine data are regularized to obtain the data feature of the target domain steering engine data; A feature alignment and feature discrimination module is used to perform feature alignment and feature discrimination on the data features using the steering gear intelligent fault diagnosis framework to obtain discrimination source domain features and discrimination target domain features; A fault diagnosis loss value calculation module, used to calculate the fault diagnosis loss value between the discrimination source domain feature and the discrimination target domain feature and the source domain steering engine data and the target domain steering engine data; A steering gear fault detection model determination module is used to adjust parameters of the steering gear intelligent fault diagnosis framework according to the fault diagnosis loss value until the fault diagnosis loss value is less than a preset loss value threshold, thereby obtaining a steering gear fault detection model corresponding to the steering gear intelligent fault diagnosis framework; The servo fault type diagnosis module is used to calculate the servo fault type corresponding to the target servo using the servo fault detection model when receiving the servo motion signal data collected by the edge intelligent chip carried by the target servo.

Images