A device fault diagnosis network model training method based on multi-source domain knowledge joint transfer

Abstract

The present application relates to the technical field of computer deep learning, and relates to a device fault diagnosis network model training method based on multi-source domain knowledge joint migration, which comprises the following steps: constructing a training sample set, a verification set and a test set; training a pre-constructed device fault diagnosis network model by using the training sample set, optimizing the parameters of the device fault diagnosis network model, and obtaining a trained device fault diagnosis network model; verifying the trained device fault diagnosis network model by using the verification set, evaluating the diagnostic performance of the trained device fault diagnosis network model, and selecting an optimal device fault diagnosis network model; and testing the optimal device fault diagnosis network model by using the test set, and evaluating the performance of the optimal device fault diagnosis network model. The present application uses the multi-source domain collaborative training of the model by using the collected multi-source domain data set containing a small amount of labels composed of a plurality of labeled source domain data sets and unlabeled source domain data sets, learns the field invariant features in the fault samples, and migrates the shared diagnostic knowledge from the labeled data and the unlabeled data to the target field, so that the trained device fault diagnosis network model can complete the cross-domain fault diagnosis task without needing to obtain the target samples in advance.

Description

Technical Field

[0001] This application belongs to the field of computer deep learning technology, specifically involving a training method for a device fault diagnosis network model based on multi-source domain knowledge joint transfer. Background Technology

[0002] Rolling bearings are critical components in twin-rotor aero-engine systems. They typically operate in high-speed, noisy environments, making them prone to failure, which can affect engine operation and even cause serious accidents. Conducting fault diagnosis research on aero-engine bearings is crucial for accurately diagnosing rolling bearing fault characteristics and significantly improves the safety and reliability of aero-engines.

[0003] In recent years, research on data-driven bearing fault diagnosis methods has developed rapidly. Various deep learning models, such as convolutional neural networks, graph convolutional networks, and long short-term memory networks, have been used for equipment fault diagnosis, achieving high performance. However, most methods rely on large amounts of labeled data, assuming that training and testing data are independent and identically distributed, and that samples in the target domain exist. However, in real-world industrial applications, obtaining sufficient labeled data is extremely difficult. Equipment operating conditions are often uncertain, making it difficult to obtain samples and fault label information under target domain conditions. Furthermore, training data is typically collected from multiple operating conditions, often resulting in a different distribution between the training data (source domain) and the testing data (target domain), exhibiting significant domain shift. This greatly reduces the performance and effectiveness of traditional fault diagnosis models, making it difficult to obtain the expected results. In actual operation, the operating conditions of aero-engines change, leading to inconsistent data sample distributions. Furthermore, using only data from a single source domain to train a fault diagnosis model may not achieve ideal diagnostic performance; however, comprehensively collecting and labeling sufficient training samples to make the fault diagnosis model robust to all operating conditions is unrealistic.

[0004] Therefore, cross-domain fault diagnosis of equipment has significant practical implications, and the ability to learn and classify domain-invariant features is crucial for constructing cross-domain fault diagnosis models. To address the challenges posed by the unknown domain in cross-domain fault diagnosis tasks, domain adaptation methods and domain generalization-based fault diagnosis methods are typically employed to mitigate performance degradation issues when dealing with data from unknown domains.

[0005] When target domain data is available, domain adaptation methods can address the domain offset problem by reducing the data distribution difference between the source and target domains. Liu et al. proposed a deep adversarial domain adaptation model for rolling bearing fault diagnosis to address the inconsistency in the distributions of the source and target domains. Xiao et al. designed an improved loss function embedding the joint maximum mean difference to achieve simultaneous alignment of marginal and conditional distributions in unsupervised cross-domain scenarios. However, single-source domain-based transfer learning strategies may not achieve ideal generalization performance due to the single source of knowledge transfer. By collecting multi-source domain data from different scenarios and employing multi-source domain-based transfer learning techniques, the generalization performance of diagnostic models can be further improved. Therefore, Gao et al. proposed a multi-source domain information fusion network that combines adversarial transfer learning techniques with fine-grained information fusion from multiple source domains to achieve feature learning and knowledge transfer between the source and target domains; Tian et al. used a multi-branch network structure to match the feature space distribution of each source and target domain separately, transferring multi-source domain diagnostic knowledge to cross-domain fault diagnosis tasks. Domain adaptation methods utilize diagnostic knowledge from known datasets to perform fault diagnosis tasks in the target domain. However, this approach assumes that data samples in the target domain are accessible during model training. Furthermore, well-trained models typically only handle fault diagnosis tasks in a single target domain, making it difficult to effectively transfer to other target domains and exhibiting limited generalization ability. Summary of the Invention

[0006] To address the problems and shortcomings of existing technologies, the present invention aims to provide a training method for a network model for equipment fault diagnosis based on joint transfer of knowledge from multiple source domains.

[0007] To achieve the objectives of this invention, the technical solution adopted is as follows:

[0008] The first aspect of this invention provides a method for training a network model for equipment fault diagnosis, comprising the following steps:

[0009] S1: Collect signal data representing equipment faults during equipment operation under different operating conditions. Based on the operating conditions, construct multiple labeled source domain datasets, unlabeled source domain datasets, and target domain datasets using the signal data. The equipment fault categories represented by the signal data in each labeled source domain dataset, unlabeled source domain dataset, and target domain dataset include all fault categories. The operating conditions corresponding to any two labeled source domain datasets are different, and the operating conditions corresponding to any two unlabeled source domain datasets are also different. Label each signal data in the labeled source domain datasets and target domain datasets to obtain the true label information of the equipment fault category corresponding to each signal data. The signal data in the unlabeled source domain datasets are not labeled.

[0010] S2: Sequentially extract the same number of signal data of the same fault category from all labeled source domain datasets and unlabeled source domain datasets to form a batch training sample. Construct multiple batch training samples according to the method for constructing a batch training sample. The multiple batch training samples form a training sample set. Sequentially extract the same number of signal data from all target domain datasets to form a batch verification sample. Construct multiple batch verification samples according to the method for constructing a batch verification sample. Randomly divide the multiple batch training samples into a verification set and a test set according to the proportion.

[0011] S3: Train the pre-built equipment fault diagnosis network model using the training sample set, optimize the parameters of the equipment fault diagnosis network model, and obtain the trained equipment fault diagnosis network model.

[0012] S4: Use the validation set to validate the trained equipment fault diagnosis network model obtained in step S3, evaluate the diagnostic performance of the trained equipment fault diagnosis network model, and select the optimal equipment fault diagnosis network model.

[0013] S5: Use the test set to test the optimal equipment fault diagnosis network model obtained in step S4 and evaluate the performance of the optimal equipment fault diagnosis network model.

[0014] According to the above training method for the equipment fault diagnosis network model, preferably, the equipment fault diagnosis network model consists of a feature extractor and a fault classifier. The feature extractor is used to extract equipment fault features contained in the signal data, and the fault classifier is used to learn and analyze the equipment fault features extracted by the feature extractor, predict the equipment fault category corresponding to the signal data, and output it. The feature extractor consists of a first Cov1D layer, a second Cov1D layer, a third Cov1D layer, a first fully connected layer, and a second fully connected layer connected in sequence. The first Cov1D layer is used to extract features from the input data signal samples (convolution, batch normalization, max pooling) to obtain a first-level feature map of the data signal samples. The input of the first Cov1D layer is the data signal sample, and the output of the first Cov1D layer is the first-level feature map. The second Cov1D layer is used to extract features from the first-level feature map output by the first Cov1D layer to obtain a second-level feature map and output it. The third Cov1D layer is used to extract features from the second-level feature map output by the second Cov1D layer to obtain a third-level feature map and output it. The first fully connected layer of the feature extractor is used to learn the nonlinear combination between features of the third-level feature map and integrate global information. The input of the first fully connected layer of the feature extractor is the flattened third-level feature map, and the output of the first fully connected layer of the feature extractor is a one-dimensional feature vector of the third-level feature map. The second fully connected layer of the feature extractor is used to map the one-dimensional feature vector output by the first fully connected layer to a low-dimensional feature space to generate a feature embedding vector.

[0015] According to the training method of the above-mentioned equipment fault diagnosis network model, preferably, the network structure of the first Cov1D layer, the second Cov1D layer and the third Cov1D layer in the feature extractor is the same, and each consists of a convolutional layer, a batch normalization layer and a max pooling layer connected in sequence.

[0016] According to the training method of the above-mentioned equipment fault diagnosis network model, preferably, the fault classifier consists of a first fully connected layer, a second fully connected layer and a Softmax output layer connected in sequence. The first fully connected layer of the fault classifier combines and transforms the feature embedding vector output by the second fully connected layer of the feature extractor to extract higher-level features. The input of the first fully connected layer of the fault classifier is the feature embedding vector output by the second fully connected layer of the feature extractor, and the output of the first fully connected layer of the fault classifier is a low-dimensional feature vector. The second fully connected layer of the fault classifier maps the low-dimensional feature vector output by the first fully connected layer of the fault classifier to a low-dimensional feature space to obtain a more refined and discriminative feature representation. The input of the second fully connected layer of the fault classifier is the low-dimensional feature vector output by the first fully connected layer of the fault classifier, and the output of the second fully connected layer of the fault classifier is a fault discrimination feature vector. The Softmax output layer converts the fault discrimination feature vector output by the second fully connected layer of the fault classifier into a probability distribution of fault categories using the Softmax function. The output of the Softmax output layer is a probability distribution vector of fault categories, where each element corresponds to a probability value of a fault category. The fault classifier selects the category with the highest probability as the predicted fault category and outputs it to achieve the fault diagnosis task.

[0017] According to the above training method for the equipment fault diagnosis network model, preferably, in step S3, the specific operation of training the pre-built equipment fault diagnosis network model using the training sample set and optimizing the parameters of the equipment fault diagnosis network model is as follows:

[0018] S301: Input the batch training samples from the training sample set into the feature extractor of the pre-built equipment fault diagnosis network model. Use the feature extractor to extract features from the batch training samples to obtain the equipment fault features for each signal data in the batch training samples. Calculate the marginal distribution difference between the equipment fault features of signal data from different source domain datasets in a batch training sample using the maximum mean difference. Calculate the conditional distribution difference between signal data samples with the same equipment fault category from different labeled source domain datasets in the training samples using the maximum mean difference. Construct the multi-source domain joint loss function L of the feature extractor based on the marginal distribution difference and the conditional distribution difference. f ;

[0019] S302: Input the device fault features of the signal data samples from the labeled source domain dataset extracted by the feature extractor into the fault classifier for detection, and obtain the predicted device fault category of the signal data samples from the labeled source domain dataset in the batch training samples. Based on the difference between the predicted device fault category and the true label, construct the supervised classification loss function L of the fault classifier using cross-entropy. y Then, based on the multi-source domain joint loss function L... f and supervised classification loss function L y Construct the total loss function L of the equipment fault diagnosis network model all Total loss function L all The calculation formula is as follows:

[0020] L all =L y +λ2L f

[0021] In the formula, L all Let L represent the total loss function. y L represents the supervised classification loss function. f This represents the joint loss function of multiple source domains, where λ² is an empirical parameter used during model training.

[0022] S303: Based on the total loss function L all The network parameters of the feature extractor are optimized and updated using the backpropagation algorithm until the total loss function L is reached. all Convergence; based on supervised classification loss function L y The network parameters of the fault classifier are optimized and updated using the backpropagation algorithm until the supervised classification loss function L is reached. y The training network model for equipment fault diagnosis is obtained after convergence.

[0023] According to the above-described training method for the equipment fault diagnosis network model, preferably, in step S301, the multi-source domain joint loss function L... f The calculation formula is as follows:

[0024]

[0025] In the formula, N is the number of source domain datasets participating in the model training process; M is the number of labeled source domain datasets; N c The number of fault categories in the labeled source domain; MMD (m) It is a measure of marginal distribution difference, MMD (c) Z is a measure of the difference in conditional distribution corresponding to the c-th fault category; k and Z pThese represent the deep feature distribution fault feature sets corresponding to the k-th labeled or unlabeled source domain and the p-th labeled or unlabeled source domain, respectively. and Let X and Y represent the fault category feature sets corresponding to the c-th fault category in the k-th and p-th labeled source domains, respectively; λ1 is the empirical parameter corresponding to the conditional distribution difference measure; X and Y correspond to different source domain datasets in the batch training samples, respectively. x i Let y be the i-th sample in the source domain dataset X; j Let be the j-th sample in the source domain dataset Y; m represents the number of samples in the source domain dataset X; n represents the number of samples in the source domain dataset Y; k(·,·) is the kernel function.

[0026] According to the above-described training method for the equipment fault diagnosis network model, preferably, in step S302, the supervised classification loss function L... y The calculation formula is as follows:

[0027]

[0028] In the formula, L y Represents the supervised classification loss function; o s This represents the device fault characteristics extracted from labeled source domain data samples; y s This represents the true label of the fault category corresponding to the signal data sample from the labeled source domain dataset; n s represents the number of samples in a batch of labeled source domain datasets; C represents the number of device fault categories; I[·] represents the probability score of the fault category of the signal data samples output by the SoftMax output layer.

[0029] Based on the above-described training method for the equipment fault diagnosis network model, preferably, in step S303, the network parameters θ of the feature extractor are optimized and updated. f The formula is as follows:

[0030]

[0031] In the formula, θ f L is the parameter of the feature extractor. y Let L represent the supervised classification loss function, ε be the learning rate of the network parameters, and L be the learning rate of the network parameters. f λ represents the joint loss function of multiple source domains, and λ2 is an empirical parameter used during model training.

[0032] Based on the training method of the equipment fault diagnosis network model described above, preferably, in step S303, the network parameters θ of the fault classifier are optimized and updated. y The formula is as follows:

[0033]

[0034] In the formula, θ y These are the parameters of the fault classifier, ε is the learning rate of the network parameters, and L is the learning rate. y This represents the supervised classification loss function.

[0035] The second aspect of the present invention provides a method for diagnosing equipment faults, specifically comprising: acquiring signal data characterizing equipment faults during equipment operation, inputting the signal data into an equipment fault diagnosis network model for detection, and obtaining fault category information of the equipment; wherein the equipment fault diagnosis network model is an equipment fault diagnosis network model trained using the training method described in the first aspect above.

[0036] A third aspect of the present invention provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the training method as described in the first aspect above, or the device fault diagnosis method as described in the second aspect above.

[0037] A fourth aspect of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the training method as described in the first aspect above, or the device fault diagnosis method as described in the second aspect above.

[0038] Compared with the prior art, the positive effects achieved by the present invention are as follows:

[0039] (1) This invention is based on a multi-source domain transfer learning strategy. It uses a multi-source domain dataset consisting of several labeled and unlabeled source domain datasets with a small number of labels to train the model. Moreover, in order to learn domain-invariant features, a multi-source domain joint difference loss function based on the maximum mean difference is designed to reduce the marginal distribution difference and conditional distribution difference between different domains, minimize domain-related information, enhance the domain invariance of the learned features, and effectively improve the domain generalization performance of the model. At the same time, the multi-source domain joint difference loss is combined with the supervised classification loss to update the network parameters together during the model training process. It learns the domain-invariant features in the fault samples and transfers the shared diagnostic knowledge from labeled and unlabeled data to the target domain. Therefore, the trained equipment fault diagnosis network model does not need to obtain the target samples in advance to complete the cross-domain fault diagnosis task.

[0040] (2) The equipment fault diagnosis method of the present invention ensures good diagnostic performance and domain generalization performance when facing unknown working condition samples; moreover, in a specific embodiment, the present invention uses the aero-engine inter-shaft bearing fault dataset to establish multiple sets of experiments under different cross-working condition scenarios. By comparing with other methods, the effectiveness of the diagnostic method of the present invention in cross-working condition fault diagnosis performance for aero-engine components is verified. Attached Figure Description

[0041] Figure 1 This is a flowchart illustrating the training method of the equipment fault diagnosis network model of the present invention.

[0042] Figure 2 This is a schematic diagram of the network structure of the equipment fault diagnosis network model of the present invention;

[0043] Figure 3 This is a schematic diagram of the network structure of the feature extractor in the equipment fault diagnosis network model of the present invention;

[0044] Figure 4 This is a schematic diagram of the network structure of the fault classifier in the equipment fault diagnosis network model of the present invention. Detailed Implementation

[0045] The present invention will be further described in detail below through specific embodiments, but this does not limit the scope of the present invention.

[0046] Example 1:

[0047] A training method for a device fault diagnosis network model (e.g.) Figure 1 (As shown), including the following steps:

[0048] S1: Collect signal data representing equipment faults during equipment operation under different operating conditions. Based on the operating conditions, construct multiple labeled source domain datasets, unlabeled source domain datasets, and target domain datasets using the signal data. The equipment fault categories represented by the signal data in each labeled source domain dataset, unlabeled source domain dataset, and target domain dataset include all fault categories. Any two labeled source domain datasets correspond to different operating conditions, and any two unlabeled source domain datasets correspond to different operating conditions. Label each signal data in the labeled source domain dataset and target domain dataset to obtain the true label information of the equipment fault category corresponding to each signal data. The signal data in the unlabeled source domain dataset is not labeled.

[0049] S2: Sequentially extract the same number of signal data of the same fault category from all labeled source domain datasets and unlabeled source domain datasets to form a batch training sample. Construct multiple batch training samples according to the construction method of the batch training sample. The multiple batch training samples form a training sample set. Sequentially extract the same number of signal data from all target domain datasets to form a batch verification sample. Construct multiple batch verification samples according to the construction method of the batch verification sample. Randomly divide the multiple batch training samples into a verification set and a test set according to the proportion.

[0050] S3: Train the pre-built equipment fault diagnosis network model using the training sample set, optimize the parameters of the equipment fault diagnosis network model, and obtain the trained equipment fault diagnosis network model.

[0051] S4: Use the validation set to validate the trained equipment fault diagnosis network model obtained in step S3, evaluate the diagnostic performance of the trained equipment fault diagnosis network model, and select the optimal equipment fault diagnosis network model.

[0052] S5: Use the test set to test the optimal equipment fault diagnosis network model obtained in step S4 and evaluate the performance of the optimal equipment fault diagnosis network model.

[0053] As a preferred implementation, in step S1 above, when constructing multiple labeled source domain datasets, unlabeled source domain datasets, and target domain datasets using signal data according to operating conditions, sample segmentation and frequency domain transformation can be performed on the signal data samples in the labeled source domain datasets, unlabeled source domain datasets, and target domain datasets to obtain more signal data samples.

[0054] As a preferred embodiment, the device fault diagnosis network model (e.g.) Figure 2 The system (as shown) consists of a feature extractor and a fault classifier. The feature extractor extracts equipment fault features from the signal data, and the fault classifier learns and analyzes the equipment fault features extracted by the feature extractor to predict and output the equipment fault category corresponding to the signal data. The feature extractor (e.g., ...) Figure 3The feature extractor consists of a first Cov1D layer, a second Cov1D layer, a third Cov1D layer, a first fully connected layer, and a second fully connected layer connected in sequence. The first Cov1D layer, the second Cov1D layer, and the third Cov1D layer in the feature extractor have the same network structure, each consisting of a convolutional layer, a batch normalization layer, and a max pooling layer connected in sequence. The first Cov1D layer is used to perform preliminary feature extraction on the input data signal sample, extracting short-time features and suppressing high-frequency noise to obtain a first-level feature map of the data signal sample. The input of the first Cov1D layer is the data signal sample, and the output of the first Cov1D layer is the first-level feature map. The second Cov1D layer is used to extract features from the first-level feature map output by the first Cov1D layer, extracting deeper local features to obtain a second-level feature map and outputting it. The third Cov1D layer is used to extract features from the second-level feature map output by the second Cov1D layer to further extract fault features from the signal data sample, obtaining a third-level feature map and outputting it. The first fully connected layer of the feature extractor is used to learn the nonlinear combination between features of the third-level feature map and integrate global information. The input of the first fully connected layer of the feature extractor is the flattened third-level feature map, and the output of the first fully connected layer of the feature extractor is the one-dimensional feature vector of the third-level feature map. The second fully connected layer of the feature extractor is used to map the one-dimensional feature vector output by the first fully connected layer to a low-dimensional feature space to generate a feature embedding vector.

[0055] The fault classifier (e.g.) Figure 4 (As shown) The fault classifier consists of a first fully connected layer, a second fully connected layer, and a Softmax output layer connected sequentially. The first fully connected layer of the fault classifier is used to combine and transform the feature embedding vector output by the second fully connected layer of the feature extractor to extract higher-level features. The input of the first fully connected layer of the fault classifier is the feature embedding vector output by the second fully connected layer of the feature extractor, and the output of the first fully connected layer of the fault classifier is a low-dimensional feature vector. The second fully connected layer of the fault classifier is used to map the low-dimensional feature vector output by the first fully connected layer of the fault classifier to a low-dimensional feature space to obtain more refined and accurate features. The fault classifier uses a highly distinctive feature representation. The input to the second fully connected layer of the fault classifier is the low-dimensional feature vector output by the first fully connected layer of the fault classifier, and the output of the second fully connected layer is the fault discrimination feature vector. The Softmax output layer transforms the fault discrimination feature vector output by the second fully connected layer of the fault classifier into a probability distribution of fault categories through the Softmax function. The output of the Softmax output layer is the probability distribution vector of fault categories, where each element corresponds to a probability value of a fault category. The fault classifier selects the category with the highest probability as the predicted fault category and outputs it, thus realizing the fault diagnosis task.

[0056] As a preferred implementation, the specific operation of training the pre-built equipment fault diagnosis network model using the training sample set in step S3 above is as follows:

[0057] S301: Input a batch of training samples from the training sample set into the feature extractor of the pre-built equipment fault diagnosis network model. The feature extractor extracts features from the batch of training samples to obtain equipment fault features for each signal data in the batch of training samples. Calculate the marginal distribution difference between the equipment fault features of signal data from different source domain datasets in a batch of training samples using the maximum mean difference. Calculate the conditional distribution difference between signal data samples with the same equipment fault category from different labeled source domain datasets in the training samples using the maximum mean difference. Construct the multi-source domain joint loss function L of the feature extractor based on the marginal distribution difference and the conditional distribution difference. f .

[0058] The multi-source domain joint loss function L f The calculation formula is as follows:

[0059]

[0060] In the formula, N is the number of source domain datasets participating in the model training process; M is the number of labeled source domain datasets; N c The number of fault categories in the labeled source domain; MMD (m) It is a measure of marginal distribution difference, MMD (c) Z is a measure of the difference in conditional distribution corresponding to the c-th fault category; k and Z p These represent the deep feature distribution fault feature sets corresponding to the k-th labeled or unlabeled source domain and the p-th labeled or unlabeled source domain, respectively. and Let X and Y represent the fault category feature sets corresponding to the c-th fault category in the k-th and p-th labeled source domains, respectively; λ1 is the empirical parameter corresponding to the conditional distribution difference measure; X and Y correspond to different source domain datasets in the batch training samples, respectively. x i Let y be the i-th sample in the source domain dataset X; j Let be the j-th sample in the source domain dataset Y; m represents the number of samples in the source domain dataset X; n represents the number of samples in the source domain dataset Y; k(·,·) is the kernel function.

[0061] S302: Input the device fault features of the signal data samples from the labeled source domain dataset extracted by the feature extractor into the fault classifier for detection, and obtain the predicted device fault category of the signal data samples from the labeled source domain dataset in the batch training samples. Based on the difference between the predicted device fault category and the true label, construct the supervised classification loss function L of the fault classifier using cross-entropy. y Then, based on the multi-source domain joint loss function L... f and supervised classification loss function L y Construct the total loss function L of the equipment fault diagnosis network model all .

[0062] The supervised classification loss function L y The calculation formula is as follows:

[0063]

[0064] In the formula, L y Represents the supervised classification loss function; o s This represents the device fault characteristics extracted from labeled source domain data samples; y s This represents the true label of the fault category corresponding to the signal data sample from the labeled source domain dataset; n s represents the number of samples in a batch of labeled source domain datasets; C represents the number of device fault categories; I[·] represents the probability score of the fault category of the signal data samples output by the SoftMax output layer.

[0065] The total loss function L all The calculation formula is as follows:

[0066] L all =L y +λ2L f

[0067] In the formula, L all Let L represent the total loss function. y L represents the supervised classification loss function. f λ represents the joint loss function of multiple source domains, and λ2 is an empirical parameter used during model training.

[0068] S303: Based on the total loss function L all The network parameters of the feature extractor are optimized and updated using the backpropagation algorithm until the total loss function L is reached. all Convergence; based on supervised classification loss function L y The network parameters of the fault classifier are optimized and updated using the backpropagation algorithm until the supervised classification loss function L is reached. yThe process converges, resulting in a trained device fault diagnosis network model. This model optimizes and updates the network parameters θ of the feature extractor. f The network parameters θ of the fault classifier y The formula is as follows:

[0069]

[0070] In the formula, θ f L is the parameter of the feature extractor. y Let L represent the supervised classification loss function, ε be the learning rate of the network parameters, and L be the learning rate of the network parameters. f This represents the joint loss function of multiple source domains, where λ² is an empirical parameter used during model training, and θ... y These are the parameters of the fault classifier, ε is the learning rate of the network parameters, and L is the learning rate. y This represents the supervised classification loss function.

[0071] S304: Repeat steps S301-S303 above until all batches of samples in the training sample set are traversed to obtain multiple trained equipment fault diagnosis network models.

[0072] Example 2:

[0073] The content of Example 2 is basically the same as that of Example 1, except that:

[0074] In this embodiment, the fault diagnosis targets the rolling bearings of an aero-engine. The signal data used to create the training, validation, and test sets are all vibration signal data collected from a real aero-engine prototype under varying speed conditions. To obtain more training samples, the vibration signal data collected under different speed conditions can be further segmented and frequency-domain transformed using equal-window partitioning. The rolling bearing fault categories corresponding to the collected vibration signal data samples include three types: 0 – healthy; 1 – inner race fault; 2 – outer race fault. These three fault categories are used as fault labels to annotate each vibration signal data in the labeled source domain dataset and target domain dataset, obtaining the true label information for each vibration signal data in the labeled source domain dataset and target domain dataset.

[0075] In one specific embodiment, to collect vibration signal data under different operating conditions of an aircraft engine, this invention sets up six detection points at different locations on the aircraft engine, corresponding to sensor numbers #1 to #6. Two displacement sensors (#1, #2) are used to measure vibration displacement response, and four acceleration sensors (#3, #4, #5, #6) are used to measure vibration acceleration response. The sensor sampling frequency is 25000Hz. Cracks are artificially added to the outer and inner rings of the bearing, including outer ring faults (depth and length 0.5mm) and inner ring faults (depth 0.5mm, length 0.5mm or 1.0mm). Tests are conducted and data samples are collected under 28 different low and high pressure rotor speed conditions. The information of the vibration signal data sample set obtained after data preprocessing is shown in Table 1 below. Based on the collected vibration signal data sample set, there are labeled source domain datasets, unlabeled source domain datasets, and target domain datasets, as shown in Table 2.

[0076] Table 1. Detailed information on the sample set of vibration signal data from the aircraft engine collected in the experiment.

[0077]

[0078] Table 2 shows the partitioning results of the labeled source domain dataset, the unlabeled source domain dataset, and the target domain dataset.

[0079]

[0080] Furthermore, when used for fault diagnosis of rolling bearings in aero-engines, the network layer parameter settings for the feature extractor and fault classifier of the equipment fault diagnosis network model are shown in Table 3 below. During training, the relevant parameters selected are as follows: the sample length of the input data is 1024, and the number of fault categories N... c =3, the maximum number of iterations N for network training tra =50, λ1=0.8, λ2=2, ∈=0.0005.

[0081] Table 3. Network layer parameter settings for the feature extractor and fault classifier in the equipment fault diagnosis network model.

[0082]

[0083] To evaluate the effectiveness of the training method for the equipment fault diagnosis network model in Embodiment 2 of the present invention, the following methods are compared with the training method of the present invention:

[0084] (1) Baseline method: In this method, the model is trained using only a single labeled source domain dataset, and the training process employs supervised classification loss L. yAs a loss function, the network parameters of the feature extractor and fault classifier are optimized and updated through the backpropagation algorithm until the supervised classification loss function L is reached. y The training converges, resulting in a trained network model for equipment fault diagnosis. This training method does not include any transfer learning strategy. The model is then directly used to diagnose the target task, yielding the final diagnostic result.

[0085] (2) Comparison with Method 1: A multi-source domain learning method without transferred knowledge. Considering the advantages of multi-source domain training, this method uses multiple labeled source domain datasets to co-train the model. Comparison with Method 1 removes the multi-source domain joint difference loss used to measure distribution differences between domains, which is used to verify the effectiveness of the multi-source domain joint difference loss in the proposed method. Specifically, multiple labeled source domain datasets are used to co-train the model, and the training process only uses supervised classification loss L. y As the loss function, the network parameters of the fault classifier are optimized and updated through the backpropagation algorithm until the supervised classification loss function L is reached. y The training network model for equipment fault diagnosis is obtained after convergence.

[0086] (3) Comparison Method 2: Multi-source domain transfer learning method. Based on Comparison Method 1, this method optimizes model parameters during training by using multi-source domain joint difference loss and supervised classification loss as the total loss. It employs a transfer learning strategy to address domain difference issues and handle unknown target tasks. Specifically, it uses multiple labeled source domain datasets for collaborative training of the model. The training process uses multi-source domain joint loss L... f And supervision, supervision classification loss L y Construct the total loss function L all The network parameters of the fault classifier are optimized and updated using the backpropagation algorithm until the total loss function L is reached. all The training network model for equipment fault diagnosis is obtained after convergence.

[0087] Each training method's device fault diagnosis network model was tested in each experimental scenario of the target task cases corresponding to the four target domain datasets listed in Table 2. The diagnostic accuracy of these device fault diagnosis network models is shown in Table 4. For the baseline method, each labeled source domain dataset was used individually for network training. Table 4 shows the accuracy of the models obtained after training in different single source domains on the target task (two tables: representing the accuracy obtained on the target task after training the model on labeled source domain dataset 1 alone; and the accuracy obtained on the target task after training the model on labeled source domain dataset 2 alone).

[0088] Table 4. Accuracy (%) of the equipment fault diagnosis network models obtained by different training methods.

[0089]

[0090] As shown in Table 4, compared with other methods, the model training method of this invention has higher diagnostic accuracy and smaller fluctuation range in all four cases, and can more effectively diagnose unknown target tasks. Specifically, in Case 1, when the baseline method is trained using only a single labeled source domain dataset B, the diagnostic accuracy for the two target tasks is 74.44% and 91.11%, respectively. When trained using only labeled source domain dataset C, the accuracy is 84.44% and 66.67%, respectively, which is low and the diagnostic accuracy differs greatly between the two target tasks. Although comparative methods 1 and 2 achieved the same diagnostic accuracy of 100% for target task 1 as the proposed method, their diagnostic accuracy for target task 2 is only 77.78% and 69.44%, respectively, which is significantly lower than the model training method of this invention, and there is a significant difference in diagnostic accuracy between different target tasks. The model training method of this invention achieves 100% accuracy in both target tasks, surpassing the other comparative methods.

[0091] The baseline method is trained using only a single labeled source domain dataset. Due to the lack of diverse knowledge, this method exhibits poor diagnostic performance when there are significant differences between the target domain and the source domain, leading to a large difference in diagnostic accuracy between the two target tasks. Comparative method 1 utilizes multiple labeled source domain datasets to train the model, achieving better diagnostic performance. Comparative method 2 further improves the model's generalization performance by integrating feature information and diagnostic knowledge through a transfer learning strategy based on multi-source domain data. While it shows similar diagnostic accuracy to the proposed method in some diagnostic tasks, significant differences in diagnostic accuracy exist between different target tasks, indicating poor generalization performance. The model training method proposed in this invention updates network parameters by combining multi-source domain joint difference loss with supervised classification loss, reducing the differences in marginal and conditional distributions between domains. This enhances the learning of domain-invariant features, shares diagnostic knowledge from multiple source domain datasets, and achieves good diagnostic performance across different target tasks.

[0092] In Case 4, the source and target domain datasets contain data samples collected from sensors at different measurement points. In the baseline method, when trained using only a single labeled source domain dataset A, the diagnostic accuracy for both target tasks is only 33.33%. Comparative method 1 achieves diagnostic accuracies of 100% and 69.44% for the two target tasks, respectively. Comparative method 2 optimizes the inter-domain boundaries, achieving accuracies of 100% and 78.33% for the same tasks, demonstrating better performance. The model training method of this invention has significant advantages over other poorly performing comparative methods, achieving accuracies of 96.67% and 100%, both greater than 96%, and the diagnostic accuracy varies little across different target tasks. The results show that the model training method of this invention, trained with limited and partially labeled multi-source domain datasets (including labeled and unlabeled source domain datasets), has good domain-invariant feature extraction capabilities for fault samples and can generalize to unknown target domains, exhibiting good diagnostic accuracy and domain generalization performance.

[0093] In the four sets of experiments, the diagnostic accuracy of the model training method of this invention reached over 96%, meeting the expected results of the target task. This fully demonstrates that the model training method of this invention can still accurately and effectively diagnose samples even when target domain samples are unavailable for network training. In summary, the model training method of this invention has a good ability to extract domain-invariant features, effectively solving the problem of difficulty in diagnosing target task fault samples due to the lack of complete training samples. It is more suitable for fault diagnosis in complex cross-condition scenarios and has broad application prospects in actual industrial fault diagnosis.

[0094] Example 3:

[0095] A method for diagnosing equipment faults includes: acquiring signal data characterizing equipment faults during equipment operation, inputting the signal data into an equipment fault diagnosis network model for detection, and obtaining fault category information of the equipment; wherein the equipment fault diagnosis network model is an equipment fault diagnosis network model trained using the training method described in Example 1.

[0096] Example 4:

[0097] A method for diagnosing aircraft engine bearing faults includes: acquiring vibration signal data of the aircraft engine during operation, inputting the vibration signal data into a device fault diagnosis network model for detection, and obtaining fault category information of the aircraft engine bearing; wherein the device fault diagnosis network model is a device fault diagnosis network model trained using the training method described in Example 2.

[0098] Example 5:

[0099] An electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement a training method as described in Embodiment 1 or Embodiment 2, or a device fault diagnosis method as described in Embodiment 3 or Embodiment 4.

[0100] Example 6:

[0101] A computer-readable storage medium storing a computer program, characterized in that, when executed by a processor, the computer program implements the training method as described in Embodiment 1 or Embodiment 2, or the device fault diagnosis method as described in Embodiment 3 or Embodiment 4.

[0102] Finally, it should be noted that the above embodiments are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may use the above technical content as inspiration to make changes or modifications. These are equivalent embodiments with variations. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the technical concept of the present invention still fall within the protection scope of the claims of the present invention.

Claims

1. A training method for a network model for equipment fault diagnosis, characterized in that, Includes the following steps: S1: Collect signal data representing equipment faults during equipment operation under different operating conditions. Based on the operating conditions, construct multiple labeled source domain datasets, unlabeled source domain datasets, and target domain datasets using the signal data. The equipment fault categories represented by the signal data in each labeled source domain dataset, unlabeled source domain dataset, and target domain dataset include all fault categories. Any two labeled source domain datasets correspond to different operating conditions, and any two unlabeled source domain datasets correspond to different operating conditions. Each signal data in the labeled source domain dataset and target domain dataset is labeled to obtain the true label information of the equipment fault category corresponding to each signal data; The signal data in the unlabeled source domain dataset is not labeled. S2: Sequentially extract the same number of signal data of the same fault category from all labeled source domain datasets and unlabeled source domain datasets to form a batch training sample. Construct multiple batch training samples according to the construction method of the batch training sample, and the multiple batch training samples form a training sample set. The same number of signal data are extracted from all target domain datasets in sequence to form a batch verification sample. Multiple batch verification samples are constructed according to the construction method of the batch verification sample. The multiple batch training samples are randomly divided into a verification set and a test set according to the proportion. S3: Train the pre-built equipment fault diagnosis network model using the training sample set, optimize the parameters of the equipment fault diagnosis network model, and obtain the trained equipment fault diagnosis network model; the equipment fault diagnosis network model consists of a feature extractor and a fault classifier; S4: Use the validation set to validate the trained equipment fault diagnosis network model obtained in step S3, evaluate the diagnostic performance of the trained equipment fault diagnosis network model, and select the optimal equipment fault diagnosis network model. S5: Test the optimal equipment fault diagnosis network model obtained in step S4 using a test set, and evaluate the performance of the optimal equipment fault diagnosis network model. In step S3, the specific operation of training the pre-built equipment fault diagnosis network model using the training sample set and optimizing the parameters of the equipment fault diagnosis network model is as follows: S301: Input the batch training samples from the training sample set into the feature extractor of the pre-built equipment fault diagnosis network model. Use the feature extractor to extract features from the batch training samples to obtain the equipment fault features for each signal data in the batch training samples. Calculate the marginal distribution difference between the equipment fault features of signal data from different source domain datasets in a batch training sample using the maximum mean difference. Calculate the conditional distribution difference between signal data samples with the same equipment fault category from different labeled source domain datasets in the training samples using the maximum mean difference. Construct the multi-source domain joint loss function L of the feature extractor based on the marginal distribution difference and the conditional distribution difference. f ; S302: Input the device fault features of the signal data samples from the labeled source domain dataset extracted by the feature extractor into the fault classifier for detection, and obtain the predicted device fault category of the signal data samples from the labeled source domain dataset in the batch training samples. Based on the difference between the predicted device fault category and the true label, construct the supervised classification loss function L of the fault classifier using cross-entropy. y Then, based on the multi-source domain joint loss function L... f and supervised classification loss function L y Construct the total loss function L of the equipment fault diagnosis network model all Total loss function L all The calculation formula is as follows: L all =L y +λ2L f In the formula, L all Let L represent the total loss function. y L represents the supervised classification loss function. f This represents the joint loss function of multiple source domains, where λ² is an empirical parameter used during model training. S303: Based on the total loss function L all The network parameters of the feature extractor are optimized and updated using the backpropagation algorithm until the total loss function L is reached. all Convergence; based on supervised classification loss function L y The network parameters of the fault classifier are optimized and updated using the backpropagation algorithm until the supervised classification loss function L is reached. y The training network model for equipment fault diagnosis is obtained after convergence.

2. The training method according to claim 1, characterized in that, The feature extractor is used to extract equipment fault features contained in the signal data, and the fault classifier is used to learn and analyze the equipment fault features extracted by the feature extractor, predict the equipment fault category corresponding to the signal data and output it; wherein, the feature extractor is composed of a first Cov1D layer, a second Cov1D layer, a third Cov1D layer, a first fully connected layer and a second fully connected layer connected in sequence.

3. The training method according to claim 2, characterized in that, The first Cov1D layer, the second Cov1D layer, and the third Cov1D layer in the feature extractor have the same network structure, each consisting of a convolutional layer, a batch normalization layer, and a max pooling layer connected in sequence.

4. The training method according to claim 3, characterized in that, The fault classifier consists of a first fully connected layer, a second fully connected layer, and a Softmax output layer connected in sequence.

5. The training method according to claim 1, characterized in that, In step S301, the multi-source domain joint loss function L f The calculation formula is as follows: In the formula, N is the number of source domain datasets participating in the model training process; M is the number of labeled source domain datasets; N c The number of fault categories in the labeled source domain; MMD (m) It is a marginal distribution difference, MMD (c) It is the difference in conditional distribution corresponding to the c-th type of fault; Z k and Z p These represent the deep feature distribution fault feature sets corresponding to the k-th labeled or unlabeled source domain and the p-th labeled or unlabeled source domain, respectively. and Let X and Y represent the fault category feature sets corresponding to the c-th fault category in the k-th and p-th labeled source domains, respectively; λ1 is the empirical parameter corresponding to the conditional distribution difference measure; X and Y correspond to different source domain datasets in the batch training samples, respectively. x i Let y be the i-th sample in the source domain dataset X; j Let be the j-th sample in the source domain dataset Y; m represents the number of samples in the source domain dataset X; n represents the number of samples in the source domain dataset Y; k(·,·) is the kernel function.

6. The training method according to claim 1, characterized in that, In step S302, the supervised classification loss function L y The calculation formula is as follows: In the formula, L y Represents the supervised classification loss function; o s This represents the device fault characteristics extracted from labeled source domain data samples; y s This represents the true label of the fault category corresponding to the signal data sample from the labeled source domain dataset; n s represents the number of samples in a batch of labeled source domain datasets; C represents the number of device fault categories; I[·] represents the probability score of the fault category of the signal data samples output by the SoftMax output layer.

7. A method for diagnosing equipment faults, characterized in that, The system acquires signal data characterizing equipment faults during equipment operation, inputs the signal data into an equipment fault diagnosis network model for detection, and obtains fault category information of the equipment; wherein, the equipment fault diagnosis network model is an equipment fault diagnosis network model trained using any of the training methods described in claims 1 to 6.

8. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and when the processor executes the computer program, it implements the training method as described in any one of claims 1 to 6, or the device fault diagnosis method as described in claim 7.

9. A computer-readable storage medium storing a computer program thereon, characterized in that, When the computer program is executed by the processor, it implements the training method as described in any one of claims 1 to 6, or the device fault diagnosis method as described in claim 7.

Images