Single-source-domain generalization intelligent diagnosis method based on targeted data enhancement

Abstract

The invention provides a single-source-domain generalization intelligent diagnosis method based on targeted data enhancement, and the method comprises the steps: collecting a state monitoring vibration signal of a key part of industrial equipment under different working conditions, and carrying out the data preprocessing, and the preprocessing comprises signal interception and normalization processing; dividing the state monitoring vibration signals into a source domain sample set and a target domain sample set; constructing a basic fault diagnosis model, wherein the model comprises a targeted enhancement chain, a distributed mixing layer, a feature extractor and a fault classifier; inputting the source domain sample set into the basic fault diagnosis model for model training to obtain a target fault diagnosis model; and inputting the target domain sample set into a target fault diagnosis model, and outputting a fault category of the sample. According to the method, the fault recognition accuracy and robustness under the unknown target working condition are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial equipment fault diagnosis, and in particular to a single-source domain generalization intelligent diagnosis method based on targeted data enhancement. BACKGROUND

[0002] With the progress of science and technology and the development of society, the structure and function of mechanical equipment in modern industry are increasingly complex, and efficient and reliable fault diagnosis is the key to ensuring safe production. Intelligent diagnosis methods based on deep learning have been widely used in this context and have shown great potential. However, in actual engineering, industrial equipment often operates under different working conditions (such as changes in speed and load), resulting in significant differences in the distribution of its condition monitoring data (one working condition can be regarded as one "domain"). This domain shift phenomenon seriously hinders the generalization ability of deep learning models in unknown working conditions. To solve the challenge of cross-domain fault diagnosis, domain generalization methods have emerged. The core goal of this type of method is to use data from multiple source domains to learn domain-invariant fault features during the training phase, thereby constructing a diagnosis model that can directly generalize to unknown target domains. Although domain generalization technology shows good prospects, its strong dependence on multi-source domain data poses a major obstacle in actual industrial scenarios. Due to the particularity of equipment operating conditions, data acquisition costs or safety factors, it is often difficult to obtain sufficient and diverse multi-condition samples, and only condition monitoring data from a single source domain is usually available. This limitation has led to the study of single-source domain generalization technology. This technology aims to train a model with strong generalization ability to any unknown target domain using only data from a single source domain.

[0003] Currently, research in this field has developed along two main directions: the first is a data augmentation-based strategy: by applying image transformations or more advanced feature-level interpolation and decoupling methods on the source domain, artificial distribution diversity is created to simulate potential target domain data; the second is a feature learning-based strategy: by designing new network structures or loss functions, the goal is to directly extract more essential and robust domain-invariant features from the data. However, both directions face serious challenges under the single-source domain setting: overly aggressive data augmentation may destroy key fault semantic information in the data; and the lack of sufficient distribution diversity in the training data makes it difficult for the model to learn truly generalizable feature representations. Therefore, how to ensure the distribution diversity and integrity of fault semantics of generated samples under the condition of a single source domain, and based on this to extract strongly generalizable fault features, has become a core bottleneck that needs to be broken through in the field of intelligent diagnosis of industrial equipment. SUMMARY

[0004] In view of the above, the present application provides a single-source domain generalization intelligent diagnosis method based on targeted data enhancement to solve the above problems.

[0005] The application provides a single-source domain generalization intelligent diagnosis method based on targeted data enhancement, comprising: collecting state monitoring vibration signals of key parts of industrial equipment under different working conditions, and performing data preprocessing, wherein the preprocessing comprises signal interception and normalization processing; dividing the state monitoring vibration signals into a source domain sample set and a target domain sample set; constructing a basic fault diagnosis model, wherein the model comprises a targeted enhancement chain, a distribution mixing layer, a feature extractor and a fault classifier; inputting the source domain sample set into the basic fault diagnosis model for model training to obtain a target fault diagnosis model; and inputting the target domain sample set into the target fault diagnosis model to output the fault category of the sample.

[0006] In another implementation mode of the application, the dividing of the state monitoring vibration signals into the source domain sample set and the target domain sample set comprises: labeling the fault category label of the state monitoring vibration signals collected under one working condition as the source domain sample set; and collecting the state monitoring vibration signals under other working conditions as the target domain sample set.

[0007] In another implementation mode of the application, the constructing of the basic fault diagnosis model, wherein the model comprises a targeted enhancement chain, a distribution mixing layer, a feature extractor and a fault classifier, comprises: constructing a targeted enhancement chain, wherein the targeted enhancement chain is used for mining fault information from the source domain sample set and generating an enhanced domain sample with diversified data distribution and rich fault semantic information; constructing a distribution mixing layer, wherein the distribution mixing layer is used for probabilistically mixing the feature statistics of the enhanced domain sample and the source domain sample to synthesize a mixed domain sample with diversified data distribution; constructing a feature extractor, wherein the feature extractor is used for extracting high-level implicit features from the source domain, the enhanced domain and the mixed domain sample; constructing a fault classifier, wherein the fault classifier is used for predicting the fault category label of the high-level implicit features; and combining the targeted enhancement chain, the distribution mixing layer, the feature extractor and the fault classifier into a feedforward neural network to construct the basic fault diagnosis model.

[0008] In another implementation mode of the application, the targeted enhancement chain is composed of a first encoder G1 and a second encoder G2; the input of the first encoder G1 is the source domain sample, and the first encoder G1 generates an enhanced domain sample with the same category and the same dimension size as the input sample by extracting key information from the input sample; the input of the second encoder G2 is the enhanced domain sample, and the second encoder G2 generates an auxiliary domain sample with the same category and the same dimension size as the input sample by extracting key information from the input sample.

[0009] In another implementation mode of the application, the distance loss between the source domain sample and the enhanced domain sample is represented as:

[0010] wherein, denotes the infimum; is and a set of joint distributions; E denotes the mathematical expectation; denotes the Frobenius norm; is a source domain sample; is an enhanced domain sample.

[0011] In another implementation of the present application, the fault semantic information consistency constraint loss imposed between the auxiliary domain and the source domain sample is expressed as:

[0012] wherein, denotes a source domain; denotes the mean absolute error; is an auxiliary domain sample.

[0013] In another implementation of the present application, the mixed domain sample is expressed as:

[0014] wherein, is the input of the mixed layer; and denote mean and standard deviation calculation, respectively; is a feature statistic mean; is a standard deviation.

[0015] In another aspect of the present application, a single-source domain generalization intelligent diagnosis system based on targeted data enhancement is provided, comprising: a data acquisition module: under different working conditions, the state monitoring vibration signals of key parts of industrial equipment are collected, and data preprocessing is performed, the preprocessing including signal interception and normalization processing; a data set division module: the state monitoring vibration signals are divided into a source domain sample set and a target domain sample set; a model training module: a basic fault diagnosis model is constructed, the model including a targeted enhancement chain, a distribution mixing layer, a feature extractor and a fault classifier; the source domain sample set is input into the basic fault diagnosis model for model training, to obtain a target fault diagnosis model; a result output module: the target domain sample set is input into the target fault diagnosis model, and the fault category of the sample is output.

[0016] In another aspect of the present application, an electronic device is provided, comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps of the single-source domain generalization intelligent diagnosis method based on targeted data enhancement according to any one of the above. In another aspect, the present invention provides a computer storage medium storing a computer program, which, when executed by a processor, implements the steps of a single-source domain generalized intelligent diagnostic method based on targeted data augmentation as described in any of the preceding claims.

[0017] This invention presents a single-source-domain generalized intelligent diagnostic method based on targeted data augmentation. It constructs a targeted augmentation chain and utilizes a dual-encoder structure to generate augmented samples with diverse distributions and complete fault information. Further, a distributed hybrid layer is designed to synthesize more abundant hybrid domain samples by probabilistically and randomly mixing the feature statistics of source and augmented domain samples. Finally, a feature extractor and a fault classifier are combined to construct a diagnostic model capable of learning strongly generalized fault features from a single source domain. This invention requires only labeled data under a single operating condition for training, significantly improving the model's fault identification accuracy and robustness under unknown target operating conditions. It effectively solves the problem of poor generalization of intelligent diagnostic models caused by the difficulty in obtaining multi-condition data in industrial settings. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. By reading the detailed description of the embodiments below, the advantages and benefits of the solutions will become clear to those skilled in the art. The accompanying drawings are only for illustrating preferred embodiments and are not intended to limit the present invention. In the accompanying drawings: Figure 1 This is a schematic diagram of a single-source domain generalized intelligent diagnostic method based on targeted data augmentation, according to an embodiment of the present invention.

[0019] Figure 2 This is a schematic diagram of the basic fault diagnosis model structure according to an embodiment of the present invention.

[0020] Figure 3 This is a schematic diagram of the target fault diagnosis model structure according to an embodiment of the present invention. Detailed Implementation

[0021] To facilitate understanding, before describing the specific embodiments of the present invention in detail, an exemplary description of the prior art of a single-source domain generalized intelligent diagnostic method based on targeted data augmentation will be provided.

[0022] Existing technologies related to this invention mainly include domain generalization methods based on feature alignment and single-source domain methods based on data augmentation. The former learns domain-invariant features by utilizing multiple source domain data and imposing domain alignment constraints, but its core drawback lies in its heavy reliance on multiple source domains, making it completely unapplicable in real-world industrial scenarios with only single-source domain data. The latter attempts to simulate distribution changes within a single source domain through generative models. However, these methods generally face a fundamental dilemma: if the distribution difference between generated samples and the source domain is pursued solely to improve generalization ability, the crucial fault semantic information in the data will inevitably be damaged, leading to a decrease in the diagnostic value of the generated samples; conversely, if strict reconstruction constraints are imposed to preserve fault features, it is difficult to generate sufficiently diverse distributions, resulting in limited improvement in generalization performance. Therefore, existing technologies have failed to solve the core problem of how to simultaneously achieve "diversified distribution of generated samples" and "complete preservation of fault semantic information" under single-source domain conditions, which is precisely the technical bottleneck that this invention aims to overcome.

[0023] In practical applications, the effective collection and utilization of large-scale industrial equipment condition monitoring data reflecting real-world working conditions remains a significant and challenging problem for existing industrial equipment fault diagnosis technologies. When only data from a single source domain is available, the performance of existing domain-generalized fault diagnosis methods deteriorates significantly. One feasible solution to the single-distribution problem is to generate samples with a new distribution based on the source domain training samples. However, existing methods face a dilemma: on the one hand, to improve the model's generalization ability, data augmentation is necessary to generate new samples that maximize the difference from the source domain distribution; on the other hand, in pursuing this distribution difference, the key fault semantic information contained in the generated samples is easily lost or distorted.

[0024] For example, enhancement methods based on Generative Adversarial Networks (GANs) struggle to control the fault modes of generated samples, potentially leading to blurred fault features; while simple data noise addition or transformation fails to produce sufficiently diverse data distributions. Therefore, how to synchronously generate samples with diverse distributions and sufficient fault semantic information from condition monitoring data under single-source domain conditions, and extract generalizable fault features accordingly, remains a core challenge that current technologies have yet to overcome.

[0025] To enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, the technical solutions in the embodiments of the present invention will be clearly and thoroughly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art should fall within the protection scope of the present invention.

[0026] Figure 1This is a schematic diagram of a single-source domain generalized intelligent diagnostic method based on targeted data augmentation, provided by an embodiment of the present invention. Figure 1 As shown, this embodiment mainly includes: S101. Under different working conditions, collect vibration signals from key parts of industrial equipment for status monitoring and perform data preprocessing, including signal interception and normalization.

[0027] S102. Divide the state monitoring vibration signal into a source domain sample set and a target domain sample set.

[0028] S103. Construct a basic fault diagnosis model, which includes a targeted enhancement chain, a distributed hybrid layer, a feature extractor, and a fault classifier.

[0029] S104. Input the source domain sample set into the basic fault diagnosis model for model training to obtain the target fault diagnosis model.

[0030] S105. Input the target domain sample set into the target fault diagnosis model and output the fault category of the sample.

[0031] This invention presents a single-source-domain generalized intelligent diagnostic method based on targeted data augmentation. It constructs a targeted augmentation chain and utilizes a dual-encoder structure to generate augmented samples with diverse distributions and complete fault information. Further, a distributed hybrid layer is designed to synthesize more abundant hybrid domain samples by probabilistically and randomly mixing the feature statistics of source and augmented domain samples. Finally, a feature extractor and a fault classifier are combined to construct a diagnostic model capable of learning strongly generalized fault features from a single source domain. This invention requires only labeled data under a single operating condition for training, significantly improving the model's fault identification accuracy and robustness under unknown target operating conditions. It effectively solves the problem of poor generalization of intelligent diagnostic models caused by the difficulty in obtaining multi-condition data in industrial settings.

[0032] In another implementation of the present invention, the step of dividing the state monitoring vibration signal into a source domain sample set and a target domain sample set includes: labeling the state monitoring vibration signal collected under one type of operating condition with a fault category label as the source domain sample set; and using the state monitoring vibration signal collected under other operating conditions as the target domain sample set.

[0033] For example, the source domain sample set is labeled with fault categories and is used to train the basic fault diagnosis model; the target domain sample set does not participate in model training and is used to verify the effectiveness and accuracy of the model.

[0034] In another implementation of the present invention, the construction of a basic fault diagnosis model includes a targeted enhancement chain, a distributed hybrid layer, a feature extractor, and a fault classifier, such as... Figure 2As shown, the method includes: constructing a targeted enhancement chain, which is used to mine fault information from the source domain sample set and generate enhanced domain samples with diverse data distributions and rich fault semantic information; constructing a distribution mixing layer, which is used to probabilistically and randomly mix the feature statistics of the enhanced domain samples and the source domain samples to synthesize a mixed domain sample with diverse data distributions; constructing a feature extractor, which is used to extract high-level latent features from the source domain, enhanced domain, and mixed domain samples; constructing a fault classifier, which is used to predict the fault category label of the high-level latent features; and combining the targeted enhancement chain, the distribution mixing layer, the feature extractor, and the fault classifier into a feedforward neural network to construct a basic fault diagnosis model.

[0035] For example, the targeted enhancement chain A can be constructed using one of the following: deep capsule network, deep residual network, fully connected network, and convolutional neural network.

[0036] The distribution mixing layer M probabilistically mixes the feature statistics of samples between the source domain and the enhancement domain, outputting mixed domain samples with diverse data distributions.

[0037] Feature extractors F can be constructed using, but are not limited to, deep capsule networks, deep residual networks, fully connected networks, and convolutional neural networks. The input to feature extractor F consists of samples from the source domain, augmentation domain, and hybrid domain. It maps the input to a high-level feature space and outputs high-level latent features.

[0038] The fault classifier C consists of a fully connected layer and a Softmax activation function. It is used to predict the fault category label of high-level latent features. Its input is the high-level latent features extracted by the feature extractor F, and its output is the fault category prediction probability of the high-level latent features.

[0039] A feedforward neural network consisting of a targeted enhancement chain A, a distributed hybrid layer M, a feature extractor F, and a fault classifier C is used to identify the fault category to which high-level hidden features belong.

[0040] In another implementation of the present invention, the targeted enhancement chain consists of a first encoder G1 and a second encoder G2; the input of the first encoder G1 is a source domain sample, and by extracting key information from the input sample, an enhancement domain sample of the same category and the same dimension size corresponding to the input sample is generated; the input of the second encoder G2 is the enhancement domain sample, and by extracting key information from the input sample, an auxiliary domain sample of the same category and the same dimension size corresponding to the input sample is generated.

[0041] In another implementation of the present invention, the distance loss between the source domain sample and the augmentation domain sample is expressed as:

[0042] in, Indicates the infimum; yes and The set of joint distributions; E represents the expected value; Denotes the Frobenius norm; For source domain samples; To enhance the domain samples.

[0043] For example, the Wasserstein distance loss between samples in the source and augmentation domains is maximized, so that the samples in the source and augmentation domains have different data distributions.

[0044] In another implementation of the present invention, the failure semantic information consistency constraint loss applied between the auxiliary domain and the source domain samples is expressed as:

[0045] in, Represents the source domain; Indicates the mean absolute error; For auxiliary domain samples.

[0046] For example, in order to ensure that the augmented domain samples contain rich fault semantic information consistent with the corresponding samples in the source domain while having a different data distribution from the source domain samples, the fault semantic information consistency constraint is indirectly applied to the augmented domain samples. That is, by applying the fault semantic information consistency constraint loss between the auxiliary domain and the source domain samples, it is ensured that the auxiliary domain samples have fault semantic information consistent with the corresponding samples in the source domain.

[0047] Since the auxiliary domain samples are generated from the augmented domain samples via the second encoder G2, it is indirectly ensured that the augmented domain samples possess the same fault semantic information as the corresponding samples in the source domain. This is achieved by directly processing the augmented domain samples... and source domain samples Applying consistency constraints on fault semantic information between samples will lead to enhancement domain samples With source domain samples The data distributions become similar, failing to achieve the goal of differentiating the data distributions of the augmented domain samples from the source domain samples. Therefore, auxiliary domain samples... This problem is overcome by indirectly imposing consistency constraints on fault semantic information.

[0048] In another implementation of the present invention, the hybrid domain sample is represented as:

[0049] in, For input to the blending layer; and These represent the calculations of the mean and standard deviation, respectively. The mean of the characteristic statistics; The standard deviation is denoted as .

[0050] For example, the step of probabilistically randomizing the sample feature statistics between the source domain and the augmentation domain includes: First, from the source domain and Enhanced Domain Small batches of samples randomly selected from the middle and The input that constitutes the distributed mixing layer M .

[0051] Then, exchange and The position is obtained by shuffling within a small batch of samples. .

[0052] Furthermore, based on and Calculate the characteristic statistic of a random mixture of probabilities—the mean. and standard deviation :

[0053]

[0054] in, and These represent the calculations of the mean and standard deviation, respectively. This represents a random parameter within the interval [0,1].

[0055] Finally, the mean of the characteristic statistics after random mixing of probabilities and standard deviation Input applied to the hybrid module To obtain small batches of mixed domain samples .

[0056] In another implementation of the present invention, the source domain sample set is input into the basic fault diagnosis model for model training to obtain the target fault diagnosis model, including: First, the source domain sample set is input into the targeted enhancement chain A, and the enhancement domain and auxiliary domain sample sets are output, maximizing the Wasserstein distance loss between the source domain and the enhancement domain. To optimize the first encoder G1 in the targeted enhancement chain, the generated enhancement domain has a different data distribution from the samples in the source domain, and a fault semantic information consistency constraint loss is applied between the auxiliary domain samples and the source domain samples. To optimize the first encoder G1 and the second encoder G2 in the targeted enhancement chain, ensuring that the auxiliary domain samples have the same fault semantic information as the corresponding samples in the source domain.

[0057] Then, the source domain and enhancement domain samples are input into the distribution mixing layer, and the output domain mixed samples are output.

[0058] Furthermore, the source domain, augmentation domain, and mixed domain samples are input into the feature extractor F to output high-level latent features, and a central loss is applied to these high-level latent features. This optimizes the feature extractor and enhances the generalization ability of high-level latent features. The center loss function aims to reduce the distance between features of the same type of fault to their class center, while increasing the distance between features of different types of faults.

[0059] Finally, the high-level latent features are input into the fault classifier C, and the cross-entropy classification loss is minimized by the cross-entropy classification loss of the fault class prediction probability output by the fault classifier C. To optimize the fault classifier C, feature extractor F, and target enhancement chain A, the model is trained under supervision.

[0060] Training stops when the number of training iterations reaches a set value.

[0061] The given optimization algorithm includes, but is not limited to, one of the following: adaptive moment estimation algorithm, stochastic gradient descent, and root mean square propagation algorithm.

[0062] like Figure 3 As shown, the target fault diagnosis model, which is composed of a feature extractor F and a fault classifier C trained from samples in the target domain sample set of industrial equipment, outputs the fault category of the sample.

[0063] The true fault category labels of the samples in the target domain sample set are only used to compare with the fault category labels predicted by the model in order to calculate the fault identification accuracy of the model. The output predicted labels correspond to the fault categories of the samples.

[0064] The core technical problem this invention aims to solve is how to overcome the dilemma between "data distribution diversity" and "fault semantic information integrity" in single-source-domain fault diagnosis. Existing domain generalization methods experience a significant performance drop when the number of source domains is insufficient; while simple data augmentation methods either fail to generate samples with significant distributional differences or damage the key fault semantic information of the samples during the generation process, resulting in limited improvement in the generalization ability of the diagnostic model.

[0065] This invention brings significant technological advancements by introducing core modules such as the "targeted enhancement chain" and the "distributed hybrid layer." First, the targeted enhancement chain, through its unique dual-encoder structure and in conjunction with Wasserstein distance loss and indirect semantic constraints, successfully resolves the contradiction between "distribution diversity" and "fault semantic integrity" in single-source domain generalization, generating enhanced samples that maintain clear fault patterns while exhibiting significant distributional differences. Second, the distributed hybrid layer further synthesizes hybrid domain samples with richer data distributions by probabilistically and randomly mixing the feature statistics of source and enhanced domain samples, effectively expanding the data distribution range for model training. Finally, these technical features work synergistically, enabling the model to extract more generalized fault features from the generated diverse samples, significantly improving the accuracy and robustness of the diagnostic model under unknown target operating conditions. This invention effectively overcomes the dependence of existing methods on multi-source domain data, achieving accurate fault diagnosis of industrial equipment under unknown target operating conditions with only single-source domain labeled data, providing a practical, highly generalized intelligent diagnostic solution for industrial equipment that can only obtain data for a single operating condition.

[0066] Example 1 Taking a reciprocating compressor vibration dataset as an example, the experiment selected four states from the dataset: normal (H), intake valve wear (I), exhaust valve wear (O), and exhaust valve spring failure (S), with fault category labels ranging from 0 to 3. The experiment included four different compressor spindle speed operating conditions, as shown in Table 1. The sample set for each operating condition constituted a domain, with 260 samples for each category under each condition. One domain from Table 1 was used as the source domain training sample set, and the other domains were used as the target domain validation sample sets for training and validating the fault diagnosis model.

[0067] Table 1 Description of the experimental bearing dataset

[0068] Step (1): Collect vibration signals from key parts of industrial equipment and perform interception and normalization processing. Use the data collected under one working condition as the source domain sample set and the data collected under other working conditions as the target domain sample set.

[0069] The collected data was truncated to a uniform sample length of 4096 data points and normalized to [-1, 1]. The preprocessed samples were then used as the input to the model.

[0070] Step (2): Construct a targeted enhancement chain A, which is used to mine fault information from the source domain sample set and generate enhanced domain samples with diverse data distribution and rich fault semantic information.

[0071] The targeted enhancement chain A consists of a first encoder G1 and a second encoder G2. The first encoder G1 consists of four one-dimensional convolutional layers, each followed by a ReLU activation function, using source domain samples. As input, the generated samples, with an output length of 4096, constitute the augmentation domain sample set. The Wasserstein distance loss between the source and augmentation domain samples is maximized, resulting in different data distributions for the samples in the source and augmentation domains.

[0072] The second encoder G2 consists of four one-dimensional convolutional layers, each followed by a ReLU activation function to enhance the domain samples. The input consists of generated samples of length 4096, which form the auxiliary domain sample set.

[0073] To ensure that the augmented domain samples contain rich fault semantic information consistent with the corresponding samples in the source domain, while having a different data distribution than the source domain samples, a fault semantic information consistency constraint is indirectly applied to the augmented domain samples. That is, by applying a fault semantic information consistency constraint loss between the auxiliary domain and the source domain samples, it is ensured that the auxiliary domain samples have fault semantic information consistent with the corresponding samples in the source domain.

[0074] Step (3): Construct a distribution mixing layer M, which randomly mixes the feature statistics of the enhancement domain and the source domain samples with probability to synthesize a mixed domain with diverse data distributions.

[0075] The distribution mixing layer M probabilistically mixes the mean and standard deviation of samples from the source domain and the enhancement domain, resulting in a mixed domain sample with diverse data distribution.

[0076] Step (4): Construct a feature extractor F, which is used to extract high-level latent features from source domain, augmentation domain and hybrid domain samples.

[0077] The feature generator G consists of five one-dimensional convolutional layers, each followed by a ReLU activation function. It takes samples from the source domain, enhancement domain, and hybrid domain as input, and then maps the input samples to a high-level feature space to output high-level latent features.

[0078] Step (5): Construct a fault classifier C, which is used to predict fault category labels of high-level latent features.

[0079] The fault classifier C uses a fully connected network with three layers. The dimensions of the hidden layers are 1024, 256, and 4, respectively. The first two fully connected layers are followed by a ReLU activation function, and the last fully connected layer is followed by a Softmax activation function. The final output of the model is a four-dimensional vector representing the predicted probability of the fault category of the input sample, which represents the predicted probability of belonging to normal (H), intake valve wear (I), exhaust valve wear (O), and exhaust valve spring failure (S).

[0080] Step (6): Combine the targeted enhancement chain A, the distributed hybrid layer M, the feature extractor F, and the fault classifier C into a feedforward neural network to construct a fault diagnosis training model, such as... Figure 2 As shown.

[0081] Step (7): Input the industrial equipment source domain sample set into the constructed basic fault diagnosis model, and train the model according to the optimization algorithm and loss function.

[0082] The model's objective loss function includes the optimized training loss for the fault classifier C, feature extractor F, and the first encoder G1 and second encoder G2 in the target enhancement chain A. The distributed mixing layer M contains no trainable parameters and requires no optimization.

[0083] The optimization algorithm uses the root mean square propagation algorithm, with a learning rate of 0.01. Model training is stopped after 180 iterations.

[0084] Step (8): Input the samples from the target domain sample set of industrial equipment into the fault diagnosis verification model composed of the trained feature extractor and fault classifier, and output the fault category of the sample.

[0085] The diagnostic accuracy results of the method of this invention, using operating condition A as the source domain and operating conditions B, C, and D as the target domains, are shown in Table 2. It can be seen that the diagnostic accuracy in all three diagnostic tasks exceeds 92%, indicating that using only a single source domain for training the fault diagnosis model constructed in this invention can be effectively used for fault diagnosis in the target domain that did not participate in model training. This demonstrates the superiority of this method in single-domain generalized intelligent diagnosis and shows broad prospects for engineering applications.

[0086] Table 2 Fault Diagnosis Results

[0087] In summary, the method of this invention generates augmentation domain samples with different data distributions through a constructed targeted augmentation chain, and achieves both diversified distribution of augmentation domain samples and high-fidelity fault information by constraining the consistency of fault semantic information between the auxiliary domain and the source domain. The constructed distribution mixing layer probabilistically and randomly mixes the feature statistics of the source domain and the augmentation domain, creating mixed domain samples with richer data distribution. By utilizing the center loss algorithm to learn generalizable domain-invariant features from the source domain, augmentation domain, and mixed domain, the generalization ability and accuracy of the model are improved, making it applicable to fault diagnosis of industrial equipment under unknown target operating conditions that were not involved in model training.

[0088] Another aspect of the present invention provides a single-source domain generalized intelligent diagnostic system based on targeted data augmentation, comprising: Data acquisition module: Under different operating conditions, it collects vibration signals from key parts of industrial equipment for status monitoring and performs data preprocessing, which includes signal interception and normalization.

[0089] Dataset partitioning module: Divides the state monitoring vibration signal into a source domain sample set and a target domain sample set.

[0090] Model training module: Constructs a basic fault diagnosis model, which includes a targeted enhancement chain, a distributed hybrid layer, a feature extractor, and a fault classifier; inputs the source domain sample set into the basic fault diagnosis model for model training to obtain the target fault diagnosis model.

[0091] Result output module: Input the target domain sample set into the target fault diagnosis model and output the fault category of the sample.

[0092] This invention presents a single-source-domain generalized intelligent diagnostic system based on targeted data augmentation. It constructs a targeted augmentation chain and utilizes a dual-encoder structure to generate augmented samples with diverse distributions and complete fault information. Further, a distributed hybrid layer is designed to synthesize more abundant hybrid domain samples by probabilistically and randomly mixing the feature statistics of source and augmented domain samples. Finally, a feature extractor and a fault classifier are combined to construct a diagnostic model capable of learning strongly generalized fault features from a single source domain. This invention requires only labeled data under a single operating condition for training, significantly improving the model's fault identification accuracy and robustness under unknown target operating conditions. It effectively solves the problem of poor generalization of intelligent diagnostic models caused by the difficulty in obtaining multi-condition data in industrial settings.

[0093] In another aspect of the present invention, the electronic device includes: a processor, a memory, and a communication bus and a communication interface.

[0094] in: The processor, memory, and communication interface communicate with each other via a communication bus.

[0095] A communication interface is used to communicate with other electronic devices or servers.

[0096] The processor is used to execute programs, specifically, to perform any of the steps of the single-source domain generalized intelligent diagnostic method based on targeted data augmentation in the above embodiments.

[0097] Specifically, the program may include program code, which includes computer operation instructions.

[0098] The processor may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application. The one or more processors included in the smart device may be processors of the same type, such as one or more CPUs; or they may be processors of different types, such as one or more CPUs and one or more ASICs.

[0099] Memory is used to store programs. Memory may include high-speed RAM, and may also include non-volatile memory, such as at least one disk drive.

[0100] Specifically, the program can be used to cause the processor to execute the steps of any of the single-source domain generalized intelligent diagnostic methods based on targeted data augmentation described in the embodiments. The specific implementation of each step in the program can be found in the corresponding descriptions of the steps and units executed by any of the single-source domain generalized intelligent diagnostic methods based on targeted data augmentation described above, and will not be repeated here. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the devices and modules described above can be referred to the corresponding process descriptions in the foregoing method embodiments.

[0101] An exemplary embodiment of this application also provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause a computer to perform the methods of various embodiments of this application.

[0102] The methods described above according to embodiments of the present invention can be implemented in hardware, firmware, or as software or computer code that can be stored in a recording medium (such as a CD-ROM, RAM, floppy disk, hard disk, or magneto-optical disk), or as computer code originally stored on a remote recording medium or a non-transitory machine-readable medium and subsequently stored on a local recording medium, downloaded via a network. Thus, the methods described herein can be processed by software stored on a recording medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware (such as an ASIC or FPGA). It is understood that the computer, processor, microprocessor controller, or programmable hardware includes storage components (e.g., RAM, ROM, flash memory, etc.) capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods described herein. Furthermore, when a general-purpose computer accesses code used to implement the methods shown herein, the execution of the code transforms the general-purpose computer into a dedicated computer for executing the methods shown herein.

[0103] Specific embodiments of the present invention have now been described. Other embodiments are within the scope of the appended claims. In some cases, the actions described in the claims can be performed in a different order and still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result.

[0104] It should be noted that all directional indications (such as up, down, left, right, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship between the components in a certain order (as shown in the figure). If the specific order changes, the directional indication will also change accordingly.

[0105] In the description of this invention, the terms "first" and "second" are used only for convenience in describing different components or names, and should not be construed as indicating or implying a sequential relationship, relative importance, or implicitly specifying the number of technical features indicated. Thus, a feature defined with "first" and "second" may explicitly or implicitly include at least one of that feature.

[0106] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention.

[0107] It should be noted that although specific embodiments of the present invention have been described in detail with reference to the accompanying drawings, this should not be construed as limiting the scope of protection of the present invention. Various modifications and variations that can be made by those skilled in the art without inventive effort within the scope described in the claims still fall within the scope of protection of the present invention.

[0108] The examples of the embodiments of the present invention are intended to concisely illustrate the technical features of the embodiments of the present invention, so that those skilled in the art can intuitively understand the technical features of the embodiments of the present invention, and are not intended to be an improper limitation of the embodiments of the present invention.

[0109] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A single-source domain generalized intelligent diagnostic method based on targeted data augmentation, characterized in that, include: Under different operating conditions, vibration signals of key parts of industrial equipment are collected for condition monitoring and data preprocessing is performed, including signal interception and normalization. The state monitoring vibration signal is divided into a source domain sample set and a target domain sample set; A basic fault diagnosis model is constructed, which includes a targeted enhancement chain, a distributed hybrid layer, a feature extractor, and a fault classifier; The source domain sample set is input into the basic fault diagnosis model for model training to obtain the target fault diagnosis model; Input the target domain sample set into the target fault diagnosis model and output the fault category of the sample.

2. The method according to claim 1, characterized in that, The step of dividing the state monitoring vibration signal into a source domain sample set and a target domain sample set includes: The vibration signals collected under one type of working condition are labeled with fault category tags to serve as the source domain sample set. Vibration signals collected under other operating conditions are used as the target domain sample set.

3. The method according to claim 1, characterized in that, The construction of the basic fault diagnosis model includes a targeted enhancement chain, a distributed hybrid layer, a feature extractor, and a fault classifier, comprising: Construct a targeted enhancement chain, which is used to mine fault information from the source domain sample set and generate enhanced domain samples with diverse data distribution and rich fault semantic information. A distribution mixing layer is constructed, which is used to probabilistically and randomly mix the feature statistics of the enhanced domain sample and the source domain sample to synthesize a mixed domain sample with diverse data distributions; Construct a feature extractor for extracting high-level latent features from source domain, augmentation domain, and hybrid domain samples; Construct a fault classifier that predicts fault category labels for high-level latent features; The targeted enhancement chain, the distributed hybrid layer, the feature extractor, and the fault classifier are combined into a feedforward neural network to construct a basic fault diagnosis model.

4. The method according to claim 3, characterized in that, The targeted enhancement chain consists of a first encoder G1 and a second encoder G2; The input of the first encoder G1 is a source domain sample. By extracting key information from the input sample, an augmented domain sample of the same category and the same dimension size that corresponds one-to-one with the input sample is generated. The input to the second encoder G2 is the enhanced domain sample. By extracting key information from the input sample, auxiliary domain samples of the same category and the same dimension size that correspond one-to-one with the input sample are generated.

5. The method according to claim 4, characterized in that, The distance loss between the source domain samples and the augmentation domain samples is expressed as: in, Indicates the infimum; yes and The set of joint distributions; E represents the expected value; Denotes the Frobenius norm; For source domain samples; To enhance the domain samples.

6. The method according to claim 4, characterized in that, The failure semantic information consistency constraint loss applied between the auxiliary domain and the source domain samples is expressed as: in, Represents the source domain; Indicates the mean absolute error; For auxiliary domain samples.

7. The method according to claim 3, characterized in that, The hybrid domain sample is represented as follows: in, For input to the blending layer; and These represent the calculations of the mean and standard deviation, respectively. The mean of the characteristic statistics; The standard deviation is denoted as .

8. A single-source domain generalized intelligent diagnostic system based on targeted data augmentation, characterized in that, include: Data acquisition module: Under different operating conditions, it collects vibration signals from key parts of industrial equipment for status monitoring and performs data preprocessing, including signal interception and normalization. Dataset partitioning module: Divides the state monitoring vibration signal into a source domain sample set and a target domain sample set; Model training module: Constructs a basic fault diagnosis model, which includes a targeted enhancement chain, a distributed mixing layer, a feature extractor, and a fault classifier; inputs the source domain sample set into the basic fault diagnosis model for model training to obtain the target fault diagnosis model; Result output module: Input the target domain sample set into the target fault diagnosis model and output the fault category of the sample.

9. An electronic device, characterized in that, include: The memory, the processor, and the computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the single-source domain generalized intelligent diagnostic method based on targeted data augmentation as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that, The computer storage medium stores a computer program, which, when executed by a processor, implements the steps of the single-source domain generalized intelligent diagnostic method based on targeted data augmentation as described in any one of claims 1 to 7.

Images