Frequency-space joint decoupling interpretable fault diagnosis method and device and storage medium

Abstract

The invention discloses a frequency-space joint decoupling interpretable fault diagnosis method and device and a storage medium, and belongs to the technical field of mechanical equipment intelligent operation and maintenance application, and the method comprises the steps: inputting a state monitoring data sample of a to-be-diagnosed machine into a pre-trained dual-domain collaborative decoupling interpretable domain generalization diagnosis model, decoupling of fault features and multi-source interference components of a state monitoring data sample to be diagnosed is achieved through the frequency domain decoupling network layer, decoupling of fault features and working condition features of the state monitoring data sample of a machine to be diagnosed is achieved through the space domain decoupling network layer, and dual-decoupling fault features are obtained; and inputting the double-decoupled fault features into a fault classifier to obtain fault category labels corresponding to multiple fault types, and determining a fault type diagnosis result according to the fault category labels. The method can solve the technical problem that in the prior art, physical interpretable feature extraction is neglected in fault and working condition feature extraction, and the reliability of a diagnosis result is limited.

Description

Technical Field

[0001] This invention relates to the field of intelligent operation and maintenance application technology for mechanical equipment, and in particular to a frequency-space joint decoupling interpretable fault diagnosis method, device and storage medium. Background Technology

[0002] The development of mechanical fault diagnosis technology is of great significance for ensuring the safe and stable operation of mechanical equipment and improving production efficiency and economic benefits. With the rapid development of the Industrial Internet of Things, big data, and GPU parallel computing capabilities, traditional signal processing-based fault diagnosis methods are struggling to cope with massive, high-dimensional, and multi-source mechanical equipment data, thus giving rise to the widespread application of deep learning in this field. Deep learning has demonstrated outstanding performance in mechanical fault diagnosis, avoiding the tediousness and biases of manual feature extraction; through structures such as convolution, recursion, and attention mechanisms, the model can maintain high diagnostic accuracy in strong noise backgrounds and has a natural adaptability to multi-sensor fusion information; furthermore, thanks to large-scale labeled data, deep networks can achieve near real-time online monitoring and remaining life prediction under fixed operating conditions.

[0003] However, industrial site conditions are complex and ever-changing, with equipment models, loads, speeds, lubrication conditions, and even ambient temperature and humidity constantly changing. This leads to significant differences in the distribution of training and testing data, resulting in deficiencies in the ability to "shift domains": it is highly dependent on a large amount of balanced labeled data, while actual fault samples are scarce and labeling costs are high; when the distribution of the source domain and the target domain differs too much, the model performance drops sharply, and the ability to transfer data across domains is weak.

[0004] As the operating environment of industrial equipment becomes increasingly complex and diverse, in order to enable fault diagnosis models to adapt to unknown environments and overcome the problem that the diagnostic performance of traditional deep learning methods usually degrades severely under varying operating conditions, and to improve the diagnostic robustness and generalization ability of models in unknown or changing environments, domain generalization fault diagnosis methods in transfer learning are often adopted.

[0005] In recent years, although domain generalization fault diagnosis methods can effectively identify mechanical faults under unknown target operating conditions, existing models usually operate in a "black box" manner, and their decision-making logic lacks transparency, which limits the credibility of the diagnostic results and lacks interpretability.

[0006] Current research indicates that neglecting the extraction of physically interpretable features in fault and operating condition feature extraction can negatively impact the model's knowledge learning performance, fault diagnosis performance, and the reliability of diagnostic results. Therefore, existing domain-generalized fault diagnosis methods suffer from the following drawbacks: 1) lack of feature interpretability leads to low reliability; 2) fault feature extraction is easily affected by operating condition coupling and exhibits poor cross-condition generalization; 3) low fault diagnosis accuracy.

[0007] Therefore, there is an urgent need for a frequency-space joint decoupling interpretable fault diagnosis method, device, and storage medium to solve the above-mentioned technical problems. Summary of the Invention

[0008] The purpose of this invention is to overcome the shortcomings of the prior art and provide a frequency-space joint decoupling interpretable fault diagnosis method, device and storage medium, which can solve the technical problem that the prior art ignores the extraction of physically interpretable features in the extraction of fault and operating condition features, thus limiting the reliability of the diagnosis results.

[0009] To achieve the above objectives, the present invention is implemented using the following technical solution:

[0010] In a first aspect, the present invention provides a frequency-space joint decoupling interpretable fault diagnosis method, comprising:

[0011] Obtain a sample of condition monitoring data for the machine to be diagnosed;

[0012] Input the condition monitoring data sample of the machine to be diagnosed into a pre-trained dual-domain collaborative decoupled interpretable domain generalized diagnostic model to obtain the mechanical fault type diagnosis result.

[0013] The dual-domain collaborative decoupling interpretable domain generalized diagnostic model includes a frequency domain decoupling network layer and a spatial domain decoupling diagnostic network. The spatial domain decoupling diagnostic network includes a spatial domain decoupling network layer and a fault classifier. The data processing procedure includes:

[0014] The frequency domain decoupling network layer decouples the fault characteristics and multi-source interference components of the condition monitoring data sample of the machine to be diagnosed. The spatial domain decoupling network layer decouples the fault characteristics and operating condition characteristics of the condition monitoring data sample of the machine to be diagnosed, resulting in dual decoupled fault characteristics.

[0015] The dual-decoupled fault features are input into the fault classifier to obtain fault category labels corresponding to multiple fault types, and the fault type diagnosis result is determined based on the fault category labels.

[0016] The training process of the dual-domain collaborative decoupling interpretable domain generalized diagnostic model includes:

[0017] Acquire samples of condition monitoring data for mechanical equipment;

[0018] The state monitoring data samples are preprocessed to obtain a set of source domain samples with multiple labels;

[0019] The source domain sample set is input into a pre-constructed dual-domain collaborative decoupling interpretable domain generalization diagnostic model, and the frequency-space dual-domain collaborative decoupling optimization strategy is executed to minimize the overall objective loss function according to the optimization algorithm. The training continues until a preset number of iterations are completed to determine the well-trained dual-domain collaborative decoupling interpretable domain generalized diagnostic model.

[0020] Furthermore, the expression for the frequency domain decoupling network layer is as follows;

[0021] ;

[0022] ;

[0023] In the formula, It is the first The frequency of the filtered output of each filter kernel High-dimensional features, It is the first Kernel functions of a variational kernel, It is the first The frequency of the filter output of each filter kernel High-dimensional features, This refers to the Fourier transform of the input samples. It is a bandwidth balancing parameter. It is the center frequency;

[0024] The spatial domain decoupling diagnostic network includes a fault feature extractor. Operating condition feature extractor Fault classifier Domain classifier And a gradient flip layer, used to decouple fault features from operating condition information and extract generalization features;

[0025] The frequency domain decoupling network layer Fault Feature Extractor and the fault classifier This forms the first feedforward neural network, used to predict the fault category label of the input sample;

[0026] The frequency domain decoupling network layer The fault feature extractor Gradient flip layer and the domain classifier This forms a second feedforward neural network used to predict the domain labels of the input samples;

[0027] The frequency domain decoupling network layer The working condition feature extractor Gradient flip layer and the fault classifier This forms a third feedforward neural network, used to predict the fault category label of the input sample;

[0028] The frequency domain decoupling network layer Operating condition feature extractor and the domain classifier This forms the fourth feedforward neural network, used to predict the domain labels of input samples.

[0029] Furthermore, the overall objective loss function Including frequency domain decoupling loss function Decoupling loss function in spatial domain The frequency domain decoupling loss function Including bandwidth-constrained loss function and reconstruction constraint loss function The spatial domain decoupling loss function Including fault classification cross-entropy loss L C Fault characteristic adversarial loss function Domain classification cross-entropy loss Working condition characteristics adversarial loss function and cosine similarity loss function .

[0030] Furthermore, the training process of the dual-domain collaborative decoupling interpretable domain generalized diagnostic model also includes:

[0031] Before implementing the frequency-space dual-domain collaborative decoupling optimization strategy, a sample self-learning strategy is first executed to obtain a discriminant value based on the difference between the initial center frequency and the updated center frequency value.

[0032] ;

[0033] in, For the first The discriminant value of the initial center frequency with an iteration count of 0. For the first An initial center frequency with an iteration count of 0. for The updated center frequency is obtained after one iteration based on the iterative formula;

[0034] The expression for updating the center frequency includes:

[0035] ;

[0036] in, For bandwidth balancing parameters, For input samples The frequency domain representation, Indicates the ordinal number of the input sample channel. Indicates the total number of channels. For frequency parameters;

[0037] If the The discriminant value for the iteration number 0 is greater than the first. If the number of iterations is 0, then the discriminant value will be... Considering a potential center frequency value, all potential center frequencies are obtained based on the frequencies of all inflection points from positive to negative discriminant values. These potential center frequencies are then ordered and a moving average is calculated using a sliding period to obtain all initial training values ​​for the center frequency parameters of the frequency domain decoupling network layer. These initial values ​​are used to output frequency-related parameters. High-dimensional features.

[0038] Furthermore, the frequency-space dual-domain collaborative decoupling optimization strategy includes:

[0039] The overall objective loss function The expression is:

[0040] ;

[0041] in, It is the first balance coefficient;

[0042] The frequency domain decoupling loss function The expression is:

[0043] ;

[0044] This is used to give the feature maps obtained by the variational kernel network layer a clear frequency domain physical meaning, and to reduce aliasing and avoid redundancy between filter kernels;

[0045] The bandwidth constraint loss function is calculated based on the bandwidth balance parameters of the variational filter kernel. The expression is:

[0046] ;

[0047] in, This represents the variational filter kernel ordinal number, and V represents the total number of variational filter kernels. For the first Bandwidth balancing parameters for each variational filter core;

[0048] The reconstruction constraint loss function is calculated based on the high-dimensional features that have removed multi-source interference and the Fourier transform of the input samples. The expression is:

[0049] ;

[0050] in, It is the sum of the results of V variational kernel filtering. It is the Fourier transform of the input sample. It is the first The frequency of the filter output of each filter kernel High-dimensional features;

[0051] The spatial domain decoupling loss function The expression is:

[0052] ;

[0053] in, It is the second balance coefficient;

[0054] Construct a fault feature adversarial loss function based on fault prediction probability and domain label. The fault feature adversarial loss function The expression is:

[0055] ;

[0056] in, , and These are fault feature extractors Fault classifier Sum Domain Classifier Model parameters, Represents the source field ordinal number. Indicates the total number of source domains. Indicates the sample ordinal number. Indicates the first The total number of samples in each source domain. For the first The first in the source domain One sample, It refers to the first The first in the source domain Domain labels for each sample, refers to samples After passing through the fault feature extractor Sum Domain Classifier The obtained fault prediction probability;

[0057] Construct an adversarial loss function based on domain prediction probability and fault category label. The working condition feature adversarial loss function The expression is:

[0058] ;

[0059] in, It is a working condition feature extractor Model parameters, yes Fault category labels, It is a sample After passing through the working condition feature extractor The domain prediction probability obtained from the fault classifier C;

[0060] A cosine similarity loss function is constructed based on the feature vectors extracted by the fault feature extractor and the feature vectors extracted by the domain feature extractor. The cosine similarity loss function The expression is:

[0061] ;

[0062] in, It is the transpose operator. Through fault feature extractor From the sample The feature vector extracted from it, Through the working condition feature extractor From the sample Feature vectors extracted from;

[0063] Construct a fault classification cross-entropy loss function based on fault category labels. The fault classification cross-entropy loss function The expressions include:

[0064] ;

[0065] Construct a cross-entropy loss function for operating condition domain classification based on domain labels. The working condition domain classification cross-entropy loss function The expressions include:

[0066] .

[0067] Further, the preprocessing of the status monitoring data samples includes:

[0068] The condition monitoring data sample is truncated into data samples of the same length and normalized to obtain the processed data sample. The sample set is divided according to the different operating conditions of the mechanical equipment. The multiple fault type samples under the same operating conditions in the processed data sample are divided into a domain to obtain multiple source domain sample sets marked with fault types.

[0069] Among them, the mechanical fault types contained in different domains are the same.

[0070] Furthermore, the optimization algorithm is one of the following: adaptive moment estimation algorithm, stochastic gradient descent algorithm, and root mean square propagation algorithm.

[0071] Secondly, the present invention provides a frequency-space joint decoupling interpretable fault diagnosis device, comprising:

[0072] The sample acquisition module is used to acquire condition monitoring data samples of the machinery to be diagnosed.

[0073] The diagnostic module is used to input the condition monitoring data samples of the machine to be diagnosed into a pre-trained dual-domain collaborative decoupled interpretable domain generalized diagnostic model to obtain the mechanical fault type diagnosis result;

[0074] The dual-domain collaborative decoupling interpretable domain generalized diagnostic model includes a frequency domain decoupling network layer and a spatial domain decoupling diagnostic network. The spatial domain decoupling diagnostic network includes a spatial domain decoupling network layer and a fault classifier. The data processing procedure includes:

[0075] The frequency domain decoupling network layer decouples the fault characteristics and multi-source interference components of the condition monitoring data sample of the machine to be diagnosed. The spatial domain decoupling network layer decouples the fault characteristics and operating condition characteristics of the condition monitoring data sample of the machine to be diagnosed, resulting in dual decoupled fault characteristics.

[0076] The dual-decoupled fault features are input into the fault classifier to obtain fault category labels corresponding to multiple fault types, and the fault type diagnosis result is determined based on the fault category labels.

[0077] The training process of the dual-domain collaborative decoupling interpretable domain generalized diagnostic model includes:

[0078] Acquire samples of condition monitoring data for mechanical equipment;

[0079] The state monitoring data samples are preprocessed to obtain a set of source domain samples with multiple labels;

[0080] The source domain sample set is input into a pre-constructed dual-domain collaborative decoupling interpretable domain generalization diagnostic model, and the frequency-space dual-domain collaborative decoupling optimization strategy is executed to minimize the overall objective loss function according to the optimization algorithm. The training continues until a preset number of iterations are completed to determine the well-trained dual-domain collaborative decoupling interpretable domain generalized diagnostic model.

[0081] Thirdly, the present invention provides an electronic terminal, including a processor and a memory connected to the processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the steps of the method described in any of the preceding claims are performed.

[0082] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the methods described above.

[0083] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0084] This invention proposes a frequency-space joint decoupling interpretable fault diagnosis method. It comprehensively considers both the frequency and spatial domains and designs a frequency-space dual-domain collaborative optimization strategy. The frequency domain decoupling network layer and the spatial domain decoupling diagnosis network are collaboratively optimized to achieve dual decoupling of fault features from multi-source interference components and operating condition information, enabling the extraction of pure fault features.

[0085] The frequency domain decoupling network layer driven by sample self-learning constructed in this invention has a designed frequency domain decoupling loss function that can extract physically interpretable features. It makes full use of the sample self-learning strategy to provide the network layer with more clearly physical initial values ​​for training parameters. The diagnostic results are easy to correspond to mechanical mechanisms, which improves the trust of operation and maintenance personnel in the model.

[0086] The model training of this invention does not require the participation of the target domain dataset, and it overcomes the shortcomings of traditional domain generalization diagnostic methods, such as the susceptibility of fault feature extraction to interference from working conditions and poor generalization across working conditions. It achieves online and effective fault diagnosis of mechanical equipment and improves accuracy. Attached Figure Description

[0087] Figure 1 This is a flowchart of an interpretable fault diagnosis method for frequency-space joint decoupling provided in the embodiments;

[0088] Figure 2 This is a schematic diagram of the structure of a frequency-space joint decoupling interpretable fault diagnosis model provided in an embodiment of the present invention;

[0089] Figure 3 An embodiment of the present invention provides a method for... Figure 2 A schematic diagram of the structure of applying a pre-trained model to a target domain sample set that has not participated in training for fault diagnosis under working conditions.

[0090] Figure 4 A schematic diagram of the confusion matrix of the target domain bearing health status predicted by the frequency-space joint decoupling interpretable fault diagnosis model provided in the embodiment of the present invention in an example;

[0091] Figure 5 This is a block diagram of a frequency-space joint decoupling interpretable fault diagnosis system provided in the embodiment. Detailed Implementation

[0092] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments and specific features in the embodiments are detailed descriptions of the technical solution of the present application, rather than limitations thereof. In the absence of conflict, the embodiments and technical features in the embodiments can be combined with each other.

[0093] In this invention, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent three cases: A alone, A and B together, or B alone. Additionally, in this invention, the character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0094] Example 1:

[0095] Figure 1 This is a flowchart of the frequency-space joint decoupling interpretable fault diagnosis method in Embodiment 1 of the present invention. This flowchart only illustrates the logical sequence of the methods described in this embodiment. Provided there are no conflicts, different methods may be used in other possible embodiments of the present invention. Figure 1 Complete the steps shown or described in the order indicated.

[0096] The frequency-space joint decoupling interpretable fault diagnosis method provided in this embodiment can be applied to a terminal and can be executed by a mechanical equipment fault identification device. This device can be implemented in software and / or hardware and can be integrated into the terminal, such as any smartphone, tablet, or computer device with communication capabilities. The method in this embodiment specifically includes the following steps:

[0097] Step 1: Obtain a sample of condition monitoring data for the machine to be diagnosed;

[0098] Step 2: Input the condition monitoring data sample of the machine to be diagnosed into the pre-trained dual-domain collaborative decoupling interpretable domain generalized diagnostic model. The dual-domain collaborative decoupling interpretable domain generalized diagnostic model includes a frequency domain decoupling network layer and a spatial domain decoupling diagnostic network. The spatial domain decoupling diagnostic network includes a spatial domain decoupling network layer and a fault classifier.

[0099] The expression for the frequency domain decoupling network layer is as follows;

[0100] ;

[0101] ;

[0102] In the formula, It is the first The frequency of the filter output of each filter kernel The high-dimensional features, i.e., the high-dimensional features after removing multi-source interference. It is the first Kernel functions of a variational kernel, It is the first The frequency of the filter output of each filter kernel High-dimensional features, This refers to the input sample (as mentioned later). Fourier transform of ) It is a bandwidth balancing parameter. It is the center frequency, used to decouple fault characteristics from multi-source interference components and extract interpretable features;

[0103] The spatial domain decoupling diagnostic network includes a fault feature extractor. Operating condition feature extractor Fault classifier Domain classifier And the Gradient-Reversal-Layer (GRL) is used to decouple fault features from operating condition information and extract generalization features;

[0104] The fault feature extractor The structure includes multiple convolutional layers, pooling layers, batch normalization, and ReLU activation functions; the fault classifier Includes a fully connected layer and a Softmax activation function; the working condition feature extractor The structure includes multiple convolutional layers, pooling layers, batch normalization, and ReLU activation functions; the domain classifier This includes fully connected layers and the Softmax activation function.

[0105] The frequency domain decoupling network layer Fault Feature Extractor and the fault classifier The first feedforward neural network is constructed to predict the fault category label of the input sample. During the training phase, the fault category label is used to calculate the fault classification cross-entropy loss function. Adversarial loss function based on operating conditions Then, training is performed based on the loss function, which is used in the testing phase to determine which type of fault it belongs to;

[0106] The frequency domain decoupling network layer The fault feature extractor Gradient flip layer and the domain classifier A second feedforward neural network is constructed to predict the domain labels of the input samples. During the training phase, the domain labels are used to calculate the domain classification cross-entropy loss function. Counter-fault loss function Then, training is performed based on the loss function, and domain labels are not used in the testing phase.

[0107] The frequency domain decoupling network layer The working condition feature extractor Gradient flip layer and the fault classifier This forms a third feedforward neural network, used to predict the fault category label of the input sample;

[0108] The frequency domain decoupling network layer Operating condition feature extractor and the domain classifier This forms the fourth feedforward neural network, used to predict the domain labels of input samples.

[0109] The four feedforward neural networks mentioned above each have their own functions: the domain feature extractor and domain classifier extract operating condition features; the fault feature extractor and fault classifier extract fault features; and the adversarial training between the domain feature extractor and fault classifier, and the adversarial training between the fault feature extractor and domain classifier, prevent feature confusion and promote decoupling. The second and third feedforward neural networks, in particular, ensure that two features do not overlap or interfere with each other.

[0110] Step 3: The fault characteristics and multi-source interference components of the condition monitoring data sample of the machine to be diagnosed are decoupled through the frequency domain decoupling network layer, and the fault characteristics and operating condition characteristics of the condition monitoring data sample of the machine to be diagnosed are decoupled through the spatial domain decoupling network layer, resulting in double decoupled fault characteristics.

[0111] Step 4: Input the dual decoupled fault features into the fault classifier to obtain fault category labels corresponding to multiple fault types, and determine the fault type diagnosis result based on the fault category labels.

[0112] Specifically, the training process of the dual-domain collaborative decoupling interpretable domain generalized diagnostic model includes:

[0113] Acquire samples of condition monitoring data for mechanical equipment;

[0114] The condition monitoring data samples are preprocessed as follows:

[0115] The condition monitoring data samples are truncated into data samples of equal length and normalized to obtain processed data samples. The sample sets are then divided according to different operating conditions of the mechanical equipment. Multiple fault type samples under the same operating conditions (e.g., load and speed) within the processed data samples are grouped into a domain, resulting in multiple source domain sample sets labeled with fault types and one unlabeled target domain sample set. The source domain sample sets can be used for input. Figure 2 In the dual-domain collaborative decoupling interpretable domain generalization diagnostic model shown, the target domain sample set can be used as input. Figure 3 The trained fault feature extractor and fault classifier shown;

[0116] Among them, the mechanical fault types contained in different domains are the same.

[0117] The source domain sample set is input into a pre-constructed dual-domain collaborative decoupling interpretable domain generalization diagnostic model. A frequency-space dual-domain collaborative decoupling optimization strategy is executed, and the overall objective loss function is minimized according to an optimization algorithm (including but not limited to adaptive moment estimation, stochastic gradient descent, and root mean square propagation algorithm). :

[0118] Wherein, the overall objective loss function The expression is:

[0119] ;

[0120] in, It is the first balance coefficient;

[0121] The frequency domain decoupling loss function The expression is:

[0122] ;

[0123] This is used to give the feature maps obtained by the variational kernel network layer a clear frequency domain physical meaning, and to reduce aliasing and avoid redundancy between filter kernels;

[0124] The bandwidth constraint loss function is calculated based on the bandwidth balance parameters of the variational filter kernel. The expression is:

[0125] ;

[0126] in, This represents the variational filter kernel ordinal number, and V represents the total number of variational filter kernels. For the first Bandwidth balancing parameters for each variational filter core;

[0127] The reconstruction constraint loss function is calculated based on the high-dimensional features that have removed multi-source interference and the Fourier transform of the input samples. The expression is:

[0128] ;

[0129] in, It is the sum of the results of V variational kernel filtering. It is the Fourier transform of the input sample. It is the first The frequency of the filter output of each filter kernel High-dimensional features;

[0130] The spatial domain decoupling loss function The expression is:

[0131] ;

[0132] in, It is the second balance coefficient;

[0133] Construct a fault feature adversarial loss function based on fault prediction probability and domain label. The fault feature adversarial loss function The expression is:

[0134] ;

[0135] in, , and These are fault feature extractors Fault classifier Sum Domain Classifier Model parameters, Represents the source field ordinal number. Indicates the total number of source domains. Indicates the sample ordinal number. Indicates the first The total number of samples in each source domain. For the first The first in the source domain One sample, It refers to the first The first in the source domain Domain labels for each sample, refers to samples After passing through the fault feature extractor Sum Domain Classifier The obtained fault prediction probability;

[0136] Construct an adversarial loss function based on domain prediction probability and fault category label. The working condition feature adversarial loss function The expression is:

[0137] ;

[0138] in, It is a working condition feature extractor Model parameters, yes Fault category labels, It is a sample After passing through the working condition feature extractor The domain prediction probability obtained from the fault classifier C;

[0139] A cosine similarity loss function is constructed based on the feature vectors extracted by the fault feature extractor and the feature vectors extracted by the domain feature extractor. The cosine similarity loss function The expression is:

[0140] ;

[0141] in, It is the transpose operator. Through fault feature extractor From the sample The feature vector extracted from it, Through the working condition feature extractor From the sample Feature vectors extracted from;

[0142] Construct a fault classification cross-entropy loss function based on fault category labels. The fault classification cross-entropy loss function The expressions include:

[0143] ;

[0144] Construct a cross-entropy loss function for operating condition domain classification based on domain labels. The working condition domain classification cross-entropy loss function The expressions include:

[0145] .

[0146] The training process continues until a preset number of iterations is reached to determine the well-trained dual-domain collaborative decoupling interpretable domain generalized diagnostic model. In this embodiment, the number of iterations can be 150.

[0147] The above summarizes the overall objective loss function. Including frequency domain decoupling loss function Decoupling loss function in spatial domain The frequency domain decoupling loss function Including bandwidth-constrained loss function and reconstruction constraint loss function The spatial domain decoupling loss function Including fault classification cross-entropy loss L C Fault characteristic adversarial loss function Domain classification cross-entropy loss Working condition characteristics adversarial loss function and cosine similarity loss function .

[0148] Furthermore, the training process of the dual-domain collaborative decoupling interpretable domain generalization diagnostic model also includes:

[0149] Before implementing the frequency-space dual-domain collaborative decoupling optimization strategy, a sample self-learning strategy is first executed. This process processes the samples of each fault type in each source domain to obtain a sample set. The discriminant value is then obtained based on the difference between the initial center frequency and the updated center frequency value.

[0150] ;

[0151] in, For the first The discriminant value of the initial center frequency with an iteration count of 0. For the first An initial center frequency with an iteration count of 0. for The updated center frequency is obtained after one iteration based on the iterative formula;

[0152] The expression for updating the center frequency includes:

[0153] ;

[0154] in, For bandwidth balancing parameters, For input samples The frequency domain representation, Indicates the ordinal number of the input sample channel. Indicates the total number of channels. For frequency parameters;

[0155] Here are the input samples that appear in this application. , sample set and the first The first in the source domain Sample Provide an explanation, among which It refers to a sample among all data samples; the sampled sample set is a subset selected from all data samples. It is a subset of samples in the sample set, ranging from large to small.

[0156] Subsequently, based on the initial center frequency and the discrimination value, the initial center frequency for the next iteration can be obtained. The expression is:

[0157] ;

[0158] in, Let be the initial center frequency for the (i+1)th iteration with a number of 0. yes Increase the frequency step size;

[0159] Repeat this process to complete the iteration.

[0160] If the The discriminant value for the iteration number 0 is greater than the first. If the number of iterations is 0, then the discriminant value will be... Considering a potential center frequency value, all potential center frequency values ​​are obtained based on the frequency values ​​of all inflection points from positive to negative values ​​of the discriminant value. These potential center frequency values ​​are then arranged in an ordered manner using a sliding period. (Sliding period is) , The center frequency value is calculated by performing a moving average on the sampling frequency to obtain all the initial training values ​​of the center frequency parameter of the frequency domain decoupling network layer, which is the center frequency in the variational kernel function mentioned above. It is used to provide the parameter values ​​required for training the frequency domain decoupled network layer, and outputs the frequency-related parameters. The high-dimensional characteristics of the frequency domain decoupling network layer give it more interpretability.

[0161] As an example, taking a rolling bearing dataset provided by a university, the preprocessed data was used as the model input. Table 1 lists the details of domains H1~H4 established based on four operating conditions in the rolling bearing dataset. Each state category in the domain has 100 samples, with a sample length of 2048 data points. The experimental dataset contains four fault categories: normal state, inner race fault, outer race fault, and rolling element fault, labeled 0~3. Bearing state data under different operating conditions were collected by adjusting the rotational speed. Four domain generalization fault diagnosis tasks were set up in the rolling bearing experimental dataset to verify the generalization performance of the proposed method, as shown in Table 2. Three fully labeled source domains were used to train the model; the target domain sample set was only used for model testing and not for model training. The average of six diagnostic accuracies was used as the diagnostic result for each task to avoid randomness in the experiment. Five comparative methods were used to analyze the fault diagnosis tasks to demonstrate the effectiveness and superiority of the proposed dual-domain collaborative interpretable domain generalization diagnostic model. Detailed descriptions are provided in Table 3. Tables 1, 2, and 3 are shown below: Table 1 Table 2 Table 3

[0162] Regarding the fault feature extractor mentioned above Using the three labeled source domain sample sets (H1-H4) from different tasks (R1-R4) in Table 2 as example inputs. The output is a high-dimensional feature of length 256; the fault classifier C and the domain classifier D are used as fault feature extractors. The extracted 256-dimensional features are used as input to output fault category labels and domain category labels, respectively. The purpose is to learn cross-domain fault knowledge and extract domain-invariant features through the Domain Adversarial Neural Network (DANN) training scheme.

[0163] Regarding the aforementioned working condition feature extractor Using the three labeled source domain sample sets (H1-H4) from different tasks (R1-R4) in Table 2 as example inputs. The output is a high-dimensional feature of length 256. The fault classifier C and the domain classifier D take the 256-dimensional features extracted by the fault feature extractor as input and output the fault category label and the domain category label respectively. The purpose is to learn cross-domain fault knowledge and extract domain-invariant features through the Domain Adversarial Neural Network (DANN) training scheme.

[0164] Using the three labeled source domain sample sets (H1-H4) from different tasks (R1-R4) in Table 2 as example inputs, these samples are fed into the mechanical equipment dual-domain collaborative decoupling interpretable domain generalized fault diagnosis model. Through a sample self-learning driven frequency domain decoupling network layer, all initial values ​​of the center frequency parameter of the frequency domain decoupling network layer are sampled and calculated. A frequency-space dual-domain collaborative optimization strategy is designed to decouple fault features and operating condition features. Based on the given loss function, the model parameters are optimized by backpropagation using the Adam optimization algorithm. The Adam optimizer weight decay rate is initialized to 0.001, the Betas parameter is (0.99, 0.99), and the learning rate μ and balance coefficient are set. and Set them to 0.0001, 0.1 and 0.01 respectively.

[0165] When implementing the sample self-learning strategy, the source domain sample sets (H1-H4) labeled in the different tasks (R1-R4) in Table 2 are used as example inputs. To overcome the difficulty of extracting weak fault features under complex multi-source interference, the sample self-learning strategy can mine all potential center frequencies of the input samples, which serve as all initial training values ​​for the center frequency parameters of the frequency domain decoupling network layer, used to output frequency-related parameters. High-dimensional features.

[0166] When implementing the frequency-space dual-domain collaborative decoupling optimization strategy, the source domain sample sets (H1-H4) of the three labels in different tasks (R1-R4) in Table 2 are used as example input data. The strategy is considered from both the frequency domain and spatial domain levels, mainly including the frequency domain decoupling network layer and the spatial domain decoupling diagnostic network.

[0167] For example, the confusion matrix of diagnostic results for different methods on the target domain test datasets (H1-H4) for different tasks (R1-R4) in Table 2 of the method of the present invention is as follows: Figure 4 As shown in Table 4, the proposed dual-domain collaborative decoupling interpretable domain generalization diagnostic method achieved the best diagnostic accuracy in all fault diagnosis tasks, with an average diagnostic accuracy of 85.8%. The average standard deviation of the diagnostic accuracy was only 0.8%, indicating the stability of the fault diagnosis model. The proposed dual-domain collaborative decoupling interpretable domain generalization diagnostic method achieved average accuracies 24.7% and 9% higher than the interpretable intelligent methods M2 and M3, respectively, proving that the proposed method is more generalizable while possessing interpretability. The proposed dual-domain collaborative decoupling interpretable domain generalization diagnostic method achieved average accuracies 11.4% and 4% higher than the domain generalization methods M4 and M5, respectively, proving the superiority of the proposed method in domain generalization diagnosis, as shown in Table 4 below. Table 4

[0168] Table 5 lists the accuracy results of the ablation experiments. The results show that removing the frequency domain decoupling network layer has a significant impact on the model performance in the four domain generalized diagnostic tasks. This verifies that the decoupling of fault components by the frequency domain decoupling network layer helps to extract generalized fault features. Table 5 is shown below: Table 5

[0169] Example 2:

[0170] Embodiment 2 of the present invention provides a frequency-space joint decoupling interpretable fault diagnosis device, such as... Figure 5 As shown, it includes:

[0171] The sample acquisition module is used to acquire condition monitoring data samples of the machinery to be diagnosed.

[0172] The diagnostic module is used to input the condition monitoring data samples of the machine to be diagnosed into a pre-trained dual-domain collaborative decoupled interpretable domain generalized diagnostic model to obtain the mechanical fault type diagnosis result;

[0173] The dual-domain collaborative decoupling interpretable domain generalized diagnostic model includes a frequency domain decoupling network layer and a spatial domain decoupling diagnostic network. The spatial domain decoupling diagnostic network includes a spatial domain decoupling network layer and a fault classifier. The data processing procedure includes:

[0174] The frequency domain decoupling network layer decouples the fault characteristics and multi-source interference components of the condition monitoring data sample of the machine to be diagnosed. The spatial domain decoupling network layer decouples the fault characteristics and operating condition characteristics of the condition monitoring data sample of the machine to be diagnosed, resulting in dual decoupled fault characteristics.

[0175] The dual-decoupled fault features are input into the fault classifier to obtain fault category labels corresponding to multiple fault types, and the fault type diagnosis result is determined based on the fault category labels.

[0176] The training process of the dual-domain collaborative decoupling interpretable domain generalized diagnostic model includes:

[0177] Acquire samples of condition monitoring data for mechanical equipment;

[0178] The state monitoring data samples are preprocessed to obtain a set of source domain samples with multiple labels;

[0179] The source domain sample set is input into a pre-constructed dual-domain collaborative decoupling interpretable domain generalization diagnostic model, and the frequency-space dual-domain collaborative decoupling optimization strategy is executed to minimize the overall objective loss function according to the optimization algorithm. The training continues until a preset number of iterations are completed to determine the well-trained dual-domain collaborative decoupling interpretable domain generalized diagnostic model.

[0180] The frequency-space joint decoupling interpretable fault diagnosis device provided in Embodiment 2 of the present invention can execute the frequency-space joint decoupling interpretable fault diagnosis method provided in Embodiment 1 of the present invention, and has the corresponding functional modules and beneficial effects of the method.

[0181] Example 3:

[0182] Embodiment 3 of the present invention also provides an electronic terminal, including a processor and a memory connected to the processor, wherein a computer program is stored in the memory, and the processor is used to perform operations according to the instructions to execute the steps of the method described in Embodiment 1.

[0183] The electronic terminal provided in Embodiment 3 of the present invention can execute the frequency-space joint decoupling and interpretable fault diagnosis method provided in Embodiment 1 of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0184] Example 4:

[0185] Embodiment 4 of the present invention also provides a computer-readable storage medium storing a computer program thereon. When the computer program is executed by a processor, it implements the steps of the method described in Embodiment 1 and has the corresponding functional modules and beneficial effects of the method.

[0186] Those skilled in the art will understand that embodiments of this application can be provided as methods, apparatus, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0187] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (devices), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0188] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0189] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0190] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A frequency-space joint decoupling interpretable fault diagnosis method, characterized in that, include: Obtain a sample of condition monitoring data for the machine to be diagnosed; Input the condition monitoring data sample of the machine to be diagnosed into a pre-trained dual-domain collaborative decoupled interpretable domain generalized diagnostic model to obtain the mechanical fault type diagnosis result. The dual-domain collaborative decoupling interpretable domain generalized diagnostic model includes a frequency domain decoupling network layer and a spatial domain decoupling diagnostic network. The spatial domain decoupling diagnostic network includes a spatial domain decoupling network layer and a fault classifier. The data processing procedure includes: The frequency domain decoupling network layer decouples the fault characteristics and multi-source interference components of the condition monitoring data sample of the machine to be diagnosed. The spatial domain decoupling network layer decouples the fault characteristics and operating condition characteristics of the condition monitoring data sample of the machine to be diagnosed, resulting in dual decoupled fault characteristics. The dual-decoupled fault features are input into the fault classifier to obtain fault category labels corresponding to multiple fault types, and the fault type diagnosis result is determined based on the fault category labels. The training process of the dual-domain collaborative decoupling interpretable domain generalized diagnostic model includes: Acquire samples of condition monitoring data for mechanical equipment; The state monitoring data samples are preprocessed to obtain a set of source domain samples with multiple labels; The source domain sample set is input into a pre-constructed dual-domain collaborative decoupling interpretable domain generalization diagnostic model, and the frequency-space dual-domain collaborative decoupling optimization strategy is executed to minimize the overall objective loss function according to the optimization algorithm. The training continues until a preset number of iterations are completed to determine the well-trained dual-domain collaborative decoupling interpretable domain generalized diagnostic model.

2. The frequency-space joint decoupling interpretable fault diagnosis method according to claim 1, characterized in that, The expression for the frequency domain decoupling network layer is as follows; ， ， In the formula, It is the first The frequency of the filtered output of each filter kernel High-dimensional features, It is the first Kernel functions of variational kernels, It is the first The frequency of the filtered output of each filter kernel High-dimensional features, This refers to the Fourier transform of the input samples. It is a bandwidth balancing parameter. It is the center frequency; The spatial domain decoupling diagnostic network includes a fault feature extractor. Operating condition feature extractor Fault classifier Domain classifier And a gradient flip layer, used to decouple fault features from operating condition information and extract generalization features; The frequency domain decoupling network layer Fault Feature Extractor and the fault classifier This forms the first feedforward neural network, used to predict the fault category label of the input sample; The frequency domain decoupling network layer The fault feature extractor Gradient flip layer and the domain classifier This forms a second feedforward neural network used to predict the domain labels of the input samples; The frequency domain decoupling network layer The working condition feature extractor Gradient flip layer and the fault classifier This forms a third feedforward neural network, used to predict the fault category label of the input sample; The frequency domain decoupling network layer Operating condition feature extractor and the domain classifier This forms the fourth feedforward neural network, used to predict the domain labels of input samples.

3. The frequency-space joint decoupling interpretable fault diagnosis method according to claim 1, characterized in that, The overall objective loss function Including frequency domain decoupling loss function Decoupling loss function in spatial domain The frequency domain decoupling loss function Including bandwidth-constrained loss function and reconstruction constraint loss function The spatial domain decoupling loss function Including fault classification cross-entropy loss L C Fault characteristic adversarial loss function Domain classification cross-entropy loss Working condition characteristics adversarial loss function and cosine similarity loss function .

4. The frequency-space joint decoupling interpretable fault diagnosis method according to claim 3, characterized in that, The training process of the dual-domain collaborative decoupling interpretable domain generalization diagnostic model also includes: Before implementing the frequency-space dual-domain collaborative decoupling optimization strategy, a sample self-learning strategy is first executed to obtain a discriminant value based on the difference between the initial center frequency and the updated center frequency value. ， in, For the first The discriminant value of the initial center frequency with an iteration count of 0. For the first An initial center frequency with an iteration count of 0. for The updated center frequency is obtained after one iteration based on the iterative formula; The expression for updating the center frequency includes: ， in, For bandwidth balancing parameters, For input samples The frequency domain representation, Indicates the ordinal number of the input sample channel. Indicates the total number of channels. For frequency parameters; If the The discriminant value for the iteration number 0 is greater than the first. If the number of iterations is 0, then the discriminant value will be... Considering a potential center frequency value, all potential center frequencies are obtained based on the frequencies of all inflection points from positive to negative discriminant values. These potential center frequencies are then ordered and a moving average is calculated using a sliding period to obtain all initial training values ​​for the center frequency parameters of the frequency domain decoupling network layer. These initial values ​​are used to output frequency-related parameters. High-dimensional features.

5. The frequency-space joint decoupling interpretable fault diagnosis method according to claim 3, characterized in that, The frequency-space dual-domain collaborative decoupling optimization strategy includes: The overall objective loss function The expression is: ， in, It is the first balance coefficient; The frequency domain decoupling loss function The expression is: ， This is used to give the feature maps obtained by the variational kernel network layer a clear frequency domain physical meaning, and to reduce aliasing and avoid redundancy between filter kernels; The bandwidth constraint loss function is calculated based on the bandwidth balance parameters of the variational filter kernel. The expression is: ， in, This represents the variational filter kernel ordinal number, and V represents the total number of variational filter kernels. For the first Bandwidth balancing parameters for each variational filter core; The reconstruction constraint loss function is calculated based on the high-dimensional features that have removed multi-source interference and the Fourier transform of the input samples. The expression is: ， in, It is the sum of the results of V variational kernel filtering. It is the Fourier transform of the input sample. It is the first The frequency of the filtered output of each filter kernel High-dimensional features; The spatial domain decoupling loss function The expression is: ， in, It is the second balance coefficient; Construct a fault feature adversarial loss function based on fault prediction probability and domain label. The fault feature adversarial loss function The expression is: ， in, , and These are fault feature extractors Fault classifier Sum Domain Classifier Model parameters, Represents the source field ordinal number. Indicates the total number of source domains. Indicates the sample ordinal number. Indicates the first The total number of samples in each source domain. For the first The first in the source domain One sample, It refers to the first The first in the source domain Domain labels for each sample, refers to samples After passing through the fault feature extractor Sum Domain Classifier The obtained fault prediction probability; Construct an adversarial loss function based on domain prediction probability and fault category label. The working condition feature adversarial loss function The expression is: ， in, It is a working condition feature extractor Model parameters, yes Fault category labels, It is a sample After passing through the working condition feature extractor The domain prediction probability obtained from the fault classifier C; A cosine similarity loss function is constructed based on the feature vectors extracted by the fault feature extractor and the feature vectors extracted by the domain feature extractor. The cosine similarity loss function The expression is: ， in, It is the transpose operator. Through fault feature extractor From the sample The feature vector extracted from it, Through the working condition feature extractor From the sample Feature vectors extracted from; Construct a fault classification cross-entropy loss function based on fault category labels. The fault classification cross-entropy loss function The expressions include: ， Construct a cross-entropy loss function for operating condition domain classification based on domain labels. The working condition domain classification cross-entropy loss function The expressions include: 。 6. The frequency-space joint decoupling interpretable fault diagnosis method according to claim 3, characterized in that, Preprocessing the condition monitoring data samples includes: The condition monitoring data sample is truncated into data samples of the same length and normalized to obtain the processed data sample. The sample set is divided according to the different operating conditions of the mechanical equipment. The multiple fault type samples under the same operating conditions in the processed data sample are divided into a domain to obtain multiple source domain sample sets marked with fault types. Among them, the mechanical fault types contained in different domains are the same.

7. The frequency-space joint decoupling interpretable fault diagnosis method according to claim 3, characterized in that, The optimization algorithm is one of the following: adaptive moment estimation algorithm, stochastic gradient descent algorithm, and root mean square transfer algorithm.

8. A frequency-space joint decoupling interpretable fault diagnosis device, characterized in that, include: The sample acquisition module is used to acquire condition monitoring data samples of the machinery to be diagnosed. The diagnostic module is used to input the condition monitoring data samples of the machine to be diagnosed into a pre-trained dual-domain collaborative decoupled interpretable domain generalized diagnostic model to obtain the mechanical fault type diagnosis result; The dual-domain collaborative decoupling interpretable domain generalized diagnostic model includes a frequency domain decoupling network layer and a spatial domain decoupling diagnostic network. The spatial domain decoupling diagnostic network includes a spatial domain decoupling network layer and a fault classifier. The data processing procedure includes: The frequency domain decoupling network layer decouples the fault characteristics and multi-source interference components of the condition monitoring data sample of the machine to be diagnosed. The spatial domain decoupling network layer decouples the fault characteristics and operating condition characteristics of the condition monitoring data sample of the machine to be diagnosed, resulting in dual decoupled fault characteristics. The dual-decoupled fault features are input into the fault classifier to obtain fault category labels corresponding to multiple fault types, and the fault type diagnosis result is determined based on the fault category labels. The training process of the dual-domain collaborative decoupling interpretable domain generalized diagnostic model includes: Acquire samples of condition monitoring data for mechanical equipment; The state monitoring data samples are preprocessed to obtain a set of source domain samples with multiple labels; The source domain sample set is input into a pre-constructed dual-domain collaborative decoupling interpretable domain generalization diagnostic model, and the frequency-space dual-domain collaborative decoupling optimization strategy is executed to minimize the overall objective loss function according to the optimization algorithm. The training continues until a preset number of iterations are completed to determine the well-trained dual-domain collaborative decoupling interpretable domain generalized diagnostic model.

9. An electronic terminal, characterized in that, It includes a processor and a memory connected to the processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, it performs the steps of the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the method according to any one of claims 1 to 7.

Images