Equipment component fault diagnosis method based on noise residual fusion strategy

Abstract

This application discloses a fault diagnosis method for equipment components based on a noise residual fusion strategy, belonging to the field of fault diagnosis technology. The method includes: acquiring noisy condition monitoring data collected from mechanical equipment operating in a noisy environment; performing standardized preprocessing to unify data distribution and eliminate dimensional differences; inputting the preprocessed noisy data into a trained denoising-diagnosis joint model; first, filtering noise through a signal denoising model to obtain denoised data; integrating the noise residual and denoised data into a comprehensive fusion feature through a fusion module; and then using a diagnostic model to complete fault prediction. This application effectively suppresses noise interference through a signal denoising model, avoids the loss of important fault information through a fusion module, and achieves collaborative optimization of denoising and diagnosis through cascaded logic in the joint model. This solves the problems of poor collaborative effect and easy information loss in existing technologies, improving the robustness and accuracy of fault diagnosis in noisy environments, and is applicable to fault monitoring and diagnosis of various industrial equipment.

Description

Technical Field

[0001] This application relates to the field of fault diagnosis technology, and in particular to a fault diagnosis method for equipment components based on a noise residual fusion strategy. Background Technology

[0002] In modern industrial systems, precision machine tools, wind turbines, high-speed trains, and other mechanical equipment occupy a crucial position, and their stable operation directly impacts productivity and economic efficiency. However, these machines often operate under harsh conditions, making core components such as bearings prone to failure, potentially leading to severe economic losses and personal injury. Therefore, achieving real-time monitoring and accurate fault diagnosis of mechanical equipment is of great significance for preventing catastrophic accidents and extending equipment lifespan.

[0003] In real-world industrial scenarios, the condition monitoring signals of mechanical equipment collected by sensors are often accompanied by a large amount of noise. This noise can mask or distort fault characteristic information, making it difficult for fault diagnosis models to accurately identify the health status of the equipment. To improve the robustness of diagnosis in noisy environments, existing technologies have proposed various solutions: some solutions directly learn fault characteristics from noisy signals, but they are difficult to effectively suppress noise interference and have limited performance in complex noisy environments; some solutions first denoise the signal before inputting it into the diagnostic model, but the denoising and diagnostic processes are independent of each other, failing to utilize their correlation, which can easily lead to the loss of important fault-related information and reduce diagnostic accuracy; other solutions use a joint learning strategy to perform denoising and diagnostic tasks simultaneously, but this violates the logical sequence of denoising as preprocessing and diagnosis as a downstream task, resulting in poor synergistic optimization between the two.

[0004] Given the shortcomings of existing technologies, there is an urgent need for a fault diagnosis solution for equipment components that can effectively integrate denoising and diagnostic tasks, avoid the loss of important information, and be applicable to noisy environments. Summary of the Invention

[0005] The purpose of this application is to provide a fault diagnosis method for equipment components based on a noise residual fusion strategy, which can significantly improve the robustness and accuracy of fault diagnosis in noisy scenarios.

[0006] To achieve the above objectives, this application provides the following solution:

[0007] A fault diagnosis method for equipment components based on a noise residual fusion strategy includes the following steps:

[0008] Acquire noisy condition monitoring data; the noisy condition monitoring data is the noisy condition monitoring data collected when the mechanical equipment is running in a noisy environment.

[0009] The noisy state monitoring data is standardized and preprocessed to obtain preprocessed noisy state monitoring data.

[0010] The preprocessed noisy condition monitoring data is input into a trained denoising-diagnosis joint model, which outputs a predicted fault type label. The denoising-diagnosis joint model includes a signal denoising model, a noise residual fusion module, and a fault diagnosis model. The signal denoising model is used to denoise the input data to obtain denoised data. The noise residual fusion module is used to fuse the denoised data and the noise residual to obtain fusion features. The fault diagnosis model is used to predict and output fault type labels based on the fusion features. The noise residual is the difference between the noisy condition monitoring data and the denoised data.

[0011] According to the specific embodiments provided in this application, the following technical effects are disclosed:

[0012] This application provides a fault diagnosis method for equipment components based on a noise residual fusion strategy. In this method, firstly, noisy condition monitoring data is collected from the mechanical equipment during operation in a noisy environment. This accurately captures the signal characteristics of the equipment under actual operating conditions, providing a data source that fits the actual industrial scenario for subsequent fault diagnosis. Then, standardized preprocessing is performed to eliminate differences in data units and the influence of extreme values, unifying the data distribution range and improving the stability and convergence speed of subsequent model processing. Finally, the data is input into a trained denoising-diagnosis joint model, outputting a predicted fault type label. In the trained denoising-diagnosis joint model, firstly... The signal denoising model denoises the input data, effectively suppressing noise interference and providing a cleaner feature base for subsequent fault diagnosis. Then, the noise residual fusion module fuses the denoised data and noise residuals, retaining important fault-related information that might have been filtered out during denoising, resulting in a more comprehensive feature representation. Finally, the fault diagnosis model predicts and outputs fault type labels based on the fused features. The entire joint model, through a cascaded logic of "denoising-fusion-diagnosis," streamlines the relationship between preprocessing and downstream tasks, achieving synergistic optimization of denoising and diagnosis tasks. This significantly improves the robustness and accuracy of fault diagnosis in noisy scenarios, addressing the problem of poor synergistic effects in existing technologies. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is a flowchart of a fault diagnosis method for equipment components based on a noise residual fusion strategy, provided as an embodiment of this application.

[0015] Figure 2 This document provides a flowchart of a method for fault diagnosis of equipment components based on a noise residual fusion strategy, which involves acquiring historical state detection datasets and constructing and training a denoising-diagnosis joint model.

[0016] Figure 3 This is a schematic diagram of the signal denoising model in an equipment component fault diagnosis method based on a noise residual fusion strategy, provided in an embodiment of this application.

[0017] Figure 4 This is a schematic diagram of the structure of a noise residual fusion module in an equipment component fault diagnosis method based on a noise residual fusion strategy, provided in an embodiment of this application.

[0018] Figure 5 This is a schematic diagram of the structure of a fault diagnosis model in an equipment component fault diagnosis method based on a noise residual fusion strategy, provided in an embodiment of this application.

[0019] Figure 6 This is a flowchart of step B6 in a fault diagnosis method for equipment components based on a noise residual fusion strategy, provided as an embodiment of this application. Detailed Implementation

[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0022] This application provides a method for fault diagnosis of equipment components based on a noise residual fusion strategy. In one exemplary embodiment, such as... Figure 1 As shown, it includes the following steps:

[0023] A1. Acquire noisy condition monitoring data; noisy condition monitoring data refers to the condition monitoring data with noise collected when mechanical equipment is running in a noisy environment.

[0024] A2. Standardize and preprocess the noisy condition monitoring data to obtain preprocessed noisy condition monitoring data. This implementation method involves standardizing and preprocessing the monitoring data according to the following formula:

[0025] .

[0026] in, This is the preprocessed status monitoring data. This is the raw condition monitoring data. The average of the monitoring data, The standard deviation of the monitoring data, It is set to a minimum value to prevent the denominator from being 0.

[0027] A3. Input the preprocessed noisy condition monitoring data into the trained denoising-diagnosis joint model, and output the predicted fault type label. The denoising-diagnosis joint model includes a signal denoising model, a noise residual fusion module, and a fault diagnosis model. The signal denoising model is used to denoise the input data to obtain denoised data. The noise residual fusion module is used to fuse the denoised data and the noise residual to obtain fusion features. The fault diagnosis model is used to predict and output the fault type label based on the fusion features. The noise residual is the difference between the noisy condition monitoring data and the denoised data.

[0028] In an exemplary embodiment of this application, before step A3, the method further includes: acquiring a historical state detection dataset, constructing and training a denoising-diagnosis joint model, and obtaining a trained denoising-diagnosis joint model, such as... Figure 2 As shown, the specific steps include:

[0029] B1. Obtain historical condition monitoring dataset; The historical condition monitoring dataset includes several clean condition monitoring data and corresponding fault type labels. Clean condition monitoring data is noise-free condition monitoring data collected during the operation of mechanical equipment.

[0030] B2. For any clean status monitoring data, Gaussian noise is added to the clean status monitoring data according to a preset signal-to-noise ratio to obtain simulated noisy status monitoring data. In this embodiment, the preset signal-to-noise ratio is as follows:

[0031] .

[0032] in, To preset the signal-to-noise ratio, The power of clean status monitoring data, The power of Gaussian noise.

[0033] .

[0034] in, For the first n Cleanliness monitoring data.

[0035] B3. Standardize and preprocess the clean state monitoring data and the simulated noisy state monitoring data respectively to obtain preprocessed clean data and preprocessed noisy data; the preprocessed noisy data, the preprocessed clean data and the corresponding fault type label constitute a sample.

[0036] B4. Based on the signal denoising model, the noise residual fusion module, and the fault diagnosis model, a denoising-diagnosis joint model is constructed. The signal denoising model is an end-to-end fully convolutional time-domain signal denoising model based on dilated convolution. The noise residual fusion module is a fusion module based on a gating mechanism. The fault diagnosis model is a diagnostic model based on convolutional neural networks and bidirectional long short-term memory neural networks.

[0037] Specifically, in this embodiment, such as Figure 3 As shown, the signal denoising model includes an input preprocessing layer, stacked dilated convolutional residual blocks, and an output post-processing layer; the input preprocessing layer is a 1×1 convolutional layer used to map the number of input channels to the number of residual channels; the number of stacked dilated convolutional residual blocks is L, and the... i The expansion factor of the layer is Each dilated convolutional residual block employs a gated activation mechanism and introduces residual connections and skip connections; the output post-processing layer includes two nonlinear transformation layers and a 1×1 convolutional layer for reconstructing clean, denoised data.

[0038] The input preprocessing layer in the signal denoising model uses a 1×1 convolutional layer to enhance feature representation capabilities for noisy input signals. ,in For batch size, For sampling length, The number of input signal channels is determined by the input preprocessing layer according to the following formula: Mapped to residual channel number :

[0039] .

[0040] Number of stacked dilated convolutional residuals in signal denoising models L =10, where the first Layer expansion factor The exponentially increasing expansion rate allows the model to obtain a sufficiently large receptive field. For the i-th layer, a 2×1 dilated convolution is first used to extract features at different scales. Then, to enhance the model's nonlinear expressive power and control the information flow, a gated activation mechanism is used within each residual block, and the convolution output is divided into filtering branches along the channel dimension. With gated branches Features are obtained using gating activation. The above process can be expressed as the following formula:

[0041] .

[0042] .

[0043] .

[0044] in, It is the Sigmoid activation function. This indicates element-wise multiplication. The dilation factor is represented as The size of the nucleus is The dilated convolution is used, where the kernel size is 2×1.

[0045] Furthermore, to accelerate model convergence and alleviate the gradient vanishing problem, the model introduces residual connections and skip connections, which are expressed as follows:

[0046] .

[0047] .

[0048] Finally, the skip connections of all layers are accumulated and aggregated at the end to obtain the final aggregated feature S.

[0049] .

[0050] The output post-processing layer in the signal denoising model contains two nonlinear transformation layers and a 1×1 convolution to restore a clean signal output. As shown in the following formula:

[0051] .

[0052] like Figure 4 As shown, the noise residual fusion module includes a residual feature extraction layer, an adaptive gating mechanism, and a weighted residual reconstruction layer. The residual feature extraction layer is a three-layer 3×1 one-dimensional convolutional network used to extract residual features of deep noise. The adaptive gating mechanism includes a gating network and an SNR perceptual factor used to determine the gating weights and control the degree of fusion of residual features. The weighted residual reconstruction layer is used to fuse the denoised data with the weighted residual features to obtain fused features.

[0053] The residual feature extraction layer in the noise residual fusion module uses a convolutional neural network to extract features, making and These represent the noisy condition monitoring data and the denoised condition monitoring data, respectively, with noise residuals. , noise residual Input into the residual feature extraction network Deep residual features are extracted Residual feature extraction network It consists of a three-layer 3×1 one-dimensional convolutional network. The first two layers are followed by batch normalization, ReLU activation, and Dropout layers. Finally, the Tanh activation function is used to limit the amplitude and prevent excessive noise features from affecting the main signal.

[0054] .

[0055] An adaptive gating mechanism in the noise residual fusion module is used to determine the degree of fusion of noise residual information. First, the denoised signal and residual features are concatenated along the channel dimension, and then a gating network is constructed. Gating network Two 1×1 convolutional layers are used to learn the interaction relationships between features. The ReLU activation function is used between layers, and finally the Sigmoid function is used to generate gating weights with values ​​between (0,1). :

[0056] .

[0057] .

[0058] in, The activation function is Sigmoid. Furthermore, considering that the probability of the residual containing useful information is low under extremely low signal-to-noise ratio conditions, an SNR sensing factor is introduced in this embodiment. and a learnable scaling factor To further control the degree of fusion. SNR sensing factor In the calculation, the model treats the denoised signal as the clean signal, the difference between the input signal and the denoised signal as noise, and uses the power of both signals to obtain the estimated signal-to-noise ratio. To map a wide range of signal-to-noise ratios to (0,1) gating coefficients, a sigmoid transform function with learnable parameters is used for the mapping, and the SNR perceptual factor is... The definition is as follows:

[0059] .

[0060] in, Use the Sigmoid activation function; k It is a scaling parameter used to control the steepness of the Sigmoid function, that is, the sensitivity of the model to changes in SNR; This is the SNR threshold parameter, representing the soft threshold for enabling the fusion mechanism. When the time is right, it tends to enable fusion; otherwise, it tends to disable it.

[0061] The estimated signal-to-noise ratio here This refers to the denoised signal obtained through a signal denoising model, assuming no prior knowledge of noise intensity (i.e., signal-to-noise ratio). With the input noisy signal The signal-to-noise ratio (SNR) is calculated as follows:

[0062] .

[0063] in, This represents the power of the clean signal, which in this case is the signal after denoising. power, This represents the power of the noise, here it is the noise residual. The power.

[0064] The weighted residual reconstruction layer in the noise residual fusion module is used to convert the denoised state monitoring data The residual characteristics obtained above Weighted fusion, final fusion features The definition is as follows:

[0065] .

[0066] in, The expression describes element-wise multiplication.

[0067] like Figure 5 As shown, the fault diagnosis model includes a feature extraction layer and a fault classification layer. The feature extraction layer consists of two layers of one-dimensional convolutional networks and two layers of bidirectional long short-term memory networks, which are used to extract temporal features of fused features. The fault classification layer is a multilayer perceptron, which is used to output the fault type prediction results.

[0068] The feature extraction layer in the fault diagnosis model is used to extract the above-mentioned fused features. The temporal features are extracted using a feature extraction layer consisting of two one-dimensional convolutional networks and two bidirectional long short-term memory networks. The two convolutional layers use wide kernels of size 15 to capture long-range vibrational patterns. Each convolutional layer is followed by batch normalization (BN) to accelerate convergence, ReLU activation to introduce nonlinearity, and a max-pooling layer to reduce feature dimensionality and maintain translation invariance. l The output of a convolutional layer can be represented as:

[0069] .

[0070] in, Represents the ReLU activation function. This represents the convolution operation. and The first Layer weights and biases. The features extracted by the convolutional network are input into a two-layer bidirectional long short-term memory network. The bidirectional long short-term memory network consists of two long short-term memory network layers in opposite directions, which can simultaneously utilize past and future contextual information, thereby more comprehensively capturing the dynamic evolution of vibration signals. To prevent overfitting, a Dropout layer is introduced between the bidirectional long short-term memory network layers.

[0071] The fault classification layer in the fault diagnosis model is used to complete the final fault type prediction and classification. A multilayer perceptron is used as the fault classification layer, which contains two hidden layers and is activated using the ReLU function. In addition, the Dropout function is used between the multilayer perceptrons to enhance the model's generalization ability. Finally, the probability of belonging to each fault category is calculated through the Softmax function, thereby outputting the predicted fault type label.

[0072] B5. Divide several samples into training dataset, validation dataset and test dataset according to a preset ratio; the training dataset is used to train and optimize the weight parameters of the denoising-diagnosis joint model, the validation dataset is used to adjust the model hyperparameters and select the model, and the test dataset is used to test and evaluate the performance of the trained denoising-diagnosis joint model.

[0073] B6. Train the denoising-diagnosis joint model using the training dataset, optimize the weight parameters of the denoising-diagnosis joint model, and obtain the trained denoising-diagnosis joint model. When training the denoising-diagnosis joint model, for any sample, use the preprocessed noisy data as the input of the signal denoising model, the preprocessed clean data as the target output of the signal denoising model, and the fault type label as the target output of the fault diagnosis model. Optimize the weight parameters of the denoising-diagnosis joint model based on the loss function values ​​of the actual output and target output of the signal denoising model and the actual output and target output of the fault diagnosis model.

[0074] In this embodiment, as Figure 6 As shown, step B6 specifically includes the following steps:

[0075] B61. For any sample in the training dataset, input the preprocessed noisy data into the signal denoising model to obtain the output denoised data.

[0076] B62. Calculate the noise residual between the noisy data and the denoised data, and input the noise residual and the denoised data into the noise residual fusion module to obtain the output fusion feature.

[0077] B63. Input the fused features into the fault diagnosis model to obtain the output predicted fault type label.

[0078] B64. Based on the signal denoising model loss function and the fault diagnosis model loss function, construct a composite loss function; the signal denoising model loss function is the mean square error loss function of the denoised data and the preprocessed clean data output by the signal denoising model; the fault diagnosis model loss function is the cross-entropy loss function of the predicted fault type label output by the fault diagnosis model and the fault type label of the sample.

[0079] Specifically, the loss function of the signal denoising model is shown in the following equation:

[0080] .

[0081] in, The loss function value of the signal denoising model. N The total number of samples in the training dataset, For the first i Denoising data for each sample For the first i Clean data for each sample.

[0082] The loss function of the fault diagnosis model is shown in the following equation:

[0083] .

[0084] in, The loss function value of the fault diagnosis model. For the first i Fault type labels for each sample For the first i Predicted fault type labels for each sample.

[0085] The composite loss function is shown in the following equation:

[0086] .

[0087] in, The value of the composite loss function. The weights of the loss function in the signal denoising model. These are the weights of the loss function for the fault diagnosis model.

[0088] B65. Optimizing the weight parameters of the denoising-diagnosis joint model based on a composite loss function. As an optional implementation, the Adam optimization algorithm is used to optimize the weight parameters of the denoising-diagnosis joint model. The learning rate of the Adam optimization algorithm is 0.001, and the number of training epochs is 100.

[0089] To better illustrate the application effect of the method provided in this application, the following experiment uses the CWRU bearing dataset to verify the fault diagnosis method for equipment components based on the noise residual fusion strategy proposed in this application.

[0090] The CWRU dataset is a commonly used bearing fault dataset, frequently used in the field of equipment component fault diagnosis. It records vibration signals under normal and various fault conditions by setting faults at different locations (inner ring, outer ring, rolling elements) and varying degrees on a motor-driven bearing, and placing accelerometers at the drive end, fan end, and base of the motor housing, at sampling frequencies of 12kHz and 48kHz. This example selects the drive end bearing fault dataset with a sampling frequency of 12kHz; detailed information about the dataset is shown in Table 1.

[0091] Table 1 CWRU bearing fault dataset

[0092]

[0093] After obtaining the aforementioned dataset, since the collected vibration data are long-time series signals, a sliding segmentation strategy can be employed to obtain more samples. In this strategy, the step size is set to 256, and the sample length is set to 2048. Furthermore, to prevent test set leakage due to sample overlap, the training, validation, and test datasets are pre-divided before the sliding segmentation strategy is applied, with a ratio of 7:1.5:1.5. After these processing steps, a total of 22,890 samples are obtained, including 16,350 training samples, 3,270 validation samples, and 3,270 test samples.

[0094] Subsequently, to simulate a noisy scenario, each of the above samples was treated as clean state monitoring data, and randomly generated Gaussian noise with a specific signal-to-noise ratio was used. Noise was added to the data to obtain noisy state monitoring data. To simulate the noise intensity of different noise scenarios, different noise levels were selected. Gaussian noise of different intensities.

[0095] The condition monitoring data was then standardized to convert it into a mean value. The standard deviation is The data is standardized to its standard distribution for subsequent processing. Data standardization is a commonly used data preprocessing method. Data standardization can not only speed up the convergence of the model, but also effectively improve the accuracy of the model.

[0096] Subsequently, by combining an end-to-end fully convolutional temporal signal denoising model based on dilated convolution, a noise residual fusion module based on a gating mechanism, and a fault diagnosis model based on convolutional neural networks and bidirectional long short-term memory neural networks, a joint denoising-diagnosis model is constructed. The signal denoising model is as follows: Figure 3 As shown, it includes: a preprocessing layer, stacked dilated convolutional residual blocks, and an output post-processing layer; the noise residual fusion module is as follows: Figure 4 As shown, it includes: a residual feature extraction layer, an adaptive gating mechanism, and a weighted residual reconstruction layer; the fault diagnosis model is as follows: Figure 5 As shown, it includes: a feature extraction layer and a fault classification layer. The specific structure and parameters of the network are shown in the figure.

[0097] In training the aforementioned denoising-diagnosis joint model using the training set, samples from the validation dataset are input into the model to obtain the fault diagnosis accuracy. The model that achieves the best validation set performance out of 100 training epochs is selected as the final model. Finally, the model's performance is tested using the test set, and accuracy (ACC) is selected as the evaluation function to assess the fault diagnosis performance. Accuracy is the proportion of correctly predicted samples out of the total samples.

[0098] To avoid the influence of random factors, each experiment was repeated 5 times, and the mean and standard deviation of the 5 experiments were recorded as the final evaluation parameters. In this example, the experimental results for different signal-to-noise ratios are shown in Table 2 below.

[0099] Table 2 Model Test Results

[0100]

[0101] As shown in Table 2 above, after multiple experiments, the accuracy of fault diagnosis of equipment components in noisy environments using the method proposed in this application is high across various signal-to-noise ratio scenarios. Even in extreme noise scenarios (signal-to-noise ratio of -10dB), the accuracy reaches 95.16%. Furthermore, the standard deviation of the accuracy is also small, indicating that the model proposed in this application has good stability. In conclusion, the method proposed in this application is effective in the field of fault diagnosis of equipment components in noisy environments.

[0102] The proposed solution employs a cascaded joint training framework to effectively integrate upstream denoising and downstream diagnostic tasks, allowing them to complement and learn from each other. Furthermore, a noise residual fusion strategy effectively prevents the loss of crucial information during denoising, resulting in excellent fault diagnosis performance in noisy environments. This approach can be widely applied to fault diagnosis of equipment components in noisy environments across various industrial sectors. In the field of fault diagnosis for equipment components in noisy environments, this application is the first to propose using a cascaded joint training framework to integrate upstream denoising and downstream diagnostic tasks, allowing them to complement and learn from each other. The noise residual fusion strategy effectively prevents the loss of crucial information during denoising, ultimately achieving excellent fault diagnosis results in noisy environments. This method demonstrates significant innovation and practicality.

[0103] Based on the same inventive concept, this application also provides an apparatus for implementing the above-described method for fault diagnosis of equipment components based on a noise residual fusion strategy. The solution provided by this apparatus is similar to the solution described in the above method, and will not be repeated here.

[0104] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0105] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0106] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0107] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Moreover, the collection, use and processing of the relevant data are carried out in compliance with the relevant data protection laws and policies of the country where the location is located, and with the authorization granted by the owner of the corresponding device.

[0108] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0109] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0110] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A fault diagnosis method for equipment components based on a noise residual fusion strategy, characterized in that, include: Acquire noisy condition monitoring data; The noisy condition monitoring data refers to the noisy condition monitoring data collected when the mechanical equipment is running in a noisy environment; The noisy state monitoring data is standardized and preprocessed to obtain preprocessed noisy state monitoring data; The preprocessed noisy condition monitoring data is input into the trained denoising-diagnosis joint model, and the predicted fault type label is output. The denoising-diagnosis joint model includes a signal denoising model, a noise residual fusion module, and a fault diagnosis model. The signal denoising model is used to denoise the input data to obtain denoised data. The noise residual fusion module is used to fuse the denoised data and the noise residual to obtain fusion features. The fault diagnosis model is used to predict and output the fault type label based on the fusion features. The noise residual is the difference between the noisy state monitoring data and the denoised data; the noise residual fusion module includes a residual feature extraction layer, an adaptive gating mechanism, and a weighted residual reconstruction layer; the residual feature extraction layer is a three-layer 3×1 one-dimensional convolutional network used to extract residual features of deep noise; the adaptive gating mechanism includes a gating network and an SNR sensing factor used to determine the gating weights and control the degree of fusion of residual features; the weighted residual reconstruction layer is used to fuse the denoised data with the weighted residual features to obtain fused features.

2. The equipment component fault diagnosis method based on noise residual fusion strategy according to claim 1, characterized in that, Also includes: Obtain the historical state detection dataset, construct and train the denoising-diagnosis joint model, and obtain the trained denoising-diagnosis joint model, specifically including: Obtain a historical condition monitoring dataset; the historical condition monitoring dataset includes several clean condition monitoring data and corresponding fault type labels, the clean condition monitoring data being noise-free condition monitoring data collected during the operation of the mechanical equipment; For any clean status monitoring data, Gaussian noise is added to the clean status monitoring data according to a preset signal-to-noise ratio to obtain simulated noisy status monitoring data. The clean state monitoring data and the simulated noisy state monitoring data are respectively standardized and preprocessed to obtain preprocessed clean data and preprocessed noisy data; the preprocessed noisy data, the preprocessed clean data and the corresponding fault type label constitute a sample; several samples constitute a training dataset; A denoising-diagnosis joint model is constructed based on a signal denoising model, a noise residual fusion module, and a fault diagnosis model. The signal denoising model is an end-to-end fully convolutional time-domain signal denoising model based on dilated convolution, the noise residual fusion module is a fusion module based on a gating mechanism, and the fault diagnosis model is a diagnostic model based on a convolutional neural network and a bidirectional long short-term memory neural network. The denoising-diagnosis joint model is trained using the training dataset, and the weight parameters of the denoising-diagnosis joint model are optimized to obtain a well-trained denoising-diagnosis joint model.

3. The equipment component fault diagnosis method based on noise residual fusion strategy according to claim 2, characterized in that, The denoising-diagnosis joint model is trained using the training dataset, and the weight parameters of the denoising-diagnosis joint model are optimized to obtain a trained denoising-diagnosis joint model, including: For any sample in the training dataset, the preprocessed noisy data is input into the signal denoising model to obtain the output denoised data. Calculate the noise residual between the noisy data and the denoised data, and input the noise residual and the denoised data together into the noise residual fusion module to obtain the output fusion feature; The fused features are input into the fault diagnosis model to obtain the output predicted fault type label; A composite loss function is constructed based on the loss function of the signal denoising model and the loss function of the fault diagnosis model. The loss function of the signal denoising model is the mean square error loss function of the denoised data and the preprocessed clean data output by the signal denoising model. The loss function of the fault diagnosis model is the cross-entropy loss function of the predicted fault type label output by the fault diagnosis model and the fault type label of the sample. The weight parameters of the denoising-diagnosis joint model are optimized based on the composite loss function.

4. The equipment component fault diagnosis method based on noise residual fusion strategy according to claim 3, characterized in that, The loss function of the signal denoising model is shown in the following equation: ； in, The loss function value of the signal denoising model. N The total number of samples in the training dataset, For the first i Denoising data for each sample For the first i Clean data for each sample; The loss function of the fault diagnosis model is shown in the following equation: ； in, The loss function value of the fault diagnosis model. For the first i Fault type labels for each sample For the first i Predicted fault type labels for each sample; The composite loss function is shown in the following equation: ； in, The value of the composite loss function. The weights of the loss function in the signal denoising model. These are the weights of the loss function for the fault diagnosis model.

5. The equipment component fault diagnosis method based on noise residual fusion strategy according to claim 4, characterized in that, When optimizing the weight parameters of the denoising-diagnosis joint model, the Adam optimization algorithm is used to optimize the weight parameters of the denoising-diagnosis joint model. The learning rate of the optimization algorithm is 0.001 and the number of training rounds is 100.

6. The equipment component fault diagnosis method based on noise residual fusion strategy according to claim 2, characterized in that, The preset signal-to-noise ratio is shown in the following formula: ； in, To preset the signal-to-noise ratio, The power of clean status monitoring data, The power of Gaussian noise.

7. The equipment component fault diagnosis method based on noise residual fusion strategy according to any one of claims 1 or 2, characterized in that, The monitoring data is standardized and preprocessed according to the following formula: ； in, This is the preprocessed status monitoring data. This is the original condition monitoring data. The average of the monitored data, The standard deviation of the monitoring data, It is set to a minimum value to prevent the denominator from being 0.

8. The equipment component fault diagnosis method based on noise residual fusion strategy according to claim 1, characterized in that, The signal denoising model includes an input preprocessing layer, stacked dilated convolutional residual blocks, and an output post-processing layer; the input preprocessing layer is a 1×1 convolutional layer used to map the number of input channels to the number of residual channels; the number of stacked dilated convolutional residual blocks is L, and the... i The expansion factor of the layer is Each dilated convolutional residual block employs a gated activation mechanism and introduces residual connections and skip connections; the output post-processing layer includes two nonlinear transformation layers and a 1×1 convolutional layer for reconstructing clean, denoised data.

9. The equipment component fault diagnosis method based on noise residual fusion strategy according to claim 1, characterized in that, The fault diagnosis model includes a feature extraction layer and a fault classification layer; the feature extraction layer consists of two layers of one-dimensional convolutional networks and two layers of bidirectional long short-term memory networks, used to extract temporal features of fused features; The fault classification layer is a multilayer perceptron, used to output fault type prediction results.

Images