Mechanical equipment fault diagnosis model construction method, mechanical equipment fault diagnosis method, mechanical equipment fault diagnosis system and mechanical equipment fault diagnosis equipment

Abstract

The invention provides a mechanical equipment fault diagnosis model construction method, a diagnosis method, a system and equipment, and relates to the technical field of mechanical fault diagnosis, and the fault diagnosis method comprises the steps: obtaining the vibration signal data of a rotating machine in real time; and inputting the vibration signal data into the fault diagnosis model to obtain the fault type of the rotating machine. According to the method, the specific fault type of the rotating machine can be accurately identified, so that measures can be taken in time for repair or maintenance, further deterioration of faults is effectively avoided, equipment safety is protected, production efficiency is improved, and maintenance cost is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of mechanical fault diagnosis, and in particular to a mechanical equipment fault diagnosis model construction method, a diagnosis method, a system and a device. Background Art

[0002] With the rapid development of smart parks, the demand for modern infrastructure and efficient management systems is increasing. Rotating machinery and equipment, as key electromechanical equipment in smart parks, are widely used in various systems such as water supply, heating, ventilation, and air conditioning. The operating status of these equipment directly affects the energy consumption, environmental quality, and maintenance costs of the park. However, during long-term operation, rotating machinery and equipment are susceptible to mechanical wear, fatigue damage, lubrication failure, and other factors, leading to failures. Failures will not only interrupt the normal operation of the equipment and increase maintenance costs, but may also have a serious impact on the overall operating efficiency and safety of the smart park.

[0003] The inventors found in their research that in the prior art, deep learning algorithms are usually used to perform fault diagnosis and identification on the vibration signals of rotating machinery. In the fault diagnosis process, the influence of the timing characteristics in the vibration signal on the fault diagnosis recognition rate is often ignored. There is a lack of effective extraction of timing characteristics. In addition, since the acquired data has complex background and noise interference, it is difficult to accurately capture the subtle feature differences that are crucial to fault type discrimination using a deep learning algorithm model, which leads to a decrease in the fault diagnosis recognition rate. Summary of the invention

[0004] In view of this, an object of the present invention is to provide a mechanical equipment fault diagnosis model construction method, diagnosis method, system and equipment to improve the accuracy and robustness of fault diagnosis under noise interference.

[0005] In a first aspect, the present invention provides a method for constructing a mechanical equipment fault diagnosis model, comprising: Acquire vibration signal data of the rotating machinery, and preprocess the vibration signal data to obtain a training data set of the vibration signal data, wherein the training data set includes a fault type and a vibration signal corresponding to the fault type; Based on the training data set, the initial fault diagnosis model is iteratively trained until the conditions are met to obtain the target fault diagnosis model. Each iterative training includes: processing the vibration signal data based on the improved convolutional neural network and the improved bidirectional gated recurrent unit to obtain the spatial characteristics and time series characteristics of the rotating machinery; fusing the spatial characteristics and time series characteristics to determine the target feature vector of the rotating machinery; based on the preset classification decision rules, determining the predicted fault type corresponding to the target feature vector; and updating the initial fault diagnosis model based on the predicted fault type and the standard fault type.

[0006] Optionally, the vibration signal data is processed based on an improved convolutional neural network to obtain spatial features of the rotating machinery, including: Based on the sparse attention matrix, the vibration signal data is processed to obtain first data; The first data is input into a convolutional neural network to obtain the spatial characteristics of the rotating machinery.

[0007] Optionally, the vibration signal data is processed based on an improved bidirectional gated cyclic unit to obtain the timing characteristics of the rotating machinery, including: Based on the time window and the sparse attention matrix, the vibration signal data is processed to obtain second data; The second data is input into the bidirectional gated cycle unit to obtain the timing characteristics of the rotating machinery.

[0008] Optionally, the spatial features and the temporal features are fused to determine the target feature vector of the rotating machinery, including: Based on the feature concatenation operation, the fused feature vector corresponding to the fusion of spatial features and temporal features is obtained; Based on the compression channel, a compressed feature vector corresponding to the fused feature vector is obtained; Based on the excitation channel, an excitation feature vector corresponding to the compressed feature vector is obtained; Based on channel-by-channel multiplication, the target feature vector corresponding to the excitation feature vector is obtained.

[0009] Optionally, the spatial features and the temporal features are fused to determine the target feature vector of the rotating machinery, further comprising: Based on the channel importance characteristics, the bias and weight values ​​of the compression channel and the excitation channel are set.

[0010] Optionally, based on the training data set, the initial fault diagnosis model is iteratively trained until a target fault diagnosis model is obtained by satisfying conditions, including: Preprocess the vibration signal data to obtain a test data set; Based on the training data set, the cross entropy loss function is used as the loss function for model training, and the fault diagnosis model is trained through the optimizer and regularization; Optimize the trained fault diagnosis model based on attention mechanism, weight decay strategy and learning rate strategy; The optimized fault diagnosis model is verified based on the test data set.

[0011] In a second aspect, the present invention provides a mechanical equipment fault diagnosis method, comprising: Obtain vibration signal data of rotating machinery in real time; The vibration signal data is input into the fault diagnosis model to obtain the fault type of the rotating machinery; wherein the fault diagnosis model is obtained by using the mechanical equipment fault diagnosis model construction method as described above.

[0012] In a third aspect, the present invention further provides a mechanical equipment fault diagnosis model construction system, comprising: A preprocessing module is used to obtain vibration signal data of the rotating machinery and preprocess the vibration signal data to obtain a training data set of the vibration signal data, wherein the training data set includes a fault type and a vibration signal corresponding to the fault type; The model training module is used to iteratively train the initial fault diagnosis model based on the training data set until the conditions are met to obtain the target fault diagnosis model. Each iterative training includes: processing the vibration signal data based on the improved convolutional neural network and the improved bidirectional gated recurrent unit to obtain the spatial characteristics and time series characteristics of the rotating machinery; fusing the spatial characteristics and time series characteristics to determine the target feature vector of the rotating machinery; determining the predicted fault type corresponding to the target feature vector based on the preset classification decision rules; and updating the initial fault diagnosis model based on the predicted fault type and the standard fault type.

[0013] In a fourth aspect, the present invention further provides a mechanical equipment fault diagnosis system, comprising: An acquisition module, used for acquiring vibration signal data of a rotating machine in real time; The diagnosis module is used to input the vibration signal data into the fault diagnosis model to obtain the fault type of the rotating machinery; wherein the fault diagnosis model is obtained by using the mechanical equipment fault diagnosis model construction method as mentioned above.

[0014] In a fifth aspect, the present invention also provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the processor implements the steps of the above-mentioned mechanical equipment fault diagnosis model construction method, or implements the steps of the above-mentioned mechanical equipment fault diagnosis method.

[0015] The embodiment of the present invention provides a method for constructing a mechanical equipment fault diagnosis model, a diagnosis method, a system and a device, which obtains vibration signal data of a rotating machine and pre-processes the vibration signal data to obtain a training data set of the vibration signal data, wherein the training data set includes a fault type and a vibration signal corresponding to the fault type; based on the training data set, the initial fault diagnosis model is iteratively trained until the conditions are met to obtain a target fault diagnosis model, and each iterative training includes: processing the vibration signal data based on an improved convolutional neural network and an improved bidirectional gated cyclic unit to obtain the spatial characteristics and time series characteristics of the rotating machine; fusing the spatial characteristics and time series characteristics to determine the target feature vector of the rotating machine; determining the predicted fault type corresponding to the target feature vector based on a preset classification decision rule; and updating the initial fault diagnosis model based on the predicted fault type and the standard fault type. The fault diagnosis model can accurately identify the specific fault type of the rotating machine, so that timely measures can be taken for repair or maintenance, effectively avoiding further deterioration of the fault, protecting equipment safety, improving production efficiency and reducing maintenance costs.

[0016] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0018] Figure 1 A schematic diagram showing a flow chart of a method for constructing a mechanical equipment fault diagnosis model provided by an embodiment of the present invention; Figure 2 A schematic diagram showing a flow chart of another method for constructing a mechanical equipment fault diagnosis model provided by an embodiment of the present invention; Figure 3 A schematic diagram of the structure of a fault analysis layer provided by an embodiment of the present invention is shown; Figure 4 A schematic diagram showing a flow chart of a mechanical equipment fault diagnosis method provided by an embodiment of the present invention; Figure 5 A schematic diagram of the structure of a mechanical equipment fault diagnosis model building system provided by an embodiment of the present invention is shown; Figure 6 A schematic diagram showing the structure of a mechanical equipment fault diagnosis system provided by an embodiment of the present invention is shown; Figure 7 A schematic structural diagram of an electronic device provided by an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present invention.

[0020] The existing methods for using deep learning algorithms to perform fault diagnosis and identification on vibration signals of rotating machinery generally include the following methods: Method 1: obtain normal variable data and fault variable data associated with the target system, preprocess these data, and then input the preprocessed normal variable data and fault variable data into a pre-trained convolutional neural network fault diagnosis model for diagnosis to obtain the fault diagnosis result of the air compression system; Method 2: obtain the original bearing fault vibration data, extract the intrinsic mode components of different frequency bands in the data, perform wavelet threshold denoising according to the permutation entropy value, and obtain the denoised and reconstructed time domain signal; perform fast Fourier transform on the obtained time domain signal, and draw the FFT-x heat map of data of different fault types; construct an improved lightweight convolutional neural network model L-CNN, input the encoded data into the model for processing and diagnosis, and obtain the fault diagnosis result; Method 3: while using a one-dimensional deep separable convolutional neural network to perform a one-dimensional convolution operation on the vibration signal of the transformer, a two-dimensional deep separable convolutional neural network is also used to extract two-dimensional features from the vibration signal of the transformer, and the one-dimensional features and the two-dimensional features are spliced ​​and input into the Softmax regression classification model to obtain the fault type. Since rotating machinery vibration signals have rich spatial and temporal characteristics, existing methods use a single grid structure, which makes it difficult to capture all the key features of the signal.

[0021] Figure 1 The following is a flow chart of a method for constructing a mechanical equipment fault diagnosis model provided by an embodiment of the present invention. Figure 1 As shown, the method comprises the following steps: Step S110: Acquire vibration signal data of the rotating machinery, and preprocess the vibration signal data to obtain a training data set of the vibration signal data, wherein the training data set includes a fault type and a vibration signal corresponding to the fault type.

[0022] The vibration signal data of the bearing is collected by an acceleration sensor arranged on the rotating machine, and the vibration signal data is a one-dimensional signal.

[0023] Further, if Figure 2 As shown, based on the preset segmentation rule, the collected vibration signal is preprocessed to obtain a training data set and a test data set, wherein the preset segmentation rule in this application is to segment the vibration signal data into data segments of 1024 lengths, and divide the segmented vibration signal data into a training data set and a test data set in a ratio of 8:2, and each data set includes at least one subset.

[0024] Furthermore, the training dataset and the test dataset may also adopt existing datasets, such as the Case Western Reserve University bearing dataset.

[0025] As rotating machinery is an important component of various mechanical equipment, it has the characteristics of a wide variety and complex structure. Common rotating machinery includes three basic components: bearings, gears and rotating shafts. Bearings are one of the most common components in various types of rotating machinery. The health of bearings plays a key role in the normal operation of mechanical equipment. Therefore, the present invention mainly focuses on fault diagnosis of bearings in rotating machinery.

[0026] Step S120, based on the training data set, iteratively train the initial fault diagnosis model until the conditions are met to obtain the target fault diagnosis model, each iterative training includes: processing the vibration signal data based on the improved convolutional neural network and the improved bidirectional gated recurrent unit to obtain the spatial characteristics and time series characteristics of the rotating machinery; fusing the spatial characteristics and the time series characteristics to determine the target feature vector of the rotating machinery; based on the preset classification decision rules, determining the predicted fault type corresponding to the target feature vector; updating the initial fault diagnosis model based on the predicted fault type and the standard fault type.

[0027] Among them, the fault diagnosis model includes a feature extraction layer, a feature fusion layer and a fault classification layer. The feature extraction layer is used to process the vibration signal data based on an improved convolutional neural network and an improved bidirectional gated recurrent unit to obtain the spatial characteristics and time series characteristics of the rotating machinery; the feature fusion layer is used to fuse the spatial characteristics and time series characteristics to determine the target feature vector of the rotating machinery; the fault classification layer is used to determine the predicted fault type corresponding to the target feature vector based on preset classification decision rules.

[0028] Furthermore, the feature extraction layer includes a spatial feature extraction channel of an improved convolutional neural network and a temporal feature extraction channel based on an improved bidirectional gated recurrent unit. The spatial features and temporal features in the vibration signal data are extracted through the two extraction channels, thereby avoiding the prior art's neglect of the temporal features of the vibration signal and the destruction of the integrity of the temporal features by the common series network, thereby enhancing the effectiveness of feature extraction.

[0029] In an optional embodiment, in step S120, the vibration signal data is processed based on the improved convolutional neural network to obtain the spatial characteristics of the rotating machinery, including: Based on the sparse attention matrix, the vibration signal data is processed to obtain first data; The first data is input into a convolutional neural network to obtain the spatial characteristics of the rotating machinery.

[0030] Specifically, the spatial feature extraction channel consists of an improved Linformer (Linearized Transformer, attention mechanism), a multi-layer basic convolution module and a fully connected layer, where the multi-layer basic convolution module includes a one-dimensional convolution layer, a batch normalization layer, an activation function layer and a maximum pooling layer. Thus, its spatial feature extraction can capture the key information of spatial features and effectively utilize the attention mechanism to improve the model's expressiveness and computational efficiency.

[0031] Linformer significantly reduces computational complexity by simplifying traditional global attention calculations into low-dimensional matrix multiplications. Its attention formula is as follows:

[0032] In the formula, is the attention matrix, is a function that converts attention weights into probability distributions, is the query matrix, The key matrix The matrix of photography to low-dimensional space, is the length of the sequence, is the dimension of the key matrix, To convert the value matrix Matrix of photography into low-dimensional space.

[0033] By sparsifying the attention matrix, the computational efficiency of the model can be further optimized. In this application, the threshold method is used to sparsify the attention matrix, that is, the attention weights above a preset threshold are retained in the attention matrix, and the part below the threshold is set to zero. The threshold can directly affect the performance of the model, so it is necessary to select a suitable threshold.

[0034] Furthermore, the threshold range can be preliminarily determined by analyzing the statistical characteristics of the training data in the training data set, and then methods such as cross-validation can be used to optimize and determine the optimal threshold.

[0035] Specifically, determine the appropriate threshold through the following steps: First, the distribution characteristics of attention weights are determined by analyzing the training data in the training set, and the range of thresholds is determined using statistical features such as mean and variance; Secondly, one or more thresholds are initially set based on the distribution characteristics of the attention weights; Then, the initially set threshold is applied to the model, and the performance of the model under different thresholds is determined by cross-validation method. The performance of the model includes but is not limited to accuracy, calculation speed during training and inference, etc.; Finally, the optimal threshold is determined based on the results of cross-validation.

[0036] Through this sparse processing, the model can reduce unnecessary calculations, thereby improving efficiency without affecting its ability to capture important information. At the same time, it can significantly reduce the computational complexity of the attention mechanism, as well as reduce memory usage and computing resource requirements.

[0037] Furthermore, for the attention matrix , define a sparse mask ,like express is retained if express is thinned, where represents the position of a certain attention matrix, Represents the row index, Represents the column index, and the sparse attention matrix is:

[0038] In the formula, is the sparse attention matrix, is a sparse mask.

[0039] In an optional embodiment, the multi-layer basic convolution module includes a one-dimensional convolution layer, a batch normalization layer, an activation function layer and a maximum pooling layer, wherein the one-dimensional convolution layer extracts feature vectors from sequence data through convolution operations; the batch normalization layer improves the stability and convergence speed of the network; the activation function layer introduces nonlinear transformations to enable the neural network to learn complex data models and relationships, and sets the activation function in the CNN (Convolutional Neural Network) channel to the ReLU function; the maximum pooling layer selects the maximum value in the local area to reduce the spatial size of the input data and extract the main features.

[0040] Furthermore, after the vibration signal data enters the CNN channel, the convolutional neural network locally perceives the input vibration signal data, and then enters the hidden layer to process the acquired partial information, extract the spatial features contained in the signal, and form a spatial feature vector.

[0041] Specifically, the vibration signal data is subjected to feature extraction using a one-dimensional convolutional neural network using the following formula:

[0042] In the formula, is the output feature vector after convolution and batch normalization, To represent batch normalization, is the convolution operation, is the spatial feature attention weight value, is the vibration signal data vector, is the spatial offset value, is a constant.

[0043] Then, the extracted feature vector is calculated by the following formula: Perform activation function and maximum pooling operations:

[0044]

[0045] In the formula, is the output feature vector after the activation function, is the ReLU activation function, is the output feature vector after the maximum pooling operation, that is, the spatial feature vector, It is the maximum pooling operation.

[0046] It should be noted that when the spatial feature extraction channel extracts the spatial feature vector of the vibration signal data, it first uses the improved Linformer to process the vibration signal data, that is, different parts of the vibration signal data are assigned different importance weights. In this application, the data that is weighted by the improved Linformer in the spatial feature extraction channel is called the first data, and then the vibration signal data with weights is input into the convolutional neural network model to obtain the spatial feature vector corresponding to the vibration signal.

[0047] In an optional embodiment, in step S120, the vibration signal data is processed based on the improved bidirectional gated cycle unit to obtain the timing characteristics of the rotating machinery, including: Based on the time window and the sparse attention matrix, the vibration signal data is processed to obtain second data; The second data is input into the bidirectional gated cycle unit to obtain the timing characteristics of the rotating machinery.

[0048] Specifically, the temporal feature extraction channel consists of a self-attention mechanism based on sparse attention and a BiGRU (Bidirectional Gated Recurrent Unit). By using the sparse attention mechanism, only some important attention weights need to be paid attention to, which greatly reduces the amount of calculation and effectively reduces the computational complexity.

[0049] Furthermore, the present application adopts a fixed window sparse mask approach to limit the focus range of local relations, thereby achieving sparse attention.

[0050] Specifically, first, determine a fixed-size window and the starting position of the window; secondly, use a sparse mask matrix to identify the position within the current window. For the position within the window, the sparse mask value is 0 or a positive number, indicating that the value after sparse masking remains unchanged. For the position outside the window, the sparse mask value is -1 or negative infinity, and it is removed. It uses a fixed window sparse mask to limit the focus range of the local relationship. The specific formula is:

[0051]

[0052] In the formula, For the The sparse mask value of the attention matrix, that is, For M, An identifier value for the sparse mask value.

[0053] Furthermore, when BiGRU is used to process one-dimensional data (vibration signal data), BiGRU consists of two GRUs (Gated Recurrent Unit, gate control recurrent unit) in the positive and negative directions. Therefore, the model learns the characteristics of time series data in two directions. For a given input sequence of vibration signal data vector , the goal of BiGRU is to output a sequence of time series feature vectors , for each time step in the sequence , the hidden state of the forward GRU The input vibration signal data vector The relationship between is calculated through a series of gating operations, where a series of gating operations include a reset gate, an update gate, a candidate hidden state, and an updated hidden state.

[0054] Among them, the reset gate determines the current input vibration signal data vector The combination with the previous memory controls the degree of forgetting, and the expression of the reset gate is:

[0055] In the formula, is the output of the reset gate, is the Sigmoid activation function, is the weight matrix of the vibration signal data vector in the reset gate, is the vibration signal data vector, is the weight matrix of the hidden state in the reset gate, For the time step The hidden state of The bias value for resetting the gate.

[0056] The update gate controls the update degree of the current hidden state and determines the fusion ratio of new information and old information. The expression of the update gate is:

[0057] In the formula, is the output of the update gate, is the weight matrix of the vibration signal data vector in the update gate, is the weight matrix of the hidden state in the update gate, is the bias value for the update gate.

[0058] The candidate hidden state is the candidate information generated based on the current input and the previous hidden state, and then combined with the update gate to generate the final hidden state. The expression of the candidate hidden state is:

[0059] In the formula, is the output of the candidate hidden state, is the hyperbolic tangent function, is the weight matrix of the vibration signal data vector in the candidate hidden state, is the weight matrix of the hidden state in the candidate hidden state, is the bias value of the candidate hidden state.

[0060] The final hidden state is calculated based on the update gate and the candidate hidden state, and its calculation formula is as follows:

[0061] In the formula, is the feature vector of the forward GRU.

[0062] Since the reverse GRU follows the same update rule as the forward GRU, but the reverse GRU processes data from the end of the sequence and moves forward, the feature vector obtained is .

[0063] Finally, for each time step , the forward GRU and the reverse GRU are combined at time step The hidden state is combined and the time series feature is output. The expression of the time series feature is:

[0064] In the formula, is the feature vector of the forward GRU A collection of The feature vector of the reverse GRU is A collection of .

[0065] It should be noted that when the timing feature extraction channel extracts the timing feature vector of the vibration signal data, it first uses the improved Linformer to process the vibration signal data, that is, different parts of the vibration signal data are assigned different importance weights. In the present application, in the timing feature extraction channel, the data to which the vibration signal data is weighted by the improved Linformer is called the third data. Then, the third data is processed by the fixed window method to obtain the second data. Finally, the second data is input into the bidirectional gated recurrent unit to obtain the timing feature vector corresponding to the vibration signal data.

[0066] In the present application, a dual-channel grid structure is set up, namely, a spatial feature extraction channel based on an improved Linformer and a convolutional neural network, and a temporal feature extraction channel based on an improved self-attention mechanism and a bidirectional gated recurrent unit, thereby avoiding the prior art's neglect of the temporal features of vibration signals and the destruction of the integrity of the temporal features by common series networks, thereby enhancing the effectiveness of feature extraction; and the improved Linformer introduces a threshold method to perform sparse processing on the attention matrix, thereby greatly improving the computational efficiency while maintaining the accuracy of the model.

[0067] In an optional embodiment, in step S120, the spatial feature and the temporal feature are fused to determine the target feature vector of the rotating machinery, including: Based on the feature concatenation operation, the fused feature vector corresponding to the fusion of spatial features and temporal features is obtained; Based on the compression channel, a compressed feature vector corresponding to the fused feature vector is obtained; Based on the excitation channel, an excitation feature vector corresponding to the compressed feature vector is obtained; Based on channel-by-channel multiplication, the target feature vector corresponding to the excitation feature vector is obtained.

[0068] In an optional embodiment, in step S120, the spatial feature and the temporal feature are fused to determine the target feature vector of the rotating machinery, and further includes: Based on the channel importance characteristics, the bias and weight values ​​of the compression channel and the excitation channel are set.

[0069] Among them, the feature fusion layer is used to fuse features from different channels to improve the performance and generalization ability of the equipment fault diagnosis model, realize information integration, and make full use of the complementarity of different features, thereby improving the equipment fault diagnosis model's ability to understand data and enhancing the robustness of the equipment fault diagnosis model.

[0070] Specifically, the feature fusion layer uses a feature concatenation operation to fuse the spatial feature vector and the temporal feature vector to obtain a target feature vector corresponding to the vibration signal data.

[0071] Furthermore, the feature splicing operation can use feature splicing to directly merge the spatial feature vector and the temporal feature vector, ensuring the integrity and diversity of the features. At the same time, feature splicing is flexible and does not require the dimensions between features to remain consistent. The representation ability of the model is increased through feature splicing. The feature splicing of the spatial feature vector and the temporal feature vector is performed through the following formula:

[0072] In the formula, is the target feature vector, is the concatenation function, is the set of spatial feature vectors, is a set of time series feature vectors.

[0073] Furthermore, based on the improved SENet (Squeeze-and-Excitation Network) model, the target feature vector is processed to enhance the recognition ability and robustness of the fault diagnosis model for vibration signal data.

[0074] Among them, the two-dimensional global average pooling in the original SENet model is replaced with one-dimensional global average pooling to obtain an improved SENet model. The improved SENet model can realize channel adaptive calibration, set different weights for different channels, adaptively learn and emphasize the importance of key features, help strengthen useful features, and suppress irrelevant features.

[0075] like Figure 3 As shown, the feature vector As the input of the improved SENet model, the target feature vector is processed by compression, excitation and channel multiplication. The output of the improved SENet model is , the improved SENet model expression is:

[0076] In the formula, is the output vector of the improved SENet model, is the Sigmoid activation function, To motivate fully connected operation, is the ReLU activation function, is the compressed full connection operation, L is the compression factor, is the feature vector In Location and Channel The characteristic value of is the compression bias value, is the excitation bias value, is the target feature vector.

[0077] In an optional embodiment, in step S120, based on a preset classification decision rule, determining the predicted fault type corresponding to the target feature vector includes: Based on the Softmax classifier, the predicted fault type corresponding to the target feature vector is determined.

[0078] Among them, the main function of the fault classification layer is to classify the input data and assign it to different fault categories. As the last layer in the neural network model, the fault classification layer has the function of classification decision-making. It classifies the input features and completes the classification recognition in fault diagnosis. At the same time, it can also be used to evaluate the performance of the model in the testing phase.

[0079] Furthermore, in this application, the Softmax classifier is selected as the fault classification layer, and the target feature vector is used as the input feature vector for classification and identification. As a standardized exponential function, the Softmax function is widely used to estimate the probability distribution of the original vibration signal mapped to different fault categories. The expression of the Softmax function is:

[0080] In the formula, To be judged as The failure probability of the class, For the The score of the class fault, for The index of is the score index and of all categories, is the total number of categories.

[0081] Further, through The value can determine the fault type that the vibration signal needs to correspond to. The fault type includes but is not limited to cracks in the outer ring of the bearing, wear on the inner ring of the bearing, and wear on the rolling element.

[0082] In an optional embodiment, in step S120, the initial fault diagnosis model is iteratively trained based on the training data set until a target fault diagnosis model is obtained by satisfying conditions, including: Preprocess the vibration signal data to obtain a test data set; Based on the training data set, the cross entropy loss function is used as the loss function for model training, and the fault diagnosis model is trained through the optimizer and regularization; Optimize the trained fault diagnosis model based on attention mechanism, weight decay strategy and learning rate strategy; The optimized fault diagnosis model is verified based on the test data set.

[0083] Specifically, the cross entropy loss function is selected as the loss function for fault diagnosis model training in this application, which can measure the difference between the predicted probability distribution and the true distribution in the fault diagnosis model; Adam is selected as the optimizer to train the network model. The learning rate of each parameter is adaptively adjusted by estimating the square root of the cumulative gradient and the first-order moment of the gradient, so that the fault diagnosis model can quickly converge to the optimal solution. Using dropout technology, during the training process, some neurons and their connections are randomly set to zero, that is, the weights are set to 0, but the bias term is usually retained, which can enhance the generalization ability of the fault diagnosis model; during testing, Dropout does not cause any actual calculation abandonment, but can scale the previously reserved weights to be redistributed, so that the entire network without Dropout behaves as if there are more neurons, thereby preventing overfitting, and setting the dropout coefficient to 0.5.

[0084] Furthermore, by introducing the attention mechanism to improve the feature extraction channel, the fault diagnosis model can be improved in its ability to capture long-distance dependencies when processing vibration signals, so that the fault diagnosis model can focus more on important information. The introduction of the attention mechanism will increase the number of fault diagnosis model parameters, making the training process more complicated and the training time longer. Therefore, a weight decay strategy, namely L2 regularization technology, is adopted to limit the size and range of the fault diagnosis model parameters.

[0085] Among them, the L2 regularization technology constrains the parameters by adding a penalty term in the loss function, which is a function of the sum of squares of all parameter weights. This prevents the fault diagnosis model parameters from being too large, reduces the model complexity, and enhances the model's generalization ability for new data.

[0086] The optimization process of L2 regularization is as follows: First, add the L2 regularization term to the original loss function to obtain the optimized loss function, which is expressed as:

[0087] In the formula, is the optimized loss function, is the original loss function, is the regularized sparse term, are the parameters of the fault diagnosis model, is the sum of parameters of all fault diagnosis models.

[0088] Then, optimization algorithms such as gradient descent are used to minimize the loss function containing the regularization term, thereby avoiding excessive parameters, preventing the fault diagnosis model from overfitting, and improving the generalization ability.

[0089] Furthermore, the parameter update expression of the fault diagnosis model is:

[0090] In the formula, is the learning rate, and the regularization term encourages the parameters to converge to a smaller value, thereby reducing the complexity of the model.

[0091] Choosing a suitable learning rate is crucial for model training. If the learning rate is set too high, the model may oscillate or even diverge near the optimal solution and eventually fail to converge. If the learning rate is set too low, the convergence speed of the model will be significantly slowed down, and more time will be required to reach or even approach the expected performance. Therefore, a multi-step decay strategy is adopted in this application. In the early stage of training, a larger learning rate is selected. As the training progresses, the learning rate is gradually reduced in order to allow the model to adjust its parameters in detail.

[0092] The method of adjusting the learning rate in stages can effectively combine the advantages of fast learning and fine-tuning, thereby improving the effect of model training.

[0093] In the mechanical equipment fault diagnosis model construction method provided by the present invention, by designing a dual-channel grid structure, a spatial feature extraction channel based on an improved Linformer and a convolutional neural network, and a temporal feature extraction channel based on an improved self-attention mechanism and a bidirectional gated recurrent unit are respectively designed to extract the spatial feature vector and the temporal feature vector of the vibration signal. The dual-channel grid structure avoids the prior art ignoring the temporal features of the vibration signal and the destruction of the integrity of the temporal features by the common series network, and enhances the effectiveness of feature extraction; using the improved Linformer, using the threshold method, and further optimizing the computational efficiency by sparsely focusing the attention matrix; improving SENet to obtain a SENet that processes one-dimensional data, so that its model is suitable for fields such as time series, increasing the effectiveness of the model to extract features, and enhancing the robustness of the method; and improving the model training optimization method to improve the speed and effectiveness of model training.

[0094] Figure 4 A schematic diagram of a mechanical equipment fault diagnosis method provided by an embodiment of the present invention is shown in FIG. Figure 4 As shown, the method comprises the following steps: Step S210, acquiring vibration signal data of the rotating machinery in real time; Step S220: input the vibration signal data into a fault diagnosis model to obtain the fault type of the rotating machinery; wherein the fault diagnosis model is obtained using the above-mentioned mechanical equipment fault diagnosis model construction method.

[0095] The vibration signal data of the rotating machinery is obtained in real time by using an acceleration sensor arranged at the driving end of the motor of the rotating machinery equipment, and the data collected by the sensor is transmitted in real time to a host computer connected to the sensor by using a data acquisition module.

[0096] Furthermore, the fault diagnosis model obtained by the above-mentioned mechanical equipment fault diagnosis model construction method is installed on the host computer. The fault diagnosis model receives real-time vibration signal data sent by the sensor in real time and analyzes the real-time vibration signal data. When it is determined that the vibration signal belongs to a certain type of fault, the maintenance personnel are notified via text messages and emails so that the abnormal rotating mechanical parts can be checked in time to avoid the occurrence of serious incidents in the future.

[0097] Figure 5 A schematic diagram of a mechanical equipment fault diagnosis model building system provided by an embodiment of the present invention is shown in FIG. Figure 5 As shown, the system includes: A preprocessing module 310 is used to obtain vibration signal data of the rotating machinery and preprocess the vibration signal data to obtain a training data set of the vibration signal data, wherein the training data set includes a fault type and a vibration signal corresponding to the fault type; The model training module 320 is used to iteratively train the initial fault diagnosis model based on the training data set until the conditions are met to obtain the target fault diagnosis model. Each iterative training includes: processing the vibration signal data based on the improved convolutional neural network and the improved bidirectional gated recurrent unit to obtain the spatial characteristics and time series characteristics of the rotating machinery; fusing the spatial characteristics and time series characteristics to determine the target feature vector of the rotating machinery; determining the predicted fault type corresponding to the target feature vector based on the preset classification decision rules; and updating the initial fault diagnosis model based on the predicted fault type and the standard fault type.

[0098] Figure 6 A schematic diagram of a mechanical equipment fault diagnosis system provided by an embodiment of the present invention is shown in FIG. Figure 6 As shown, the system includes: An acquisition module 410 is used to acquire vibration signal data of a rotating machine in real time; The diagnosis module 420 is used to input the vibration signal data into the fault diagnosis model to obtain the fault type of the rotating machinery; wherein the fault diagnosis model is obtained by using the above-mentioned mechanical equipment fault diagnosis model construction method.

[0099] The implementation principle and technical effects of the system provided in the embodiment of the present application are the same as those of the aforementioned method embodiment. For the sake of brief description, for matters not mentioned in the system embodiment, reference may be made to the corresponding contents in the aforementioned method embodiment.

[0100] like Figure 7As shown, an electronic device 600 provided in an embodiment of the present application includes: a processor 601, a memory 602 and a bus, the memory 602 stores machine-readable instructions executable by the processor 601, when the electronic device is running, the processor 601 and the memory 602 communicate through the bus, and the processor 601 executes the machine-readable instructions to perform the steps of the above-mentioned mechanical equipment fault diagnosis model construction method or the steps of the above-mentioned mechanical equipment fault diagnosis method.

[0101] Specifically, the above-mentioned memory 602 and processor 601 can be general-purpose memory and processor, which are not specifically limited here. When the processor 601 runs the computer program stored in the memory 602, it can execute the steps of the above-mentioned mechanical equipment fault diagnosis model construction method or the steps of the above-mentioned mechanical equipment fault diagnosis method.

[0102] The processor 601 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit or software instructions in the processor 601. The above processor 601 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present application can be implemented or executed. The general processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the embodiments of the present application can be directly embodied as a hardware decoding processor for execution, or a combination of hardware and software modules in the decoding processor for execution. The software module may be located in a storage medium mature in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory 602, and the processor 601 reads the information in the memory 602 and completes the steps of the above method in combination with its hardware.

[0103] Corresponding to the steps of the above-mentioned method for constructing a mechanical equipment fault diagnosis model or the steps of the above-mentioned mechanical equipment fault diagnosis method, an embodiment of the present application also provides a computer-readable storage medium, which stores machine-executable instructions. When the computer-executable instructions are called and executed by the processor, the computer-executable instructions prompt the processor to execute the steps of the above-mentioned method for constructing a mechanical equipment fault diagnosis model or the steps of the above-mentioned mechanical equipment fault diagnosis method.

[0104] The mechanical equipment fault diagnosis model construction system or the mechanical equipment fault diagnosis system provided in the embodiments of the present application can be specific hardware on the equipment or software or firmware installed on the equipment. The device provided in the embodiments of the present application, its implementation principle and the technical effect produced are the same as those in the aforementioned method embodiments. For the sake of brief description, the parts not mentioned in the device embodiment can refer to the corresponding contents in the aforementioned method embodiments. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the systems, devices and units described above can all refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here.

[0105] In the embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0106] For another example, the flowchart and block diagram in the accompanying drawings show the possible architecture, functions and operations of the device, method and computer program product according to multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of the code, and a part of the module, program segment or code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of the boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0107] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0108] In addition, each functional unit in the embodiments provided in the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0109] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including a number of instructions to enable an electronic device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store program codes.

[0110] It should be noted that similar numbers and letters represent similar items in the following figures. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. In addition, the terms "first", "second", "third", etc. are only used to distinguish the description and are not to be understood as indicating or implying relative importance.

[0111] Finally, it should be noted that the above embodiments are only specific implementation methods of the present application, which are used to illustrate the technical solutions of the present application, rather than to limit them. The protection scope of the present application is not limited thereto. Although the present application is described in detail with reference to the aforementioned embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the aforementioned embodiments within the technical scope disclosed in the present application, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application. They should all be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A method for constructing a mechanical equipment fault diagnosis model, characterized in that: include: Acquire vibration signal data of the rotating machinery, and preprocess the vibration signal data to obtain a training data set of the vibration signal data, wherein the training data set includes a fault type and a vibration signal corresponding to the fault type; Based on the training data set, the initial fault diagnosis model is iteratively trained until the conditions are met to obtain the target fault diagnosis model, and each iterative training includes: processing the vibration signal data based on an improved convolutional neural network and an improved bidirectional gated recurrent unit to obtain the spatial characteristics and time series characteristics of the rotating machinery; fusing the spatial characteristics and the time series characteristics to determine the target feature vector of the rotating machinery; based on a preset classification decision rule, determining the predicted fault type corresponding to the target feature vector; and updating the initial fault diagnosis model based on the predicted fault type and the standard fault type.

2. The method for constructing a mechanical equipment fault diagnosis model according to claim 1, characterized in that: The vibration signal data is processed based on the improved convolutional neural network to obtain the spatial characteristics of the rotating machinery, including: Based on the sparse attention matrix, the vibration signal data is processed to obtain first data; The first data is input into a convolutional neural network to obtain the spatial characteristics of the rotating machinery.

3. The method for constructing a mechanical equipment fault diagnosis model according to claim 1, characterized in that: The vibration signal data is processed based on the improved bidirectional gated cycle unit to obtain the timing characteristics of the rotating machinery, including: Processing the vibration signal data based on the time window and the sparse attention matrix to obtain second data; The second data is input into a bidirectional gated cycle unit to obtain a timing characteristic of the rotating machinery.

4. The method for constructing a mechanical equipment fault diagnosis model according to claim 1, characterized in that: The spatial feature and the temporal feature are integrated to determine the target feature vector of the rotating machinery, including: Based on the feature concatenation operation, a fused feature vector corresponding to the fusion of the spatial feature and the temporal feature is obtained; Based on the compression channel, obtaining a compressed feature vector corresponding to the fused feature vector; Based on the excitation channel, obtaining an excitation feature vector corresponding to the compression feature vector; Based on channel-by-channel multiplication, the target feature vector corresponding to the excitation feature vector is obtained.

5. The method for constructing a mechanical equipment fault diagnosis model according to claim 4, characterized in that: The spatial feature and the temporal feature are integrated to determine the target feature vector of the rotating machinery, further comprising: Based on the channel importance characteristics, the bias values ​​and weight values ​​of the compression channel and the excitation channel are set.

6. The method for constructing a mechanical equipment fault diagnosis model according to claim 1, characterized in that: Based on the training data set, the initial fault diagnosis model is iteratively trained until the target fault diagnosis model is obtained by satisfying the conditions, including: Preprocessing the vibration signal data to obtain a test data set; Based on the training data set, a cross entropy loss function is used as a loss function for model training, and the fault diagnosis model is trained through an optimizer and regularization; Optimizing the trained fault diagnosis model based on an attention mechanism, a weight decay strategy, and a learning rate strategy; Based on the test data set, the optimized fault diagnosis model is verified.

7. A mechanical equipment fault diagnosis method, characterized in that: include: Obtain vibration signal data of rotating machinery in real time; The vibration signal data is input into a fault diagnosis model to obtain the fault type of the rotating machinery; wherein the fault diagnosis model is obtained by using the mechanical equipment fault diagnosis model construction method according to any one of claims 1 to 6.

8. A mechanical equipment fault diagnosis model construction system, characterized in that: include: A preprocessing module, used to obtain vibration signal data of the rotating machinery, and preprocess the vibration signal data to obtain a training data set of the vibration signal data, wherein the training data set includes a fault type and a vibration signal corresponding to the fault type; A model training module is used to iteratively train the initial fault diagnosis model based on the training data set until the conditions are met to obtain the target fault diagnosis model, and each iterative training includes: processing the vibration signal data based on an improved convolutional neural network and an improved bidirectional gated recurrent unit to obtain the spatial characteristics and time series characteristics of the rotating machinery; fusing the spatial characteristics and the time series characteristics to determine the target feature vector of the rotating machinery; determining the predicted fault type corresponding to the target feature vector based on a preset classification decision rule; and updating the initial fault diagnosis model based on the predicted fault type and the standard fault type.

9. A mechanical equipment fault diagnosis system, characterized in that: include: An acquisition module, used for acquiring vibration signal data of a rotating machine in real time; A diagnostic module is used to input the vibration signal data into a fault diagnosis model to obtain the fault type of the rotating machinery; wherein the fault diagnosis model is obtained by using the mechanical equipment fault diagnosis model construction method as described in any one of claims 1 to 6.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the processor implements the steps of the mechanical equipment fault diagnosis model construction method described in any one of claims 1 to 6, or implements the steps of the mechanical equipment fault diagnosis method described in claim 7.

Images