Fault diagnosis method for multi-rate data of hydraulic system

Abstract

The invention relates to a fault diagnosis method for multi-rate data of a hydraulic system. The method comprises the steps that multi-sensor signals in the hydraulic system are obtained, wherein the multi-sensor signals comprise pressure, temperature and / or vibration signals collected by sensors of different types, and the multi-sensor signals are preprocessed; based on the preprocessed multi-sensor signal, obtaining a first fault feature corresponding to a high sampling rate signal and a second fault feature corresponding to a low sampling rate signal in the multi-sensor signal; fusing the first fault feature and the second fault feature to obtain a multi-rate fusion feature; and performing fault classification based on the multi-rate fusion features. According to the invention, accurate fault identification can be carried out on fault components of the hydraulic system, and the problem that multi-rate signals are difficult to input into a single neural network for reasoning training is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of hydraulic systems, and in particular to a fault diagnosis method for multi-rate data of hydraulic systems. Background Art

[0002] The Industrial Internet, with its powerful deep perception, intelligent analysis, efficient data processing and system integration capabilities, uses data analysis as the decision-making center and creates a series of intelligent solutions for intelligent manufacturing, networked collaboration, customized production and service expansion that are suitable for different industrial occasions. With the in-depth application and rapid progress of artificial intelligence, big data and 5G technologies, the Industrial Internet is gradually showing its important role in promoting high-quality economic growth in the manufacturing industry, promoting the coordinated development of advanced manufacturing and modern service industries, and accelerating the realization of the strategy of building a strong manufacturing country, becoming a key path and powerful driver to achieve these goals.

[0003] Industrial equipment constitutes the core element of the Industrial Internet and its platform. These key equipment play a vital role in industrial production. They are not only related to the smoothness of the production process, the stability of the industrial system, and the quality of products, but also directly affect the competitiveness of production enterprises. Over time, the uninterrupted operation of the equipment and the harsh working environment may eventually cause failures. Since industrial production is a complex interactive process, the failure of key equipment may cause product quality to decline and the defective rate to increase, which in turn affects the normal operation of the production system and may even cause serious safety accidents. Therefore, using advanced fault detection technology to obtain the fault status of equipment and conduct scientific and reasonable maintenance in a timely manner is a key measure to reduce time costs, prevent accidents, and enhance corporate profitability.

[0004] At present, data-driven intelligent fault diagnosis methods have been widely studied, and these studies are based on the data having the same sampling rate. However, in actual engineering, most hydraulic systems are complex combinations involving various hydraulic, mechanical and electrical components, and require multiple types of sensors to monitor the physical variables in the system, which makes the data monitored by the hydraulic system have multi-rate characteristics such as periodicity and information distribution asymmetry. Periodicity means that the variables at each sampling rate are continuous and complete; information distribution asymmetry means that there are differences in the information contained in variables at different sampling rates. In industrial systems, high-rate variables are mostly process variables, which usually contain limited process information, while low-rate variables are mostly quality-related variables, and the information in them may be more valuable.

[0005] The methods for fault detection using multi-rate data samples are mainly upsampling and downsampling. The upsampling method is to convert low-rate variables into high-rate signals using interpolation or other methods. The downsampling method is to sample high-rate variables at a lower sampling rate so that all variables obey the same sampling rate. However, both methods have inevitable disadvantages. In the upsampling method, the interpolated data may not reflect the original data distribution well. In the downsampling method, the information loss caused by the reduction in sampling rate may have adverse effects, especially when the sampling interval is large. Therefore, it is necessary to develop a new framework to process multi-rate data samples. Summary of the invention

[0006] To this end, the present invention provides a fault diagnosis method for multi-rate data of a hydraulic system, which can accurately identify faulty components of the hydraulic system and solve the problem that multi-rate signals are difficult to input into a single neural network for inference training.

[0007] In order to solve the above technical problems, the present invention provides a fault diagnosis method for multi-rate data of a hydraulic system, which is applied to a hydraulic system and comprises:

[0008] Acquire multi-sensor signals in the hydraulic system, wherein the multi-sensor signals include pressure, temperature and / or vibration signals collected by different types of sensors;

[0009] Preprocessing the multi-sensor signals;

[0010] Based on the preprocessed multi-sensor signal, obtaining a first fault feature corresponding to a high sampling rate signal and a second fault feature corresponding to a low sampling rate signal in the multi-sensor signal;

[0011] fusing the first fault feature and the second fault feature to obtain a multi-rate fusion feature;

[0012] Fault classification is performed based on the multi-rate fusion features.

[0013] In one embodiment of the present invention, the method for preprocessing the multi-sensor signal includes:

[0014] Filling missing data values ​​in the multi-sensor signal to ensure data integrity;

[0015] Use singular value decomposition method to denoise the padded data;

[0016] Based on the low sampling rate data in the multi-sensor signal after denoising, the same physical quantity monitored by multiple sensors of the same type at different positions is fused using variance contribution rate.

[0017] In one embodiment of the present invention, the method for obtaining a first fault feature corresponding to a high sampling rate signal and a second fault feature corresponding to a low sampling rate signal in the multi-sensor signal based on the preprocessed multi-sensor signal includes:

[0018] Processing the high sampling rate signal using a residual shrinkage separable convolutional network to obtain the first fault feature;

[0019] Perform feature extraction in the time domain and frequency domain on the low sampling rate signal to obtain the second fault feature.

[0020] In one embodiment of the present invention, a high sampling rate signal is processed using a residual shrinkage separable convolutional network to obtain the first fault feature, including:

[0021] The high sampling rate signal is processed by using the depth-separable convolution layer in the residual shrinkage separable convolutional network; wherein,

[0022] The expression of depth-wise separable convolution is as follows:

[0023]

[0024] Among them, x is the input of the deep convolutional layer; W d represents the convolution kernel; M and n represent the spatial dimensions of the convolution kernel; (i, j) is the position coordinate on the input feature; ⊙ represents element-wise multiplication;

[0025] In the point-by-point convolution layer, multiple convolution kernels of size 1 are used to convolve the feature map output by the depth convolution layer into a new feature map along its channel dimension, as shown in the following formula:

[0026]

[0027] Among them, x p Represents the input of the point-by-point convolutional layer; Represents the convolution kernel of point-by-point convolution; C and c give the number of channels in the input feature and output feature respectively; Represents the pixel value of the cth channel at position (i, j) of the input feature map;

[0028] Therefore, the calculation process of the depth-separable convolutional layer on the input data is expressed as:

[0029] SeparConv(x,W d ,W p )=PointwiseConv (i,j) (W p ,DepthwiseConv (i,j) (W d ,x)).

[0030] In one embodiment of the present invention, the method of fusing the first fault feature and the second fault feature to obtain a multi-rate fusion feature includes:

[0031] Sum the feature vectors of the first fault feature and the second fault feature to obtain a summed vector;

[0032] Constructing a multi-branch module based on a combination of the summed vector, the convolutional layer, and the global average pooling layer;

[0033] Activating the multi-branch module through a sigmoid activation function and multiplying it with the sum vector to obtain a fusion weight;

[0034] The multi-rate fusion feature is obtained based on the fusion weight, the first fault feature and the second fault feature.

[0035] In one embodiment of the present invention, the summation vector is calculated as follows:

[0036]

[0037] Among them, h(x) is the first fault, l(x) is the second fault feature, and Xout is the summation vector.

[0038] In one embodiment of the present invention, the calculation process of the fusion weight is as follows:

[0039]

[0040]

[0041] Among them, M(l(x), h(x)) is the fusion weight, MB is the multi-branch module, GAP is the global average pooling layer, Conv is the convolutional layer, and Xout is the summation vector.

[0042] In one embodiment of the present invention, the multi-rate fusion feature is calculated as follows:

[0043]

[0044] Among them, M(l(x), h(x)) is the fusion weight, and the dotted line is 1-M(l(x), h(x)).

[0045] In one embodiment of the present invention, the expression of the sigmoid activation function is as follows:

[0046]

[0047] In one embodiment of the present invention, fault classification is performed based on the multi-rate fusion feature, including:

[0048] Inputting the multi-rate fusion features into a multi-layer perceptron for fault classification;

[0049] Multi-class cross entropy is used as the loss function to evaluate the difference between the true category and the predicted category.

[0050] The above technical solution of the present invention has the following advantages compared with the prior art:

[0051] The fault diagnosis method for multi-rate data of hydraulic system described in the present invention accurately identifies the faulty components of hydraulic system, and solves the problem that multi-rate signals are difficult to input into a single neural network for inference training. In terms of feature fusion, a multi-rate feature fusion module based on gated attention mechanism is provided for fault feature information at different levels, and fault semantic information at different levels is aggregated through an adaptive weight allocation mechanism to provide the model with fault diagnosis accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to make the contents of the present invention more clearly understood, the present invention is further described in detail below based on specific embodiments of the present invention in conjunction with the accompanying drawings.

[0053] Figure 1 A flowchart of hydraulic system fault diagnosis based on multi-rate signals used in an embodiment of the present invention.

[0054] Figure 2 Schematic diagram of the structure of a depth-wise separable convolutional layer according to an embodiment of the present invention.

[0055] Figure 3 This is a structural diagram of the residual shrinkage separable convolution module according to an embodiment of the present invention.

[0056] Figure 4 This is a diagram of a multi-rate feature fusion module based on a gated attention mechanism according to an embodiment of the present invention.

[0057] Figure 5 Schematic diagram of a hydraulic system data collection test platform according to an embodiment of the present invention.

[0058] Figure 6 (a)-(d) are confusion matrix diagrams of fault diagnosis results of different components of the hydraulic system according to an embodiment of the present invention.

[0059] Figure 7 This is a signal diagram of adding different signal-to-noise ratios to the pressure signal according to an embodiment of the present invention.

[0060] Figure 8(a)-(d) are comparison diagrams of fault identification accuracy of four components, namely, cooler, valve, pump, and accumulator, under different signal-to-noise ratios using different methods of the embodiments of the present invention.

[0061] Fig. 9 (a)-(c) in the figure are diagrams of the mapping distribution of data after network transformation by visualizing data using T-SNE in an embodiment of the present invention. DETAILED DESCRIPTION

[0062] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it, but the embodiments are not intended to limit the present invention.

[0063] In the present invention, if directions (up, down, left, right, front and back) are described, it is only for the convenience of describing the technical solution of the present invention, and does not indicate or imply that the technical features referred to must have a specific orientation, be constructed and operate in a specific orientation. Therefore, it cannot be understood as a limitation of the present invention.

[0064] In the present invention, "several" means one or more, "multiple" means more than two, "greater than", "less than", "exceed" and the like are understood to exclude the number itself; "above", "below", "within" and the like are understood to include the number itself. In the description of the present invention, if there is a description of "first" or "second", it is only used for the purpose of distinguishing technical features, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features or implicitly indicating the order of the indicated technical features.

[0065] In the present invention, unless otherwise clearly defined, the words "set", "install", "connect" and the like should be understood in a broad sense, for example, they can be directly connected or indirectly connected through an intermediate medium; they can be fixedly connected or detachably connected or integrally formed; they can be mechanically connected or electrically connected or able to communicate with each other; they can be the internal connection of two elements or the interaction relationship between two elements. Those skilled in the art can reasonably determine the specific meanings of the above words in the present invention in combination with the specific content of the technical solution.

[0066] Reference Figure 1 As shown, a fault diagnosis method for multi-rate data of a hydraulic system in an embodiment of the present invention is applied to a hydraulic system, comprising:

[0067] S1. Acquire multi-sensor signals in a hydraulic system, wherein the multi-sensor signals include pressure, temperature and / or vibration signals collected by different types of sensors.

[0068] This embodiment uses a public hydraulic system data set to perform fault identification on the four components of the hydraulic system, namely, the cooler, valve, pump, and accumulator. Figure 5 , as shown in Table 1 and Table 2.

[0069] Table 1: Fault components and fault severity of the hydraulic system in this embodiment

[0070]

[0071] Table 2: Hydraulic system sensors and sensor sampling rates

[0072]

[0073] S2. Preprocess the multi-sensor signals to avoid loss of key information and eliminate the influence of redundant information to improve signal quality.

[0074] It should be noted that, in this embodiment, the multi-sensor signal includes a high sampling rate signal (such as not less than 100 Hz) and a low sampling rate signal (less than 100 Hz).

[0075] In other embodiments, specific sampling rate ranges of high and low sampling rate signals may also be divided accordingly.

[0076] Specifically, step S2 includes:

[0077] Filling missing values ​​in the multi-sensor signal to ensure data integrity; such as using interpolation, mean filling, or inference based on historical data;

[0078] The filled data is denoised using the singular value decomposition method; singular value decomposition decomposes the data matrix into multiple components, retains the main components, removes the noise components, and ensures the accuracy and reliability of the data;

[0079] Based on the low sampling rate data in the multi-sensor signal after denoising, the same physical quantity monitored by multiple sensors of the same type at different positions is fused using the variance contribution rate; by calculating the contribution rate to the overall variance, important information is retained and redundant data is eliminated, thereby ensuring that key feature information is effectively retained and improving signal quality.

[0080] S3, based on the preprocessed multi-sensor signal, obtaining a first fault feature corresponding to a high sampling rate signal and a second fault feature corresponding to a low sampling rate signal in the multi-sensor signal. Specifically comprising:

[0081] Processing a high sampling rate signal (such as a 100 Hz pressure signal) using a residual shrinkage separable convolutional network (module) to obtain the first fault feature;

[0082] The low sampling rate signal is subjected to feature extraction in the time domain (mean, standard deviation, skewness) and frequency domain (such as power spectrum density, etc.) to obtain the second fault feature.

[0083] Residual shrinkage separable convolution module reference Figure 3 As shown in the figure, the residual structure is used to modularly build a residual shrinkage separable convolutional network, which has fewer parameters and faster inference speed, and is used to capture the deep fault semantic features of the system's high-rate data. At the same time, the statistical feature module extracts the fused low sampling rate signal for time and frequency characteristic analysis to obtain multi-angle fault features.

[0084] Reference Figure 3 As shown, the residual shrinkage separable convolution module includes:

[0085] Separable Convolution: Separable convolution is usually divided into depth convolution and point-by-point convolution, which splits the calculation of standard convolution into two steps to reduce the computational complexity. Through these convolutional layers, multi-scale feature extraction can be performed on the input features.

[0086] Activation function (Leaky ReLU);

[0087] Batch Normalization (BN): used to standardize the features of the convolutional layer output and reduce the internal covariate shift during training.

[0088] Adaptive Average Pooling (AdaptiveAvg Pool): Reduce the input feature map to the specified output size;

[0089] Fully connected layer (FC): used to perform further linear mapping on the extracted features.

[0090] Sigmoid function: activation function, used to compress the output result into the range of [0,1].

[0091] The network structure combines the information on different feature paths by weighting or directly adding / multiplying them through addition, multiplication, and threshold operations.

[0092] The input high sampling rate data (X) passes through multiple residual shrinkage separable convolution block modules, and the module contains multiple separable convolution layers and Leaky ReLU activation to gradually extract more abstract fault features.

[0093] The features are normalized through the batch normalization layer, and then high-level features are extracted.

[0094] At the same time, in another path, the input features undergo adaptive average pooling and then generate feature weights through a fully connected layer and a Sigmoid function.

[0095] Finally, the network weights or fuses the features from different paths, processes the information, and outputs y.

[0096] Deep features of high-rate data (such as rapidly changing signals in a pressure sensor, which can indicate signs of sudden failure in a hydraulic system).

[0097] The time and frequency domain characteristics of low-rate data (such as long-term trend changes in temperature and vibration signals, which can indicate the hidden dangers of chronic faults).

[0098] Reference Figure 2 As shown in the figure, the high sampling rate signal is processed by the depthwise separable convolutional layer in the residual shrinkage separable convolutional network. Replacing the depthwise separable convolutional layer with the traditional convolutional layer can reduce the number of network parameters and achieve network lightweight.

[0099] Among them, the expression of depth-separable convolution is as follows:

[0100]

[0101] Among them, x is the input of the deep convolutional layer; W d represents the convolution kernel; M and n represent the spatial dimensions of the convolution kernel; (i, j) is the position coordinate on the input feature; ⊙ represents element-wise multiplication;

[0102] In the point-by-point convolution layer, multiple convolution kernels of size 1 are used to convolve the feature map output by the depth convolution layer into a new feature map along its channel dimension, as shown in the following formula:

[0103]

[0104] Among them, x p Represents the input of the point-by-point convolutional layer; Represents the convolution kernel of point-by-point convolution; C and c give the number of channels in the input feature and output feature respectively; Represents the pixel value of the cth channel at position (i, j) of the input feature map;

[0105] Therefore, the calculation process of the depth-separable convolutional layer on the input data is expressed as:

[0106] SeparConv(x,W d ,W p)=PointwiseConv (i,j) (W p ,DepthwiseConv (i,j) (W d ,x)).

[0107] S4. Fuse the first fault feature and the second fault feature to obtain a multi-rate fusion feature. Figure 4 As shown, in this embodiment, a multi-rate feature fusion module based on gated attention is used to solve the multi-rate feature fusion problem, specifically including:

[0108] Sum the feature vectors of the first fault feature and the second fault feature to obtain a summed vector;

[0109] Constructing a multi-branch module based on a combination of the summed vector, the convolutional layer, and the global average pooling layer;

[0110] Activating the multi-branch module through a sigmoid activation function and multiplying it with the sum vector to obtain a fusion weight;

[0111] The multi-rate fusion feature is obtained based on the fusion weight, the first fault feature and the second fault feature.

[0112] Specifically, the summation vector is calculated as follows:

[0113]

[0114] Among them, h(x) is the first fault, l(x) is the second fault feature, and Xout is the summation vector.

[0115] Specifically, the calculation process of the fusion weight is as follows:

[0116]

[0117]

[0118] Among them, M(l(x), h(x)) is the fusion weight, MB is the multi-branch module, GAP is the global average pooling layer, Conv is the convolutional layer, and Xout is the summation vector.

[0119] Specifically, the multi-rate fusion feature is calculated as follows:

[0120]

[0121] Among them, M(l(x), h(x)) is the fusion weight, and the dotted line ( Figure 4) is 1-M(l(x), h(x)), and M(l(x), h(x)) is a fusion weight consisting of real numbers between 0 and 1.

[0122] Specifically, the attention mechanism can adaptively determine the weight of the input. The expression of the sigmoid activation function is as follows:

[0123]

[0124] The sigmoid function can be used to convert the output value of the classification into a probability distribution ranging from [0, 1] and 1. The input x = [x1, x2] is decomposed into importance weights α = [α1, α2] and multiplied with the original input. The weight of the input is adaptively determined to improve the effect of the prediction model.

[0125] A multi-rate feature fusion module based on a gated attention mechanism is used to capture fault information at different levels. At the same time, an adaptive weight mechanism is used to aggregate feature information at different levels to improve the effect of model diagnosis: the attention mechanism can adaptively determine the input weight.

[0126] The use of residual shrinkage separable convolutional networks improves the generalization and robustness of the network: residual connections can effectively alleviate the problems of gradient vanishing and gradient exploding, improve the training efficiency of the network, and facilitate the modular design of the network, thereby increasing the feature extraction ability of the network. The shrinkage block is a structure that enhances useful information and suppresses redundant information by introducing a soft thresholding mechanism.

[0127] S5. Based on the multi-rate fusion features, fault classification is performed. Specifically, it includes:

[0128] The multi-rate fusion features are input into a multi-layer perceptron (MLP) for fault classification; the multi-layer perceptron processes the features step by step through fully connected layers, learns complex patterns in the data, and thus performs fault classification;

[0129] During the training process, multi-class cross entropy is used as the loss function to evaluate the difference between the true category and the predicted category to evaluate the model.

[0130] The training loss is defined as the difference between the labeled true and predicted categories, with multi-class cross entropy as the loss function. The training loss is calculated by the difference between the true category and the model's predicted category. The goal of the model is to continuously reduce these differences, thereby improving the accuracy of fault classification. Multi-class cross entropy is used as a loss function to guide the model to optimize classification performance.

[0131] Through the above method, when processing sampling rate imbalance (such as the sampling ratio of high-rate data to low-rate data is 1:100), a high fault diagnosis accuracy can still be maintained.

[0132] from Figure 6 It can be clearly seen that in the diagnosis of coolers and valves, the classification accuracy reaches 100%, while in the fault identification of pumps and accumulators, only a few samples are misplaced into adjacent categories.

[0133] Table 3 shows a comparison of the fault diagnosis accuracy of different methods for the four components of the hydraulic system cooler, valve, pump, and accumulator. It can be found that compared with Multi-rate CNN, the method of the present invention improves the pump and accumulator fault detection by 0.57% and 0.20% respectively.

[0134] Table 3: Fault diagnosis accuracy of hydraulic system components by different methods

[0135]

[0136] In order to verify the accuracy of the model in a noisy environment, an additive Gaussian white noise with a power of Pn is added to the original signal to simulate a noisy environment. The power of the noise is proportional to the power of the original signal Ps, and the signal-to-noise ratio (SNR) is introduced to describe the ratio of Ps to Pn. The calculation of the signal-to-noise ratio and power is:

[0137]

[0138]

[0139] Depend on Figure 7 The data distribution of pressure signals under different signal-to-noise ratios is shown. The XGBoost and CNN algorithms are used to compare with the framework. Figure 8 As shown. In general, the accuracy of fault detection increases with the increase of signal-to-noise ratio. The noise environment has different effects on the fault diagnosis results of different components. In the detection of coolers and valves, as the signal-to-noise ratio increases, the detection accuracy of the three methods changes little. In the fault diagnosis of pumps and accumulators with more complex fault scenarios, the noise environment has a greater impact on the diagnosis accuracy. However, compared with XGBoost and CNN, this framework is less affected. As the signal-to-noise ratio increases, the accuracy of the framework is close to that in a noise-free environment. In the fault diagnosis of hydraulic accumulators, when the signal-to-noise ratio is 15db, the accuracy of this framework is far different from that of XGBoost and CNN, which proves the robustness of the proposed framework.

[0140] To visualize the effectiveness of the proposed framework, Fig. 9t-SNE is used to visualize the distribution of cooler test samples in two-dimensional space. It can be found that in this figure, data points of different categories are separated from each other, indicating that the framework has strong nonlinear mapping capabilities and can effectively identify fault categories.

[0141] By comparing with other models and methods, the effectiveness of this method is verified.

[0142] Aiming at the hydraulic system failure problem of multi-rate signals, the present invention proposes a fault diagnosis framework for multi-rate data of hydraulic system, and conducts simulation experiments based on public hydraulic system data. Through the experiments, the effectiveness of the algorithm model in the present invention is proved.

[0143] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0144] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0145] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0146] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0147] Finally, it should be noted that the above specific implementation methods are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to examples, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention, which should be included in the scope of the claims of the present invention.

Claims

1. A fault diagnosis method for multi-rate data of a hydraulic system, applied to a hydraulic system, characterized in that: include: Acquire multi-sensor signals in the hydraulic system, wherein the multi-sensor signals include pressure, temperature and / or vibration signals collected by different types of sensors; Preprocessing the multi-sensor signals; Based on the preprocessed multi-sensor signal, obtaining a first fault feature corresponding to a high sampling rate signal and a second fault feature corresponding to a low sampling rate signal in the multi-sensor signal; fusing the first fault feature and the second fault feature to obtain a multi-rate fusion feature; Fault classification is performed based on the multi-rate fusion features.

2. A fault diagnosis method for multi-rate data of a hydraulic system according to claim 1, characterized in that: The method for preprocessing the multi-sensor signal comprises: Filling missing data values ​​in the multi-sensor signal to ensure data integrity; Use singular value decomposition method to denoise the padded data; Based on the low sampling rate data in the multi-sensor signal after denoising, the same physical quantity monitored by multiple sensors of the same type at different positions is fused using variance contribution rate.

3. A fault diagnosis method for multi-rate data of a hydraulic system according to claim 1, characterized in that: The method for obtaining a first fault feature corresponding to a high sampling rate signal and a second fault feature corresponding to a low sampling rate signal in the multi-sensor signal based on the preprocessed multi-sensor signal includes: Processing the high sampling rate signal using a residual shrinkage separable convolutional network to obtain the first fault feature; Perform feature extraction in the time domain and frequency domain on the low sampling rate signal to obtain the second fault feature.

4. A fault diagnosis method for multi-rate data of a hydraulic system according to claim 3, characterized in that: The high sampling rate signal is processed by using a residual shrinkage separable convolutional network to obtain the first fault feature, including: The high sampling rate signal is processed by using the depth-separable convolution layer in the residual shrinkage separable convolutional network; wherein, The expression of depth-wise separable convolution is as follows: Among them, x is the input of the deep convolutional layer; W d Represents the convolution kernel; M and n represent the spatial dimensions of the convolution kernel; (i, j) is the position coordinate on the input feature; ⊙ represents element-wise multiplication; In the point-by-point convolution layer, multiple convolution kernels of size 1 are used to convolve the feature map output by the depth convolution layer into a new feature map along its channel dimension, as shown in the following formula: Among them, x p Represents the input of the point-by-point convolutional layer; W c p Represents the convolution kernel of point-by-point convolution; C and c give the number of channels in the input feature and output feature respectively; Represents the pixel value of the cth channel at position (i, j) of the input feature map; Therefore, the calculation process of the depth-separable convolutional layer on the input data is expressed as: SeparConv(x,W d ,W p )=PointwiseConv (i,j) (W p ,DepthwiseConv (i,j) (W d ,x)).

5. A fault diagnosis method for multi-rate data of a hydraulic system according to claim 1, characterized in that: The method of fusing the first fault feature and the second fault feature to obtain a multi-rate fusion feature includes: Sum the feature vectors of the first fault feature and the second fault feature to obtain a summed vector; Constructing a multi-branch module based on a combination of the summed vector, the convolutional layer, and the global average pooling layer; Activating the multi-branch module through a sigmoid activation function and multiplying it with the sum vector to obtain a fusion weight; The multi-rate fusion feature is obtained based on the fusion weight, the first fault feature and the second fault feature.

6. A fault diagnosis method for multi-rate data of a hydraulic system according to claim 5, characterized in that: The summation vector is calculated as follows: Among them, h(x) is the first fault, l(x) is the second fault feature, and Xout is the summation vector.

7. A fault diagnosis method for multi-rate data of a hydraulic system according to claim 5, characterized in that: The calculation process of the fusion weight is as follows: Among them, M(l(x), h(x)) is the fusion weight, MB is the multi-branch module, GAP is the global average pooling layer, Conv is the convolutional layer, and Xout is the summation vector.

8. A fault diagnosis method for multi-rate data of a hydraulic system according to claim 5, characterized in that: The multi-rate fusion feature is calculated as follows: Among them, M(l(x), h(x)) is the fusion weight.

9. A fault diagnosis method for multi-rate data of a hydraulic system according to claim 5, characterized in that: The expression of the sigmoid activation function is as follows:

10. A fault diagnosis method for multi-rate data of a hydraulic system according to claim 1, characterized in that: Based on the multi-rate fusion features, fault classification is performed, including: Inputting the multi-rate fusion features into a multi-layer perceptron for fault classification; Multi-class cross entropy is used as the loss function to evaluate the difference between the true category and the predicted category.

Images