A motor fault detection method and system based on a multi-branch neural network

Abstract

The application discloses a motor fault detection method and system based on a multi-branch neural network, wherein the multi-branch neural network in the method is obtained through the following steps: obtaining a multi-dimensional data set from a motor, and preprocessing the multi-dimensional data set to obtain a motor training set; inputting the motor training set into a local branch framework to obtain a first branch feature set; inputting the motor training set into a global branch framework to obtain a second branch feature set; inputting the motor training set into a time sequence branch framework to obtain a third branch feature set; performing attention fusion on the second branch feature set and the third branch feature set according to the first branch feature set to obtain a fusion data set; and performing parameter updating on an initialized multi-branch neural network according to the fusion data set to obtain a trained multi-branch neural network. The detection method can effectively improve the detection efficiency and detection accuracy of motor faults. The application relates to the technical field of deep learning.

Description

Technical Field

[0001] This invention relates to the field of deep learning technology, and in particular to a method and system for detecting motor faults based on a multi-branch neural network. Background Technology

[0002] As a typical rotating machine, electric motors often operate under high speed and high load conditions. Timely fault detection of electric motors is beneficial to ensuring the stability and safety of industrial production.

[0003] Currently, traditional motor fault detection methods mainly use Support Vector Machine (SVM) and Random Forest models to diagnose motor faults. These methods often rely on manually extracted features, which are inefficient. In addition, real industrial environments are often accompanied by non-stationary operating conditions such as start-stop, speed change, and impact loads. These non-stationary operating conditions can easily lead to misdiagnosis of motors, resulting in low accuracy.

[0004] Therefore, the problems existing in the current technology still need to be solved and optimized. Summary of the Invention

[0005] The purpose of this invention is to at least partially solve one of the technical problems existing in the related art.

[0006] Therefore, one objective of this invention is to provide a motor fault detection method based on a multi-branch neural network, which can effectively improve the detection efficiency and accuracy of motor faults.

[0007] Another objective of this application is to provide a motor fault detection system based on a multi-branch neural network.

[0008] To achieve the above-mentioned technical objectives, the technical solutions adopted in the embodiments of this application include:

[0009] In a first aspect, embodiments of this application provide a motor fault detection method based on a multi-branch neural network, including:

[0010] Acquire motor signal data;

[0011] The motor signal data is input into a trained multi-branch neural network to obtain motor fault detection results.

[0012] The multi-branch neural network includes a local branch framework, a global branch framework, and a time-series branch framework. These three frameworks are connected in parallel. The trained multi-branch neural network is obtained through the following steps:

[0013] Obtain a multidimensional dataset from the motor and preprocess the multidimensional dataset to obtain the motor training set;

[0014] The motor training set is input into the local branch framework to obtain a first branch feature set, which is used to characterize the set of local feature information corresponding to the motor training set.

[0015] The motor training set is input into the global branch framework to obtain the second branch feature set, which is used to characterize the set of global context information corresponding to the motor training set.

[0016] The motor training set is input into the time series branching framework to obtain the third branch feature set, which is used to characterize the set of time series information corresponding to the motor training set.

[0017] Based on the first branch feature set, attention fusion is performed on the second branch feature set and the third branch feature set to obtain a fused dataset;

[0018] Based on the fused dataset, the parameters of the initialized multi-branch neural network are updated to obtain the trained multi-branch neural network.

[0019] In addition, the motor fault detection method according to the above embodiments of this application may also have the following additional technical features:

[0020] Furthermore, in one embodiment of this application, the multidimensional dataset includes several subsets of sensor data, each subset corresponding to a different vibration sensor. The preprocessing of the multidimensional dataset to obtain the motor training set includes:

[0021] According to the diffusion entropy calculation function, the weights of the sensor data subsets are calculated to obtain the sensor weights corresponding to each sensor data subset.

[0022] Based on the sensor weights, all the sensor data subsets are fused to obtain a fused dataset;

[0023] Noise is added to the fused dataset to obtain the motor training set.

[0024] Furthermore, in one embodiment of this application, the global branching framework includes several global context shrinking networks, which are connected in series. The global context shrinking networks are used to perform the following steps:

[0025] Obtain the input dataset and perform a nonlinear transformation on the input dataset to obtain a linearly transformed dataset;

[0026] Global context features are extracted from the linear transformation dataset to obtain the global dataset;

[0027] Based on the linear transformation dataset, noise reduction and extraction are performed on the global dataset to obtain the output dataset corresponding to the global context shrinking network.

[0028] Furthermore, in one embodiment of this application, the global context feature extraction is performed on the linear transformation dataset to obtain a global dataset:

[0029] The linear transformation dataset is combined with its features to obtain a combined dataset;

[0030] A first bottleneck transformation is performed on the combined dataset to obtain a first bottleneck dataset, and a second bottleneck transformation is performed on the combined dataset to obtain a second bottleneck dataset;

[0031] The second bottleneck dataset is subjected to soft thresholding to obtain the first soft threshold dataset;

[0032] The global dataset is obtained by element-wise multiplication of the first bottleneck dataset based on the first soft threshold dataset.

[0033] Further, in one embodiment of this application, the step of performing noise reduction and extraction on the global dataset based on the linear transformation dataset to obtain the output dataset corresponding to the global context shrinking network includes:

[0034] Based on the global dataset, soft thresholding is performed on the linear transformation dataset to obtain a second soft threshold dataset;

[0035] Based on the linear transformation dataset, the second soft threshold dataset is concatenated to obtain the output dataset.

[0036] Further, in this embodiment of the application, the step of performing attention fusion on the second branch feature set and the third branch feature set based on the first branch feature set to obtain a fused dataset includes:

[0037] Based on the third branch feature set, the second branch feature set is concatenated to obtain the concatenated feature set;

[0038] Based on the spliced ​​feature set, the first branch feature set is fused with attention weighting to obtain the fused dataset.

[0039] Secondly, embodiments of this application provide a motor fault detection system based on a multi-branch neural network, comprising:

[0040] The acquisition module is used to acquire motor signal data;

[0041] The processing module is used to input the motor signal data into a trained multi-branch neural network to obtain motor fault detection results;

[0042] The multi-branch neural network includes a local branch framework, a global branch framework, and a time-series branch framework. These three frameworks are connected in parallel. The trained multi-branch neural network is obtained through the following steps:

[0043] Obtain a multidimensional dataset from the motor and preprocess the multidimensional dataset to obtain the motor training set;

[0044] The motor training set is input into the local branch framework to obtain a first branch feature set, which is used to characterize the set of local feature information corresponding to the motor training set.

[0045] The motor training set is input into the global branch framework to obtain the second branch feature set, which is used to characterize the set of global context information corresponding to the motor training set.

[0046] The motor training set is input into the time series branching framework to obtain the third branch feature set, which is used to characterize the set of time series information corresponding to the motor training set.

[0047] Based on the first branch feature set, attention fusion is performed on the second branch feature set and the third branch feature set to obtain a fused dataset;

[0048] Based on the fused dataset, the parameters of the initialized multi-branch neural network are updated to obtain the trained multi-branch neural network.

[0049] Furthermore, in this embodiment of the application, the global branching framework includes several global context shrinking networks, which are connected in series. The global context shrinking networks are used to perform the following steps:

[0050] Obtain the input dataset and perform a nonlinear transformation on the input dataset to obtain a linearly transformed dataset;

[0051] Global context features are extracted from the linear transformation dataset to obtain the global dataset;

[0052] Based on the linear transformation dataset, noise reduction and extraction are performed on the global dataset to obtain the output dataset corresponding to the global context shrinking network.

[0053] Thirdly, embodiments of this application also provide an electronic device, including:

[0054] At least one processor;

[0055] At least one memory for storing at least one program;

[0056] When the at least one program is executed by the at least one processor, the at least one processor implements the method of the first aspect described above.

[0057] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a processor-executable program, which, when executed by the processor, is used to implement the method of the first aspect described above.

[0058] The advantages and beneficial effects of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application:

[0059] This application discloses a motor fault detection method and system based on a multi-branch neural network. The detection method acquires motor signal data; inputs the motor signal data into a trained multi-branch neural network to obtain a motor fault detection result. The multi-branch neural network includes a local branch framework, a global branch framework, and a time-series branch framework, which are connected in parallel. The trained multi-branch neural network is obtained through the following steps: acquiring a multidimensional dataset from the motor and preprocessing the multidimensional dataset to obtain a motor training set; inputting the motor training set into the local branch framework to obtain a first branch feature set; and so on. A first branch feature set is used to represent the set of local feature information corresponding to the motor training set. The motor training set is input into the global branch framework to obtain a second branch feature set, which represents the set of global context information corresponding to the motor training set. The motor training set is input into the time series branch framework to obtain a third branch feature set, which represents the set of time series information corresponding to the motor training set. Based on the first branch feature set, attention fusion is performed on the second branch feature set and the third branch feature set to obtain a fused dataset. Based on the fused dataset, the parameters of the initialized multi-branch neural network are updated to obtain the trained multi-branch neural network. This detection method, based on three parallel local branch frameworks, a global branch framework, and a time series branch framework, can extract accurate and reliable feature information more quickly. By performing attention fusion on the feature information output by the three branch frameworks, it effectively improves the detection efficiency and accuracy of motor faults. Attached Figure Description

[0060] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the following description is provided with accompanying drawings of the relevant technical solutions in the embodiments of this application or the prior art. It should be understood that the accompanying drawings described below are only for the purpose of clearly illustrating some embodiments of the technical solutions in this application. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0061] Figure 1 A schematic diagram of a motor fault detection method based on a multi-branch neural network provided in an embodiment of this application;

[0062] Figure 2 A schematic diagram of a training process for a multi-branch neural network provided in an embodiment of this application;

[0063] Figure 3 A schematic diagram of the structure of a motor fault detection system based on a multi-branch neural network provided in an embodiment of this application;

[0064] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0065] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application. The step numbers in the following embodiments are set only for ease of explanation, and there is no limitation on the order between the steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0066] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0067] Currently, traditional motor fault detection methods mainly employ Support Vector Machines (SVM) and Random Forest models for motor fault diagnosis. These methods often rely on manually extracted features, resulting in low efficiency. Furthermore, real-world industrial environments typically involve non-stationary operating conditions such as start-stop, speed changes, and impact loads, introducing significant noise into the acquired motor signals. These methods have limited performance in handling strong noise signals, easily leading to misdiagnosis and low accuracy. Additionally, since single-sensor signals can be affected by noise interference, compromising the integrity of the information, the accuracy and reliability of motor fault diagnosis relying on a single sensor are often unsatisfactory.

[0068] In view of this, embodiments of the present invention provide a motor fault detection method based on a multi-branch neural network. This detection method is based on the preprocessing of a multidimensional dataset, specifically calculating the sensor weights corresponding to each sensor through a diffusion entropy calculation function, and performing data fusion based on the sensor weights. This can fully leverage the correlation and complementarity of data collected by multiple sensors, significantly improving the reliability of sensor data fusion. Furthermore, this detection method is based on parallel training of local branch frameworks, global branch frameworks, and time-series branch frameworks, and fuses the feature sets output by the three branch frameworks based on an attention mechanism. This allows it to provide relatively accurate and reliable diagnostic information even under the influence of high-noise environments, effectively improving the accuracy and efficiency of motor fault detection.

[0069] Reference Figure 1 and Figure 2 In this embodiment of the application, a motor fault detection method based on a multi-branch neural network includes:

[0070] Step 110: Acquire motor signal data;

[0071] Step 120: Input the motor signal data into the trained multi-branch neural network to obtain the motor fault detection result;

[0072] In this embodiment, the signal data about the motor collected by the sensor can be input into a trained multi-branch neural network. The trained multi-branch neural network can then predict whether the motor has a fault, thereby obtaining a motor fault detection result that records whether the motor has a fault.

[0073] The multi-branch neural network includes a local branch framework, a global branch framework, and a time-series branch framework. These three frameworks are connected in parallel. The trained multi-branch neural network is obtained through the following steps:

[0074] Step 130: Obtain the multidimensional dataset from the motor and preprocess the multidimensional dataset to obtain the motor training set;

[0075] In some embodiments, the multidimensional dataset includes several subsets of sensor data, each subset corresponding to a different vibration sensor. The preprocessing of the multidimensional dataset to obtain a motor training set includes:

[0076] A1. Calculate the weights of the sensor data subsets according to the diffusion entropy calculation function to obtain the sensor weights corresponding to each sensor data subset.

[0077] A2. Based on the sensor weights, perform data fusion on all the sensor data subsets to obtain a fused dataset;

[0078] A3. Add noise to the fused dataset to obtain the motor training set.

[0079] In this embodiment, the multidimensional data set of the motor can be obtained by deploying multiple high-precision vibration sensors at different locations on the motor, such as bearings, bases, or transmission components. Each vibration sensor converts mechanical vibration into electrical signals through its built-in sensitive element. These electrical signals are then processed through preliminary amplification and filtering to obtain the multidimensional data set.

[0080] Understandably, the dispersion entropy calculation function is used to describe the degree of irregularity in a pattern sequence. Specifically, for the electrical signal sampling points corresponding to a certain vibration sensor, its time series X = {x1, x2, ..., x...} j ,…,x N}, x j Let be the time subsequence of the j-th electrical signal sampling point, and N be the number of sampling points. Map this time subsequence to a new sequence Y using a normal distribution function, Y = {y}. j},j=1,2,…,N,y j ∈(0,1);

[0081]

[0082] Where μ is the expected value, σ is the standard deviation, and y j For x j The corresponding new time subsequence; t is the integration variable, and exp is the natural exponential function.

[0083] Then for y j Perform a linearization transformation so that y j When assigned to an integer class, the equivalent formula for linearization transformation can be expressed as:

[0084]

[0085] in, For y j The corresponding integer class ranges from [1,2,…,c]; int is the rounding function, and c is the class number.

[0086] Calculate the embedding vector for each integer class. Each embedding vector with embedding dimension m and time delay d is created according to the following formula.

[0087]

[0088] Each embedding vector Mapping to distribution pattern in d represents the time delay, m represents the embedding dimension, and c represents the embedding dimension. m This represents the total number of potential scattering patterns when the embedding dimension is m.

[0089] For each scattering pattern, the formula for calculating its relative frequency is:

[0090]

[0091] Where Num is the summation function used to solve for the scattering pattern. Assigned to the embedding vector Quantity; This refers to the relative frequency.

[0092] The equivalent expression for the diffusion entropy calculation function can be:

[0093]

[0094] Where DE(X) is the diffusion entropy calculation function, and ln represents the natural logarithm.

[0095] Understandably, a larger DE value indicates greater sequence complexity and a higher degree of irregularity. Specifically, step A1 can first calculate the diffusion entropy (DE) value of the subset of sensor data corresponding to each vibration sensor. For a specific acquired signal within the subset of sensor data, the equivalent expression for its diffusion entropy value is:

[0096] B i =DE(X) i ), i = 1 ~ n

[0097] Among them, B i X is the dispersion entropy value corresponding to the i-th acquired vibration signal. i DE() is the diffusion entropy calculation function, and n is the total number of vibration sensors.

[0098] After obtaining the dispersion entropy value of all signals collected by each sensor, the redundancy of the corresponding vibration sensor can be calculated based on the set of dispersion entropy values ​​corresponding to each subset of sensor data. The equivalent expression for the redundancy is:

[0099] D i =|1-B i |,i=1~n

[0100] Among them, D i This represents the redundancy of the dispersion entropy value of the i-th vibration sensor.

[0101] By allocating weights to the subsets of sensor data corresponding to each vibration sensor based on the redundancy, the equivalent expression for the sensor weight of a certain subset of sensor data is:

[0102]

[0103] Among them, w i It is the sensor weight of the i-th subset of sensor data.

[0104] It should be noted that the data fusion in step A2 can be based on the sensor weights corresponding to each subset of sensor data, performing weighted fusion on the subsets to obtain a fused dataset corresponding to multiple subsets of sensor data. Step A3 can first obtain a preset signal-to-noise ratio (SNR) value, and then, based on this preset SNR value, use a random noise generation algorithm to generate a noise signal of corresponding intensity. The generated noise signal is then superimposed on the fused dataset to obtain the motor training set.

[0105] Step 140: Input the motor training set into the local branch framework to obtain the first branch feature set, which is used to characterize the set of local feature information corresponding to the motor training set;

[0106] In this embodiment of the application, the local branch framework is used to extract deep feature representations from the motor training set and capture local features and detailed information in the motor training set. It may include multiple residual neural networks (ResNet), which are connected in series. The output of the previous residual neural network is used as the input of the next residual neural network, and the output of the last residual neural network is used as the first branch feature set.

[0107] Step 150: Input the motor training set into the global branch framework to obtain the second branch feature set. The second branch feature set is used to characterize the set of global context information corresponding to the motor training set.

[0108] In this embodiment, the global branching framework sequentially includes an input layer, a convolutional (Conv) layer, multiple global context shrinking networks, a batch normalization layer, a ReLU activation layer, a global average pooling (GAP) layer, and a fully connected (FC) layer. The motor training set undergoes adaptive noise reduction and signal feature extraction through the Conv layer and multiple global context shrinking networks.

[0109] In some embodiments, step 150, the global branching framework, includes several global context shrinking networks, which are connected in series. The global context shrinking networks are used to perform the following steps:

[0110] B1. Obtain the input dataset and perform a nonlinear transformation on the input dataset to obtain a linearly transformed dataset;

[0111] In this embodiment of the application, the input dataset of the first global context shrinking network in the series relationship can be the motor training set; while the input data of the second and subsequent global context shrinking networks in the series relationship can be the output dataset of the previous global context shrinking network.

[0112] Understandably, step B1 could involve inputting the input dataset into two cascaded Conv layers in a global context shrinking network for non-linear transformation, and using the features output by the second Conv layer as the linearly transformed dataset.

[0113] B2. Perform global context feature extraction on the linear transformation dataset to obtain the global dataset;

[0114] Further, in step B2, global contextual features are extracted from the linear transformation dataset to obtain a global dataset:

[0115] B21. Perform feature combination on the linear transformation dataset to obtain a combined dataset;

[0116] B22. Perform a first bottleneck transformation on the combined dataset to obtain a first bottleneck dataset, and perform a second bottleneck transformation on the combined dataset to obtain a second bottleneck dataset;

[0117] B23. Perform soft thresholding on the second bottleneck dataset to obtain the first soft threshold dataset;

[0118] B24. Based on the first soft threshold dataset, perform element-wise multiplication on the first bottleneck dataset to obtain the global dataset.

[0119] In this embodiment of the application, step B21 is used to combine all features in the linear transformation dataset into global context features. Specifically, the feature combination in step B21 can be based on a 1×1 convolutional layer to perform a nonlinear transformation on the linear transformation dataset, then use the Softmax activation function to obtain the corresponding attention weights, and finally obtain a combined dataset representing global context feature information through attention pooling.

[0120] Understandably, the first and second bottleneck transformations in step B22 are used to capture the interdependencies between channels, thereby compressing the dimensions of global context features and reducing the burden on multi-branch neural networks. Specifically, the first bottleneck transformation in step B22 can be to input the combined dataset sequentially into a 1×1 convolutional layer, a layer norm, a 1×1 convolutional layer, a ReLU layer, an absolute value function layer, and a global bottleneck pooling (GAP) layer to obtain the first bottleneck dataset; the second bottleneck transformation in step B22 can be to input the combined dataset into another bottleneck transformation branch, which consists of a 1×1 convolutional layer, a layer norm, a 1×1 convolutional layer, and a ReLU layer sequentially to obtain the second bottleneck dataset.

[0121] Step B23, the soft thresholding process, can involve sequentially inputting the second bottleneck dataset into an absolute value function layer, a global bottleneck pooling (GAP) layer, and two fully connected layers. Then, the output of the second fully connected layer is scaled to the range (0, 1) using the sigmoid function to obtain the first soft threshold dataset. Step B24 can involve performing element-wise multiplication of each feature element in the first soft threshold dataset with each feature element in the first bottleneck dataset to obtain the global dataset.

[0122] B3. Based on the linear transformation dataset, perform noise reduction and extraction on the global dataset to obtain the output dataset corresponding to the global context shrinking network.

[0123] Further, step B3, denoising and extracting the global dataset based on the linear transformation dataset to obtain the output dataset corresponding to the global context shrinking network, includes:

[0124] B31. Based on the global dataset, perform soft thresholding on the linear transformation dataset to obtain a second soft threshold dataset;

[0125] B32. Based on the linear transformation dataset, the second soft threshold dataset is concatenated to obtain the output dataset.

[0126] In this embodiment, the global dataset in step B31 is used to provide dynamic soft thresholds for different channels to the soft thresholding module in the global branch framework. This soft thresholding module sets features within a threshold interval to zero, achieving noise suppression. Specifically, the soft threshold extraction in step B31 can involve inputting a linear transformation dataset into an updated soft thresholding module. The threshold interval in this updated soft thresholding module is determined based on the latest received global dataset, thus obtaining a second soft threshold dataset. Step B32 can be based on a shortcut identity connection, performing element-wise addition on each feature element in the linear transformation dataset and each feature element in the second soft threshold dataset to obtain the output dataset.

[0127] It should be noted that for the last global context shrinking network in the chain, its output dataset can be sequentially input into the batch normalization layer, ReLU activation layer, global average pooling (GAP) layer and fully connected (FC) layer to obtain the second branch feature set output by the global branch framework.

[0128] Step 160: Input the motor training set into the time series branching framework to obtain the third branch feature set, which is used to characterize the set of time series information corresponding to the motor training set;

[0129] In this embodiment, the time series branching framework can be constructed based on the Grey Wolf Algorithm (GWO) to optimize the gated recurrent unit (GRU) neural network. By applying the global search capability of the Grey Wolf Algorithm (GWO) to the parameter optimization of the gated recurrent unit (GRU), the optimal gate parameters of the gated recurrent unit are found. Then, based on the update gate and reset gate in the gated recurrent unit with the optimal gate parameters, the long-term dependency information contained in the sequence data of the input motor training set is captured, thereby obtaining a set with time series information of the motor training set (i.e., the third branch feature set).

[0130] Step 170: Based on the first branch feature set, perform attention fusion on the second branch feature set and the third branch feature set to obtain a fused dataset;

[0131] In some embodiments, step 170, performing attention fusion on the second branch feature set and the third branch feature set based on the first branch feature set to obtain a fused dataset, includes:

[0132] C1. Based on the third branch feature set, perform feature concatenation on the second branch feature set to obtain a concatenated feature set;

[0133] C2. Based on the spliced ​​feature set, the first branch feature set is fused with attention weighting to obtain the fused dataset.

[0134] In this embodiment, the first branch feature set can be used as the key and value, the second branch feature set as the query representation, and the third branch feature set as the dynamic time series weights in the weighted fusion process of the attention mechanism. Specifically, step C1 can be a concatenation operation on the third branch feature set and the second branch feature set to obtain a concatenated feature set; step C2 can be using the concatenated feature set as the first input of the attention mechanism module, the first branch feature set as the second input of the attention mechanism module, and calculating the attention weights and performing weighted fusion to obtain the fused dataset.

[0135] Step 180: Update the parameters of the initialized multi-branch neural network according to the fused dataset to obtain the trained multi-branch neural network.

[0136] Understandably, for machine learning models, the accuracy of model predictions can be measured by a loss function. A loss function is defined on a single training data point and measures the prediction error of that data point. Specifically, the loss value is determined by the label of the individual training data point and the model's prediction result for that data. However, in actual training, a training dataset contains many data points. Therefore, a cost function is generally used to measure the overall error of the training dataset. The cost function is defined on the entire training dataset and calculates the average prediction error of all training data points, providing a better measure of the model's prediction performance. For general machine learning models, the aforementioned cost function, plus a regularization term to measure model complexity, can serve as the training objective function. Based on this objective function, the loss value of the entire training dataset can be calculated. Many types of loss functions are commonly used, such as 0-1 loss, squared loss, absolute loss, logarithmic loss, and cross-entropy loss, which will not be elaborated upon here. In this embodiment, any one of these loss functions can be selected to determine the training loss value, such as the cross-entropy loss function. Based on the training loss value, the backpropagation algorithm is used to update the model parameters. After several iterations, a trained multi-branch neural network can be obtained. The specific number of iterations can be preset, or training can be considered complete when the accuracy requirement is met on the test set.

[0137] The following describes in detail, with reference to the accompanying drawings, a motor fault detection system based on a multi-branch neural network according to an embodiment of this application.

[0138] Reference Figure 3 The motor fault detection system based on a multi-branch neural network proposed in this application includes:

[0139] Acquisition module 101 is used to acquire motor signal data;

[0140] Processing module 102 is used to input the motor signal data into a trained multi-branch neural network to obtain motor fault detection results;

[0141] The multi-branch neural network includes a local branch framework, a global branch framework, and a time-series branch framework. These three frameworks are connected in parallel. The trained multi-branch neural network is obtained through the following steps:

[0142] Obtain a multidimensional dataset from the motor and preprocess the multidimensional dataset to obtain the motor training set;

[0143] The motor training set is input into the local branch framework to obtain a first branch feature set, which is used to characterize the set of local feature information corresponding to the motor training set.

[0144] The motor training set is input into the global branch framework to obtain the second branch feature set, which is used to characterize the set of global context information corresponding to the motor training set.

[0145] The motor training set is input into the time series branching framework to obtain the third branch feature set, which is used to characterize the set of time series information corresponding to the motor training set.

[0146] Based on the first branch feature set, attention fusion is performed on the second branch feature set and the third branch feature set to obtain a fused dataset;

[0147] Based on the fused dataset, the parameters of the initialized multi-branch neural network are updated to obtain the trained multi-branch neural network.

[0148] In some embodiments, the global branching framework includes several global context shrinking networks, which are connected in series. The global context shrinking networks are used to perform the following steps:

[0149] Obtain the input dataset and perform a nonlinear transformation on the input dataset to obtain a linearly transformed dataset;

[0150] Global context features are extracted from the linear transformation dataset to obtain the global dataset;

[0151] Based on the linear transformation dataset, noise reduction and extraction are performed on the global dataset to obtain the output dataset corresponding to the global context shrinking network.

[0152] It is understood that the content of the above method embodiments is applicable to this system embodiment. The specific functions implemented in this system embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0153] Reference Figure 4 This application also provides an electronic device, including:

[0154] At least one processor 201;

[0155] At least one memory 202 is used to store at least one program;

[0156] When the at least one program is executed by the at least one processor 201, the at least one processor 201 implements the method embodiment described above.

[0157] Similarly, it can be understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0158] This application also provides a computer-readable storage medium storing a program executable by a processor 201, which, when executed by the processor 201, is used to implement the above-described method embodiments.

[0159] Similarly, the content of the above method embodiments is applicable to the present computer-readable storage medium embodiments. The specific functions implemented by the present computer-readable storage medium embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0160] In some alternative embodiments, the functions / operations mentioned in the block diagrams may not occur in the order shown in the operation diagrams. For example, depending on the functions / operations involved, two consecutively shown blocks may actually be executed substantially simultaneously, or the blocks may sometimes be executed in reverse order. Furthermore, the embodiments presented and described in the flowcharts of this application are provided by way of example to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is changed and sub-operations described as part of a larger operation are executed independently.

[0161] Furthermore, although this application is described in the context of functional modules, it should be understood that, unless otherwise stated to the contrary, one or more of the functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding this application. Rather, given the properties, functions, and internal relationships of the various functional modules in the apparatus disclosed herein, the actual implementation of the module will be understood within the scope of conventional technology for an engineer. Therefore, those skilled in the art can implement the application set forth in the claims using ordinary techniques without excessive experimentation. It is also understood that the specific concepts disclosed are merely illustrative and not intended to limit the scope of this application, which is determined by the full scope of the appended claims and their equivalents.

[0162] If a function is implemented as 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 this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods in the embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0163] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0164] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which programs can be printed, because programs can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0165] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0166] In the foregoing description of this specification, the references to terms such as "one embodiment," "another embodiment," or "some embodiments," etc., indicate that a specific feature, structure, material, or characteristic described in connection with an embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0167] Although embodiments of this application have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the claims and their equivalents.

[0168] The above is a detailed description of the preferred embodiments of this application, but this application is not limited to the embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of this application, and these equivalent modifications or substitutions are all included within the scope defined by the claims of this application.

Claims

1. A motor fault detection method based on a multi-branch neural network, characterized in that, include: Acquire motor signal data; The motor signal data is input into a trained multi-branch neural network to obtain motor fault detection results. The multi-branch neural network includes a local branch framework, a global branch framework, and a time-series branch framework. These three frameworks are connected in parallel. The trained multi-branch neural network is obtained through the following steps: Obtain a multidimensional dataset from the motor and preprocess the multidimensional dataset to obtain the motor training set; The motor training set is input into the local branch framework to obtain a first branch feature set, which is used to characterize the set of local feature information corresponding to the motor training set. The motor training set is input into the global branch framework to obtain the second branch feature set, which is used to characterize the set of global context information corresponding to the motor training set. The motor training set is input into the time series branching framework to obtain the third branch feature set, which is used to characterize the set of time series information corresponding to the motor training set. Based on the first branch feature set, attention fusion is performed on the second branch feature set and the third branch feature set to obtain a fused dataset; Based on the fused dataset, the parameters of the initialized multi-branch neural network are updated to obtain the trained multi-branch neural network. The global branching framework includes several global context shrinking networks, which are interconnected. These global context shrinking networks are used to perform the following steps: Obtain the input dataset and perform a nonlinear transformation on the input dataset to obtain a linearly transformed dataset; Global context features are extracted from the linear transformation dataset to obtain the global dataset; Based on the linear transformation dataset, noise reduction and extraction are performed on the global dataset to obtain the output dataset corresponding to the global context shrinking network; The global context feature extraction is performed on the linear transformation dataset to obtain the global dataset: The linear transformation dataset is combined with its features to obtain a combined dataset; A first bottleneck transformation is performed on the combined dataset to obtain a first bottleneck dataset, and a second bottleneck transformation is performed on the combined dataset to obtain a second bottleneck dataset; The second bottleneck dataset is subjected to soft thresholding to obtain the first soft threshold dataset; The global dataset is obtained by element-wise multiplication of the first bottleneck dataset based on the first soft threshold dataset.

2. The motor fault detection method according to claim 1, characterized in that, The multidimensional dataset includes several subsets of sensor data, each subset corresponding to a different vibration sensor. The preprocessing of the multidimensional dataset to obtain the motor training set includes: According to the diffusion entropy calculation function, the weights of the sensor data subsets are calculated to obtain the sensor weights corresponding to each sensor data subset. Based on the sensor weights, all the sensor data subsets are fused to obtain a fused dataset; Noise is added to the fused dataset to obtain the motor training set.

3. The motor fault detection method according to claim 1, characterized in that, The step of denoising and extracting the global dataset based on the linear transformation dataset to obtain the output dataset corresponding to the global context shrinking network includes: Based on the global dataset, soft thresholding is performed on the linear transformation dataset to obtain a second soft threshold dataset; Based on the linear transformation dataset, the second soft threshold dataset is concatenated to obtain the output dataset.

4. The motor fault detection method according to claim 1, characterized in that, The step of performing attention fusion on the second branch feature set and the third branch feature set based on the first branch feature set to obtain a fused dataset includes: Based on the third branch feature set, the second branch feature set is concatenated to obtain the concatenated feature set; Based on the spliced ​​feature set, the first branch feature set is fused with attention weighting to obtain the fused dataset.

5. A motor fault detection system based on a multi-branch neural network, characterized in that, include: The acquisition module is used to acquire motor signal data; The processing module is used to input the motor signal data into a trained multi-branch neural network to obtain motor fault detection results; The multi-branch neural network includes a local branch framework, a global branch framework, and a time-series branch framework. These three frameworks are connected in parallel. The trained multi-branch neural network is obtained through the following steps: Obtain a multidimensional dataset from the motor and preprocess the multidimensional dataset to obtain the motor training set; The motor training set is input into the local branch framework to obtain a first branch feature set, which is used to characterize the set of local feature information corresponding to the motor training set. The motor training set is input into the global branch framework to obtain the second branch feature set, which is used to characterize the set of global context information corresponding to the motor training set. The motor training set is input into the time series branching framework to obtain the third branch feature set, which is used to characterize the set of time series information corresponding to the motor training set. Based on the first branch feature set, attention fusion is performed on the second branch feature set and the third branch feature set to obtain a fused dataset; Based on the fused dataset, the parameters of the initialized multi-branch neural network are updated to obtain the trained multi-branch neural network. The global branching framework includes several global context shrinking networks, which are interconnected. These global context shrinking networks are used to perform the following steps: Obtain the input dataset and perform a nonlinear transformation on the input dataset to obtain a linearly transformed dataset; Global context features are extracted from the linear transformation dataset to obtain the global dataset; Based on the linear transformation dataset, noise reduction and extraction are performed on the global dataset to obtain the output dataset corresponding to the global context shrinking network; The global context feature extraction is performed on the linear transformation dataset to obtain the global dataset: The linear transformation dataset is combined with its features to obtain a combined dataset; A first bottleneck transformation is performed on the combined dataset to obtain a first bottleneck dataset, and a second bottleneck transformation is performed on the combined dataset to obtain a second bottleneck dataset; The second bottleneck dataset is subjected to soft thresholding to obtain the first soft threshold dataset; The global dataset is obtained by element-wise multiplication of the first bottleneck dataset based on the first soft threshold dataset.

6. An electronic device, characterized in that, include: At least one processor; At least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the method as described in any one of claims 1-4.

7. A computer-readable storage medium storing a processor-executable program, characterized in that, The processor-executable program, when executed by the processor, is used to implement the method as described in any one of claims 1-4.

Images