Equipment state diagnosis method and system of multi-channel fusion neural network, electronic equipment and computer readable medium

Abstract

The invention discloses a multi-channel fusion neural network equipment state diagnosis method and system, electronic equipment and a computer readable medium, and the method comprises the steps: collecting data of a sensor disposed on the equipment, carrying out the preprocessing, and constructing equipment state multi-channel data; inputting the equipment state multi-channel data into a multi-component signal analysis model, wherein an empirical mode decomposition model is used to obtain an equipment state decomposition signal; inputting the equipment state decomposition signal into a non-negative matrix decomposition model, and separating out an independent source equipment state signal; performing time-frequency domain feature extraction on the independent source equipment state signal in combination with improved Hilbert-Huang transform to obtain a time-frequency domain feature result, and reconstructing an equipment state time-frequency graph from the time-frequency domain feature result; and inputting the equipment state time-frequency graph into the multi-scale sparse time-frequency convolutional network model to obtain an equipment state diagnosis result. According to the invention, equipment fault diagnosis is realized by acquiring the multi-channel data of the equipment, and the diagnosis accuracy is improved.

Description

Technical Field

[0001] The present invention belongs to the field of equipment status fault diagnosis, and in particular relates to an equipment status diagnosis method and system of a multi-channel fusion neural network, an electronic device, and a computer-readable medium. Background Art

[0002] With the rapid development of science and technology, the number of industrial equipment in the world has exploded. According to data from market research firm Statista, the number of industrial Internet of Things (IIoT) devices in the world has exceeded 15 billion in 2022, and it is expected that by 2030, the number will exceed 30 billion. The huge scale of equipment poses unprecedented challenges to the condition monitoring and management of equipment. In order to ensure the normal operation of these devices, reduce downtime and reduce maintenance costs, equipment condition diagnosis technology urgently needs to be improved, and will gradually become an important pillar in the era of smart manufacturing and Industry 4.0. Equipment condition diagnosis is a key technology that determines the health status of equipment and predicts potential failures by real-time monitoring and analysis of equipment operation data. According to McKinsey, companies can reduce maintenance costs by 20%-30% and reduce equipment downtime by about 50% by implementing effective equipment condition diagnosis and predictive maintenance. In modern industrial environments, the improvement of fault diagnosis technology can help companies save millions of dollars in costs and significantly improve production efficiency.

[0003] Traditional equipment maintenance models usually rely on regular inspections and manual maintenance, and their limitations are obvious. For example, the global manufacturing industry suffers direct losses of more than $5 billion each year due to equipment failures. A large part of these losses is caused by the failure to detect potential problems in the equipment in a timely manner. Therefore, how to use advanced technical means to monitor the real-time status of equipment and conduct accurate diagnosis has become the focus of attention of major companies. With the advancement of sensor technology, equipment status diagnosis technology has made great progress. Nowadays, most industrial equipment is equipped with various sensors, such as temperature sensors, pressure sensors, and vibration sensors. These sensors can collect various key data in the operation of the equipment in real time and transmit the data to the central processing system for analysis through the Internet of Things platform. Using advanced data analysis technologies, such as big data analysis and machine learning algorithms, equipment status diagnosis can deeply mine and analyze massive amounts of equipment data to identify abnormal behaviors and potential failure modes of equipment. With the rapid development of deep learning technology, its application in equipment status diagnosis has received widespread attention. By constructing a multi-layer neural network, feature extraction and pattern recognition of complex data, it has shown great potential in the field of equipment fault diagnosis. Many studies have shown that deep learning algorithms can automatically extract features from large amounts of equipment data and accurately classify and predict the health status of equipment. For example, convolutional neural networks (CNNs) and long short-term memory networks (LSTMs) have been widely used in fault diagnosis of mechanical equipment, and can effectively identify problems such as bearing failures and motor anomalies. However, in existing equipment status diagnosis, the analysis of multiple sensor data is isolated, resulting in poor correlation analysis between sensor data, and the original analysis is in the time domain, making it difficult to analyze data characteristics at a higher dimensional level, making the accuracy of equipment fault status diagnosis challenging.

[0004] Therefore, due to the problems of existing traditional methods and artificial intelligence methods such as the correlation of multi-channel sensor data of equipment fault status and the effective mining of fault status in the time domain, a new fault status diagnosis method is urgently needed to further improve the accuracy of fault diagnosis. Summary of the invention

[0005] In order to solve the above technical problems and overcome the shortcomings of the prior art, the present invention provides a device status diagnosis method and system of a multi-channel fusion neural network, an electronic device, and a computer-readable medium, which introduces a fault diagnosis method of a multi-channel fusion neural network, so that a better recognition effect can be obtained in the field of equipment fault diagnosis, and the diagnostic performance can be greatly improved; it can improve the accuracy of fault detection, reduce maintenance costs, and improve the operational safety and reliability of equipment in the factory. In addition, the implementation of this method can also provide valuable data support for equipment manufacturers and maintenance service providers, helping to optimize product design and maintenance strategies.

[0006] The technical solution of the present invention is as follows:

[0007] According to a first aspect, the present invention provides a device status diagnosis method of a multi-channel fusion neural network, comprising:

[0008] Step S1: collecting sensor data arranged at key parts of the equipment and preprocessing them to construct multi-channel data of equipment status;

[0009] Step S2: inputting the multi-channel data of the device state into a multi-component signal analysis model, the multi-component signal analysis model comprising an empirical mode decomposition model and a non-negative matrix decomposition model, wherein the empirical mode decomposition model is used to obtain a device state decomposition signal from the multi-channel data of the device state; inputting the device state decomposition signal into the non-negative matrix decomposition model to separate an independent source device state signal; performing time-frequency domain feature extraction on the independent source device state signal in combination with an improved Hilbert-Huang transform to obtain a time-frequency domain feature result, and reconstructing a device state time-frequency diagram from the time-frequency domain feature result;

[0010] Step S3: Input the device status time-frequency diagram into a multi-scale sparse time-frequency convolutional network model to obtain a device status diagnosis result.

[0011] According to a second aspect, a device status diagnosis system of a multi-channel fusion neural network is provided, comprising the following modules:

[0012] Equipment failure multi-channel data module: collects sensor data arranged at key parts of the equipment, performs pre-processing, and constructs multi-channel data of equipment status;

[0013] The device status diagnosis module of the multi-channel fusion neural network: the multi-channel data of the device status is input into a multi-component signal analysis model, the multi-component signal analysis model includes an empirical mode decomposition model and a non-negative matrix decomposition model, wherein the empirical mode decomposition model is used to obtain a device status decomposition signal from the multi-channel data of the device status; the device status decomposition signal is input into the non-negative matrix decomposition model to separate the independent source device status signal; the time-frequency domain feature extraction of the independent source device status signal is performed in combination with the improved Hilbert-Huang transform to obtain the time-frequency domain feature result, and the device status time-frequency diagram is reconstructed from the time-frequency domain feature result; the device status time-frequency diagram is input into a model based on a multi-scale sparse time-frequency convolutional network to obtain the device status diagnosis result.

[0014] According to a third aspect, the present invention provides an electronic device, comprising a memory and a processor, wherein the memory is used to store a computer program executable by the processor, and when the processor executes the computer program, it implements the aforementioned device status diagnosis method of a multi-channel fusion neural network.

[0015] According to a fourth aspect, the present invention provides a computer-readable medium having a computer program stored thereon, and when the computer program is executed, the aforementioned device status diagnosis method of a multi-channel fusion neural network is implemented.

[0016] Compared with the prior art, the present invention has the following beneficial effects:

[0017] (1) The present invention provides a device status diagnosis method or system based on a multi-channel fusion neural network, which can identify the device fault status. For the first time, through the multi-channel fusion method, the multi-dimensional information provided by multiple sensors can be integrated to make up for the fault characteristics that may be missed by a single signal source; this method enhances the comprehensiveness and robustness of the device fault diagnosis system, enabling the device status diagnosis to cope with more complex and diverse fault modes.

[0018] (2) The present invention provides a device status diagnosis method or system based on a multi-channel fusion neural network, which helps to capture the changes in the signal in both time and frequency dimensions by converting time series data into a time-frequency diagram. This method is more intuitive and effective than traditional time domain or frequency domain analysis, especially when processing non-stationary signals. It can more accurately identify potential fault characteristics and improve the accuracy of equipment fault diagnosis.

[0019] (3) The present invention provides a device status diagnosis method or system based on a multi-channel fusion neural network, which can extract key features of the device operating status at different time and frequency scales through the application of a multi-scale sparse time-frequency convolutional network. Compared with traditional convolutional neural networks, multi-scale convolutional networks can more comprehensively capture the multi-scale features of device failures, significantly improving the sensitivity and accuracy of the equipment fault diagnosis system.

[0020] (4) The present invention provides a method or system for diagnosing equipment status using a multi-channel fusion neural network. The design of multi-scale sparse convolution makes the model more efficient when processing high-dimensional data, thus avoiding the problem of excessive computational complexity. At the same time, sparse connections reduce the risk of overfitting and enhance the generalization ability of the model under different working conditions, thereby improving the reliability and applicability of the equipment fault diagnosis system in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 It is a flow chart of a device status diagnosis method of a multi-channel fusion neural network in an embodiment of the present invention;

[0022] Figure 2 This is a flow chart of step S2 in a device status diagnosis method of a multi-channel fusion neural network in an embodiment of the present invention;

[0023] Figure 3This is a flow chart of step S22 in a device status diagnosis method of a multi-channel fusion neural network in an embodiment of the present invention;

[0024] Figure 4 This is a flow chart of step S23 in a device status diagnosis method of a multi-channel fusion neural network in an embodiment of the present invention;

[0025] Figure 5 This is a flow chart of step S3 in a device status diagnosis method of a multi-channel fusion neural network in an embodiment of the present invention;

[0026] Figure 6 It is a structural block diagram of a device status diagnosis system of a multi-channel fusion neural network in an embodiment of the present invention. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is further described in detail below through specific implementations and in conjunction with the accompanying drawings.

[0028] Embodiment 1

[0029] Figure 1 FIG. 1 is a flow chart of a device status diagnosis method of a multi-channel fusion neural network in an embodiment of the present invention. Figure 1 As shown, the method includes:

[0030] Step S1: Collect sensor data arranged on the equipment and perform preprocessing to construct multi-channel data of equipment status.

[0031] The device status diagnosis method of the multi-channel fusion neural network provided by the present invention is based on the sensor measurement signal data of the device failure. The installed sensors include vibration sensors, sound sensors, temperature sensors, current and voltage sensors, position sensors and load sensors; the device status data of the above sensors are collected and preprocessing operations are used, including filter denoising, normalization, missing value supplementation, to construct multi-channel data of the device status. This method enhances the comprehensiveness and robustness of the diagnostic system.

[0032] Step S2: inputting the multi-channel data of the device state into a multi-component signal analysis model, the multi-component signal analysis model comprising an empirical mode decomposition model and a non-negative matrix decomposition model, wherein the empirical mode decomposition model is used to obtain a device state decomposition signal from the multi-channel data of the device state; inputting the device state decomposition signal into the non-negative matrix decomposition model to separate the independent source device state signal from the device state decomposition signal; extracting the time-frequency domain features of the independent source device state signal in combination with an improved Hilbert-Huang transform to obtain a time-frequency domain feature result, and reconstructing a device state time-frequency diagram from the time-frequency domain feature result;

[0033] Among them, the empirical mode decomposition model is an adaptive signal time-frequency processing method. In essence, it is a method for decomposing signals, decomposing signals into the superposition of independent components, and decomposing signals according to the time scale characteristics of the data itself. The core of the non-negative matrix decomposition model is to decompose a non-negative matrix into two smaller non-negative matrices. The content of the model is well known to those skilled in the art and will not be elaborated here.

[0034] Step S3: Input the device status time-frequency diagram into a multi-scale sparse time-frequency convolutional network model to obtain a device status diagnosis result.

[0035] The device status time-frequency diagram is input into the multi-scale sparse time-frequency convolutional network model to effectively improve the accuracy of device status diagnosis. The design of multi-scale sparse convolution makes the model more efficient when processing high-dimensional data and avoids the problem of excessive computational complexity.

[0036] like Figure 2 As shown, in one embodiment, the above step S2: input the device state multi-channel data into a multi-component signal analysis model, the multi-component signal analysis model includes an empirical mode decomposition model and a non-negative matrix decomposition model, wherein the empirical mode decomposition model is used to obtain a device state decomposition signal from the device state multi-channel data; input the device state decomposition signal into the non-negative matrix decomposition model to separate the independent source device state signal; extract the time-frequency domain features of the independent source device state signal in combination with the improved Hilbert-Huang transform to obtain the time-frequency domain feature results, and reconstruct the device state time-frequency diagram from the time-frequency domain feature results, including:

[0037] Step S21: The device status multi-channel data , input into the empirical mode decomposition model, decomposed to obtain the trend term and the intrinsic mode functions ,in ; and, according to the following formula (1), and Perform addition reconstruction to obtain the device state decomposition signal ;

[0038] (1)

[0039] Step S22: Decomposing the device status signal Input into the non-negative matrix factorization model to separate the independent source device status signals ;

[0040] Step S23: The independent source device status signal The improved Hilbert-Huang transform model is used to extract time-frequency domain features to obtain the time-frequency domain feature results, and the device status time-frequency diagram is reconstructed from the time-frequency domain feature results.

[0041] Among them, the empirical mode decomposition model decomposes the preprocessed multi-channel data of the equipment state to obtain the intrinsic mode functions (IMFs). The purpose of this process is to decompose the complex equipment state signal into several simple, single-frequency oscillation modes; the decomposed signal obtained by the empirical mode decomposition model is input into the non-negative matrix decomposition model, which can separate independent source signals; through the non-negative matrix decomposition model, the independent characteristic signals of the equipment state can be identified and extracted; the independent source equipment state signal separated by the non-negative matrix decomposition model is combined with the improved Hilbert-Huang transform (HHT) to extract time-frequency domain features; the improved HHT can accurately capture the local characteristics of the signal and improve the resolution of time-frequency analysis. The equipment state time-frequency diagram is reconstructed from the time-frequency domain feature results, including: reconstructing the extracted time-frequency domain feature results to generate a two-dimensional time-frequency image of the equipment state, and performing a pooling operation on the two-dimensional time-frequency image features to compress the feature information and highlight the most important features, and fusing the equipment state time-frequency diagram of multi-scale fault characteristics through splicing operations, which provides an intuitive visual representation and rich feature information for subsequent fault diagnosis.

[0042] like Figure 3 As shown, in one embodiment, the above step S22: inputting the device state decomposition signal into the non-negative matrix decomposition model to separate and obtain the independent source device state signal includes:

[0043] Step S221: Define the problem and initialize: Decompose the device state signal through the non-negative matrix factorization model Decomposed into two non-negative matrices and , so that ,in is the original non-negative matrix, that is, the approximate device state decomposition signal; randomly initialize the two matrices and and ensure and is a non-negative matrix;

[0044] Step S222: Iterative update: fixed matrix , by optimizing the matrix To minimize the loss function, and use the gradient descent update rule to update the training weights; fix the matrix , by optimizing the matrix To minimize the loss function, and use the gradient descent update rule to update the training weights;

[0045] Step S223: Convergence judgment: After each iteration, the current loss function value is calculated; when the change of the loss function is less than a preset threshold, or the maximum number of iterations is reached, whichever condition is reached first, the algorithm converges and meets the convergence condition, and the algorithm is terminated;

[0046] Step S224: Reconstructing the independent source device status signal: Matrix Each column in is represented as an independent source signal. The columns in Multiply and reconstruct the independent source device status signal .

[0047] Before inputting the device state decomposition signal into the non-negative matrix factorization (NMF) model, the device state decomposition signal needs to be preprocessed to ensure the signal quality and reduce interference factors in data analysis; the intrinsic mode functions (IMFs) obtained by decomposition are constructed into a data matrix, in which each row represents an IMF and each column represents the signal value at a time point; in the NMF model, two non-negative matrices need to be initialized, namely the basis matrix and the coefficient matrix ; Basis matrix Each column of represents a potential signal component, and the coefficient matrix Each row of represents the weight of these components in the device state decomposition signal; when the NMF model converges, from the basis matrix The independent components are extracted from the original signal, which represent the independent source device status signals. Each independent source device status signal is a component of the original signal and reflects the behavior of the device at a specific frequency.

[0048] like Figure 4 As shown, in one embodiment, the above step S23: the independent source device status signal The improved Hilbert-Huang transform model is used to extract time-frequency domain features to obtain the time-frequency domain feature results, and the device state time-frequency diagram is reconstructed from the time-frequency domain feature results, including:

[0049] Step S231: The independent source device status signal Add white noise , get the independent source device status signal with added noise disturbance ;

[0050] Step S232: Add the independent source device status signal with noise disturbance Perform empirical mode decomposition model decomposition to obtain the intrinsic mode function , and then take the eigenmode function The average value of , according to the following formula (2), the eigenmode function is obtained The ensemble average decomposition signal :

[0051] (2)

[0052] in, It is After adding noise IMF , is the total number of times noise is added.

[0053] Step S233: According to the following formula (3), for each Calculate permutation entropy at different time scales , select the eigenmode function with the lowest permutation entropy As a valid signal;

[0054] (3)

[0055] in, Represents the dimension used in constructing the vector of eigenmode functions when performing phase space reconstruction. is the time scale, is the number of possible orderings of different elements. Is the arrangement The probability of occurrence, indicating that each permutation corresponds to a specific ordering of the subsequences after phase space reconstruction, where It is a specific arrangement.

[0056] Step S234: The intrinsic mode function Use the improved normalized Hilbert transform to improve the accuracy of the instantaneous frequency and reduce the constraints of the Bedrosian theorem through the normalization operation, that is, calculate the normalized instantaneous phase and instantaneous frequency ;

[0057] in, Represents the Hilbert transform model obtained at time The analytical signal of Represents the analysis signal The real part of Represents the analysis signal The modulus value, Represents the arccosine function, used to calculate angles; represents the derivative of the normalized instantaneous phase with respect to time;

[0058] Step S235: The instantaneous frequency Perform statistical analysis, use average value operation to extract time-frequency domain features, obtain time-frequency domain feature extraction results, and use the feature splicing and fusion method to reconstruct the device status time-frequency feature map;

[0059] Among them, in order to improve the performance of the Hilbert-Huang transform model (HHT), the present invention introduces an improvement measure of the normalization operation and designs an improved Hilbert transform model to improve the accuracy of instantaneous frequency estimation and capture the time-frequency characteristics of the signal; after applying the improved HHT to each IMF, a series of instantaneous frequencies and instantaneous amplitudes are obtained. These time-frequency domain features can reflect the changes of the device status signal at different times and frequencies. From the extracted time-frequency domain features, the most representative and informative features are selected. The time-frequency diagram of the device status is reconstructed using the selected time-frequency domain features. This step involves mapping the two-dimensional time-frequency data (time-frequency) to a graph to intuitively display the behavior of the device status signal at different times and frequencies. The reconstructed device status time-frequency feature graph is analyzed and features are extracted to identify the changing trend of the device status and potential failure modes. The training process of the Hilbert-Huang transform model includes: using the cross entropy loss function, according to the following formula (4), in which the sparse regularization term is combined, the model training optimization is performed:

[0060] (4)

[0061] in, Indicates the category type, that is, the number of categories of device status in the classification problem (such as normal equipment, faulty equipment, etc.), Indicates The true label of the class, Indicates The predicted probability of the class, represents the regularization coefficient, represents the number of layers of the convolutional network, Take the logarithmic function, Indicates Layer The weight matrix of the convolution kernel, No. The type of layer.

[0062] like Figure 5 As shown, in one embodiment, the above step S3: using a multi-scale sparse time-frequency convolutional network model to obtain a device status diagnosis result for the device status time-frequency diagram specifically includes:

[0063] Step S31: performing a convolution operation on the device state time-frequency feature graph using a multi-scale convolution kernel, and obtaining the device state time-frequency feature according to the following formula (5);

[0064] (5)

[0065] in, represents the convolution output, It is Tier The convolution kernel of channels, is the feature vector of the time-frequency feature map, is the convolution operation weight vector OK, is the convolution operation weight vector OK, is the feature vector of the time-frequency feature map OK, is the feature vector of the time-frequency feature map List, is the activation function, is the bias term;

[0066] Step S32: using sparse regularization on the device state time-frequency feature graph, according to the following formula (6), the convolution layer output is sparsely processed to obtain a sparse convolution output result;

[0067] (6)

[0068] in, represents the regularization coefficient, Indicates Tier The convolution kernel of channels;

[0069] Step S33: The sparse convolution output results are fused using a feature cascade operation to obtain a fused feature map result according to the following formula (7): ;

[0070] (7)

[0071] in, Indicates Feature maps of different scales;

[0072] Step S34: Introduce the feature map result into the attention mechanism, and perform weighted processing on the fused feature map according to the following formula (8) to obtain the feature map after attention weighting ;

[0073] (8)

[0074] in, Indicates the first , The scoring function of the position feature, It is a feature map No. OK, It is a feature map No. List;

[0075] Step S35: The attention-weighted feature map Max pooling is used for multi-scale downsampling, and the downsampled feature map is input into the fully connected layer, where the fully connected layer flattens the feature map into a one-dimensional feature vector, outputs the probability of the fault category through the softmax activation function, and performs fault diagnosis of the equipment status.

[0076] Among them, the multi-scale convolutional network can extract multi-scale features from different resolutions of the time-frequency graph through convolution kernels of different sizes, and capture the state changes of the equipment in different time-frequency ranges; convolution kernels of different scales can extract features of different time scales, and capture the equipment status of rapid changes in the short term or long-term trends; in order to ensure the sparsity of the features, the weight of the convolution kernel is limited by introducing L1 regularization, so that the output feature map is more sparse, so that only the key features related to the fault or health state are retained. Each layer of the convolutional network extracts time-frequency features of different scales. In order to generate a comprehensive state representation of the equipment, the features of different scales are fused. In order to further improve the focus on key features, the attention mechanism is adopted to enable the model to better focus on the important parts of the time-frequency graph. The attention mechanism assigns different weights to different positions on the time-frequency graph to highlight the characteristic areas of the equipment's operating status. After attention weighting, multi-scale downsampling is used to reduce the dimension of the feature map while retaining important features; and the pooling operation can compress the amount of data, reduce the computational complexity, and retain key information.

[0077] Embodiment 2

[0078] like Figure 6 As shown, a device status diagnosis system of a multi-channel fusion neural network includes the following modules:

[0079] Constructing a multi-channel data module 41 for equipment failure, collecting sensor data arranged at key parts of the equipment, and pre-processing it to construct multi-channel data of equipment status;

[0080] The device status diagnosis module 42 of the multi-channel fusion neural network is used to input the multi-channel data of the device status into a multi-component signal analysis model, wherein the device status decomposition signal is obtained by using the empirical mode decomposition model; the device status decomposition signal is input into a non-negative matrix decomposition model to separate the independent source device status signal; the time-frequency domain features of the independent source device status signal are extracted in combination with the improved Hilbert-Huang transform to obtain the time-frequency domain feature results, and the device status time-frequency diagram is reconstructed from the time-frequency domain feature results; the device status time-frequency diagram is subjected to a multi-scale sparse time-frequency convolutional network model to obtain the device status diagnosis result.

[0081] Optionally, the present invention also provides an electronic device (computer, server, smart phone, network device, etc.), including a memory and a processor, wherein the memory is used to store a computer program executable by the processor, and when the processor executes the computer program, the above-mentioned method embodiment or device embodiment is implemented.

[0082] Optionally, the present invention also provides a program product, such as a computer-readable storage medium, including a program, which is used to execute the above method embodiment or device embodiment when executed by a processor.

[0083] The above embodiments are provided only for the purpose of describing the present invention, and are not intended to limit the scope of the present invention. The scope of the present invention is defined by the appended claims. Various equivalent substitutions and modifications made without departing from the spirit and principles of the present invention should all be included within the scope of the present invention.

Claims

1. A device status diagnosis method based on a multi-channel fusion neural network, characterized in that: include: Step S1: collecting sensor data arranged on the equipment and preprocessing it to construct multi-channel data of equipment status; Step S2: inputting the multi-channel data of the device state into a multi-component signal analysis model, the multi-component signal analysis model comprising an empirical mode decomposition model and a non-negative matrix decomposition model, wherein the empirical mode decomposition model is used to obtain a device state decomposition signal from the multi-channel data of the device state; inputting the device state decomposition signal into the non-negative matrix decomposition model to separate an independent source device state signal; performing time-frequency domain feature extraction on the independent source device state signal in combination with an improved Hilbert-Huang transform to obtain a time-frequency domain feature result, and reconstructing a device state time-frequency diagram from the time-frequency domain feature result; Step S3: Input the device status time-frequency diagram into a multi-scale sparse time-frequency convolutional network model to obtain a device status diagnosis result.

2. The device status diagnosis method of a multi-channel fusion neural network according to claim 1 is characterized in that: The step S2: inputting the multi-channel data of the device state into a multi-component signal analysis model, the multi-component signal analysis model includes an empirical mode decomposition model and a non-negative matrix decomposition model, wherein the empirical mode decomposition model is used to obtain a device state decomposition signal from the multi-channel data of the device state; inputting the device state decomposition signal into the non-negative matrix decomposition model to separate an independent source device state signal; extracting time-frequency domain features of the independent source device state signal in combination with an improved Hilbert-Huang transform to obtain a time-frequency domain feature result, and reconstructing a device state time-frequency diagram from the time-frequency domain feature result, including: Step S21: The device status multi-channel data , input into the empirical mode decomposition model, decomposed to obtain the trend term and the intrinsic mode functions ,in ; and, according to the following formula (1), and Perform addition reconstruction to obtain the device state decomposition signal : (1) Step S22: Decomposing the device status signal Input into the non-negative matrix factorization model to separate the independent source device status signals ; Step S23: The independent source device status signal The improved Hilbert-Huang transform model is used to extract time-frequency domain features to obtain the time-frequency domain feature results, and the device status time-frequency diagram is reconstructed from the time-frequency domain feature results.

3. The device status diagnosis method of a multi-channel fusion neural network according to claim 2 is characterized in that: The step S22: decomposing the device status signal Input into the non-negative matrix factorization model to separate the independent source device status signals ,include: Step S221: Define the problem and initialize: Decompose the device status signal Decomposed into two non-negative matrices and matrix , so that the matrix , is the original non-negative matrix; randomly initialize the two matrices and matrix , and ensure that the matrix and matrix is a non-negative matrix; Step S222: Iterative update: fix the matrix , by optimizing the matrix To minimize the loss function, and use the gradient descent update rule to update the training weights; fix the matrix , by optimizing the matrix To minimize the loss function, and use the gradient descent update rule to update the training weights; Step S223: Convergence judgment: After each iteration, the loss function value of the current training is calculated; when the change of the loss function is less than a preset threshold, or the maximum number of iterations is reached, whichever condition is reached first, the algorithm converges and meets the convergence condition, and the algorithm is terminated; Step S224: Reconstruct the independent source device status signal: Reconstruct the matrix Each column in is represented as an independent source signal, through the matrix The columns in the matrix are the same as Multiply and reconstruct the independent source device status signal .

4. The device status diagnosis method of a multi-channel fusion neural network according to claim 2 is characterized in that: The step S23: the independent source device status signal The improved Hilbert-Huang transform model is used to extract time-frequency domain features to obtain the time-frequency domain feature results, and the device state time-frequency diagram is reconstructed from the time-frequency domain feature results, including: Step S231: The independent source device status signal Add white noise , get the independent source device status signal with added noise disturbance ; Step S232: Add the independent source device status signal with noise disturbance Perform empirical mode decomposition model decomposition to obtain the intrinsic mode function , and then take the eigenmode function The average value of is obtained according to the following formula (2): The ensemble average decomposition signal : (2) in, It is After adding noise IMF , is the total number of times noise is added; Step S233: According to the following formula (3), for each Calculate permutation entropy at different time scales , select the eigenmode function with the lowest permutation entropy As a valid signal; (3) in, represents the dimension used to construct the vector of the eigenmode function when reconstructing the phase space, is the time scale, is the number of possible orderings of different elements. Is the arrangement The probability of occurrence, indicating that each permutation corresponds to a specific ordering of the subsequences after the phase space reconstruction; Step S234: The intrinsic mode function Using the improved normalized Hilbert transform model, the normalized instantaneous phase is calculated and instantaneous frequency ; in, represents the original signal, Represents the Hilbert transform model obtained at time The analytical signal of Represents the analysis signal The real part of Represents the analysis signal The modulus value, Represents the arccosine function, used to calculate angles; represents the derivative of the normalized instantaneous phase with respect to time; Step S235: The instantaneous frequency Statistical analysis is performed, and the time-frequency domain features are extracted using the average value operation to obtain the time-frequency domain feature extraction results, and the time-frequency feature diagram of the equipment status is reconstructed.

5. The device status diagnosis method of a multi-channel fusion neural network according to claim 1 is characterized in that: The step S3: inputting the device status time-frequency diagram into a multi-scale sparse time-frequency convolutional network model to obtain a device status diagnosis result, including: Step S31: performing a convolution operation on the device state time-frequency feature graph using a multi-scale convolution kernel, and obtaining a device state time-frequency feature result according to the following formula (5); (5) in, represents the convolution output, It is Tier The convolution kernel of channels, is the feature vector of the time-frequency feature map, is the convolution operation weight vector OK, is the convolution operation weight vector OK, is the feature vector of the time-frequency feature map OK, is the feature vector of the time-frequency feature map List, is the activation function, is the bias term; Step S32: using sparse regularization on the device state time-frequency feature graph, and performing a sparse operation on the convolution layer output according to the following formula (6) to obtain a sparse convolution output result; (6) in, represents the regularization coefficient, Indicates Tier The convolution kernel of channels; Step S33: The sparse convolution output results are fused using a feature cascade operation, and the fused feature map result is obtained according to the following formula (7): ; (7) in, Indicates Feature maps of different scales; Step S34: The feature map result Introducing the attention mechanism operation, according to the following formula (8), we get the feature map after attention weighting ; (8) in, Indicates the first , The scoring function of the position feature, It is a feature map No. OK, It is a feature map No. List; Step S35: The attention-weighted feature map Max pooling is used for multi-scale downsampling, and the downsampled feature map is input into the fully connected layer, where the fully connected layer flattens the feature map into a one-dimensional feature vector, outputs the probability of the fault category through the softmax activation function, and performs fault diagnosis of the equipment status.

6. A multi-channel fusion neural network equipment status diagnosis system, characterized in that: Includes the following modules: Equipment failure multi-channel data module: collects sensor data arranged at key parts of the equipment, performs pre-processing, and constructs multi-channel data of equipment status; The device status diagnosis module of the multi-channel fusion neural network: the multi-channel data of the device status is input into a multi-component signal analysis model, the multi-component signal analysis model includes an empirical mode decomposition model and a non-negative matrix decomposition model, wherein the empirical mode decomposition model is used to obtain a device status decomposition signal from the multi-channel data of the device status; the device status decomposition signal is input into the non-negative matrix decomposition model to separate the independent source device status signal; the time-frequency domain feature extraction of the independent source device status signal is performed in combination with the improved Hilbert-Huang transform to obtain the time-frequency domain feature result, and the device status time-frequency diagram is reconstructed from the time-frequency domain feature result; the device status time-frequency diagram is input into a model based on a multi-scale sparse time-frequency convolutional network to obtain the device status diagnosis result.

7. An electronic device, characterized in that: It comprises a memory and a processor, wherein the memory is used to store a computer program executable by the processor, and when the processor executes the computer program, it implements the device status diagnosis method of the multi-channel fusion neural network described in any one of claims 1 to 5.

8. A computer-readable medium, characterized in that The computer-readable medium stores a computer program, and when the computer program is executed, the device status diagnosis method of the multi-channel fusion neural network according to any one of claims 1 to 5 is implemented.

Images