Fire-fighting pressure reducing valve fault diagnosis method based on convolutional neural network

Abstract

The invention provides a fire-fighting pressure reducing valve fault diagnosis method based on a convolutional neural network, and the method comprises the steps: obtaining data, and constructing a basic data set; and performing data reconstruction on the basic data set in a time dimension and a space dimension based on Cmor wavelet transform to obtain a reconstructed time-frequency graph, and determining a training set. And performing feature extraction on the data in the training set based on the residual structure chart neural network. And optimizing the cross entropy loss function based on the weight of the normal state and the abnormal state of the pressure reducing valve to obtain an optimized cross entropy loss function. And continuously training the basic fire-fighting pressure-reducing valve fault diagnosis model by using an optimized cross entropy loss function by using data in the training set, and adjusting parameters of the basic fire-fighting pressure-reducing valve fault diagnosis model based on a result of the optimized cross entropy loss function. And obtaining a fire-fighting pressure-reducing valve fault diagnosis model until a training cut-off condition is reached. According to the method provided by the invention, the diagnosis model with relatively high accuracy can be trained.

Description

Technical Field

[0001] The present application relates to the technical field of fault diagnosis, and in particular to a fire-fighting pressure reducing valve fault diagnosis method based on convolutional neural network. Background Art

[0002] With the continuous development of science and technology, the fire protection field has also made great progress. As a key link in the fire water supply system, the fire pressure reducing valve is responsible for reducing the high-pressure stable city tap water to a pressure suitable for fire water, ensuring the stability of the flow and pressure of fire water. The normal operation of the fire pressure reducing valve is directly related to the effectiveness of the fire protection system. With the acceleration of urbanization, the increase in high-rise buildings and large public facilities, and the continuous development and improvement of the fire protection system, the performance and reliability of the fire pressure reducing valve have also been significantly improved. The existing technology for the diagnosis of fire pressure reducing valve failures is often based on classification models. The fire pressure reducing valve failures are classified by the classification model to obtain the diagnosis results. However, the existing classification model has a low diagnostic accuracy rate for fire pressure reducing valve failures.

[0003] Therefore, how to improve the diagnostic accuracy of fire pressure reducing valve failure has become a technical problem that technical personnel in this field need to solve urgently. Summary of the invention

[0004] Based on the above problems, the present application provides a fire pressure reducing valve fault diagnosis method based on convolutional neural network, so as to improve the diagnosis accuracy of fire pressure reducing valve faults.

[0005] The present application provides a fire pressure reducing valve fault diagnosis method based on convolutional neural network, the method comprising: Acquire a normal data set of a sample pressure reducing valve in a normal state and a fault data set of the sample pressure reducing valve in a fault state, wherein the normal data includes normal vibration data and normal pressure data, and the fault data set includes abnormal vibration data and abnormal pressure data; Building a basic data set based on the normal data set and the fault data set, wherein the data in the basic data set carries a state label, and the state label includes a normal state label and an abnormal state label; Based on Cmor wavelet transform, the basic data set is reconstructed in time dimension and space dimension to obtain a reconstructed time-frequency diagram; Determining a training set based on the reconstructed time-frequency graph; Performing feature extraction on the data in the training set based on a residual structure graph neural network to obtain a first feature extraction result, wherein the residual structure graph neural network is a residual structure graph neural network after feature learning is performed on a first feature extraction result corresponding to an abnormal state label in the first feature extraction result; Obtain a basic fire pressure reducing valve fault diagnosis model, wherein the model type of the basic fire pressure reducing valve fault diagnosis model is a densely connected network model, and the loss function corresponding to the basic fire pressure reducing valve fault diagnosis model is a cross entropy loss function; The cross entropy loss function is optimized based on the weights of the normal state and the abnormal state of the pressure reducing valve to obtain an optimized cross entropy loss function; Continuously training the basic fire pressure reducing valve fault diagnosis model using the optimized cross entropy loss function using the data in the training set, and adjusting the parameters of the basic fire pressure reducing valve fault diagnosis model based on the result of the optimized cross entropy loss function until the training cutoff condition is reached to obtain the fire pressure reducing valve fault diagnosis model; The fault of the pressure reducing valve is diagnosed based on the fire-fighting pressure reducing valve fault diagnosis model.

[0006] In a possible implementation, the reconstructing the basic data set in the time dimension and the space dimension to obtain the reconstructed time-frequency diagram based on the Cmor wavelet transform is performed by the following formula: ; Among them, W(t) represents the time-frequency domain of the time-frequency diagram, t represents the time parameter, b represents the space parameter, exp represents the exponential function, and f represents the frequency.

[0007] In a possible implementation, extracting features from the data in the training set based on the residual structure graph neural network to obtain a first feature extraction result includes: A multi-way parallel network of a residual structure graph neural network is used to perform feature extraction on the data in the training set to obtain a first feature extraction result, wherein the multi-way parallel network includes at least three feature extraction modules, and the convolution kernel of each feature extraction module is one of the three scales of 1×1, 3×3 or 5×5.

[0008] In a possible implementation, the residual structure graph neural network is used to extract features from the data in the training set to obtain a first feature extraction result by using the following formula: ; Among them, T(1) represents the feature extraction result of the first type of abnormal state, T(2) represents the feature extraction result of the second type of abnormal state, and T(3) represents the feature extraction result of the third type of abnormal state. represents the activation function, D represents the degree matrix, X(1) represents the weight of the first type of abnormal state, X(2) represents the weight of the second type of abnormal state, X(3) represents the weight of the third type of abnormal state, S and J represent the input array and the unit matrix respectively, and (S+J) represents the adjacent matrix with self-loop.

[0009] In a possible implementation, the cross entropy loss function is optimized based on the weights of the normal state and the abnormal state of the pressure reducing valve to obtain the optimized cross entropy loss function by the following formula: ; ; in, represents the penalty coefficient, Indicates the number of samples of different categories, represents the category with the largest number of N samples to be classified, Q represents the total number of samples, l(·) represents the judgment function, represents the actual output of the basic fire pressure reducing valve fault diagnosis model, and n represents the nth category; The optimized cross entropy loss function is expressed by the following formula: ; Among them, L represents the optimized cross entropy loss function, Q represents the total number of samples, and N represents N samples to be classified. represents the number of samples of different categories, l(·) represents the judgment function, represents the actual output of the basic fire pressure reducing valve fault diagnosis model, n represents the nth category, represents the weight matrix of the i-th category, represents the weight matrix of the nth category, represents the bias of the i-th category, represents the bias of the nth category, and exp(·) represents the exponential function.

[0010] In a possible implementation, diagnosing a pressure reducing valve fault based on the fire pressure reducing valve fault diagnosis model includes: Inputting the to-be-diagnosed data of the fire-fighting pressure-reducing valve to be diagnosed into the fire-fighting pressure-reducing valve fault diagnosis model to obtain the diagnosis data; Based on the diagnostic data, it is determined whether the fire-fighting pressure reducing valve to be diagnosed is faulty, and if a fault occurs, the fault type is diagnosed.

[0011] The present application also provides a fire pressure reducing valve fault diagnosis device based on a convolutional neural network, the device comprising: A first acquisition module, used to acquire a normal data set of a sample pressure reducing valve in a normal state and a fault data set of the sample pressure reducing valve in a fault state, wherein the normal data includes normal vibration data and normal pressure data, and the fault data set includes abnormal vibration data and abnormal pressure data; A construction module, configured to construct a basic data set based on the normal data set and the fault data set, wherein the data in the basic data set carries a state label, and the state label includes a normal state label and an abnormal state label; A reconstruction module, used for reconstructing the basic data set in the time dimension and the space dimension respectively based on Cmor wavelet transform to obtain a reconstructed time-frequency diagram; A determination module, used for determining a training set based on the reconstructed time-frequency graph; A feature extraction module, used for performing feature extraction on the data in the training set based on a residual structure graph neural network to obtain a first feature extraction result, wherein the residual structure graph neural network is a residual structure graph neural network after feature learning is performed on a first feature extraction result corresponding to an abnormal state label in the first feature extraction result; A second acquisition module is used to acquire a basic fire pressure reducing valve fault diagnosis model, wherein the model type of the basic fire pressure reducing valve fault diagnosis model is a densely connected network model, and the loss function corresponding to the basic fire pressure reducing valve fault diagnosis model is a cross entropy loss function; An optimization module, used for optimizing the cross entropy loss function based on the weights of the normal state and the abnormal state of the pressure reducing valve to obtain an optimized cross entropy loss function; A training module, used to continuously train the basic fire pressure reducing valve fault diagnosis model using the optimized cross entropy loss function using the data in the training set, and adjust the parameters of the basic fire pressure reducing valve fault diagnosis model based on the result of the optimized cross entropy loss function until the training cutoff condition is met to obtain the fire pressure reducing valve fault diagnosis model; The diagnosis module is used to diagnose the fault of the pressure reducing valve based on the fire pressure reducing valve fault diagnosis model.

[0012] In a possible implementation, the reconstruction module is specifically used to reconstruct using the following formula: ; Among them, W(t) represents the time-frequency domain of the time-frequency diagram, t represents the time parameter, b represents the space parameter, exp represents the exponential function, and f represents the frequency.

[0013] The present application also provides an electronic device, the electronic device comprising a processor and a memory: The memory is used to store a computer program and transmit the computer program to the processor; The processor is used to execute the steps of the above-mentioned fire pressure reducing valve fault diagnosis method based on convolutional neural network according to the instructions in the computer program.

[0014] The present application also provides a computer-readable storage medium, which is used to store a computer program. When the computer program is executed by an electronic device, it implements the steps of the above-mentioned fire pressure reducing valve fault diagnosis method based on convolutional neural network.

[0015] Compared with the prior art, this application has the following beneficial effects: The fire pressure reducing valve fault diagnosis method based on convolutional neural network provided by itself reconstructs the data in time and frequency by combining Cmor wavelet transform. In actual working conditions, most of the fire pressure reducing valves are in normal state, resulting in a scarcity of fault data. Traditional models are often difficult to accurately identify the fault type when the data is unbalanced. This application can expand the data used for training through time-frequency reconstruction, and use convolutional neural network to extract more refined features, which can better capture the differences in the time and space dimensions of fault data, thereby effectively improving the accuracy of fault diagnosis. This application adopts a residual structure graph neural network, which can perform deep learning in the feature extraction process, especially feature learning for abnormal states, effectively avoiding the problem of insufficient abnormal data processing in traditional methods. Through the optimization of the residual network, the model can adaptively adjust the network structure, extract more effective feature information, and improve the robustness and diagnostic ability of the model. This application combines Cmor wavelet transform with convolutional neural network for the first time to perform time-frequency reconstruction and feature extraction, so that the model can simultaneously utilize information in time and space dimensions. Through the combination of time-frequency analysis and deep learning, the operating status of the pressure reducing valve can be analyzed more comprehensively, potential fault characteristics can be discovered, and the shortcomings of traditional methods in multi-dimensional data processing can be made up. This application proposes a training method based on optimizing the cross-entropy loss function. By optimizing the weights of normal and abnormal states, the model can better cope with complex fault diagnosis scenarios. During the training process, the model parameters are adjusted based on the continuously optimized loss function to ensure accurate identification of different fault states during the model training process. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0017] Figure 1 A flow chart of a fire pressure reducing valve fault diagnosis method based on a convolutional neural network provided in an embodiment of the present application; Figure 2 A schematic diagram of the structure of a fire pressure reducing valve fault diagnosis device based on a convolutional neural network provided in an embodiment of the present application. DETAILED DESCRIPTION

[0018] As described above, the existing technology for diagnosing fire pressure reducing valve faults is often based on a classification model, which is used to classify fire pressure reducing valve faults and obtain diagnosis results. However, the existing classification model has a low accuracy rate in diagnosing fire pressure reducing valve faults.

[0019] After research, it was found that with the development of science and technology, the quality of fire pressure reducing valves is getting better and better, and naturally the probability of fire pressure reducing valve failure is getting lower and lower. The low failure rate of fire pressure reducing valves leads to a lack of enough fire pressure reducing valve failure samples in the training model, which ultimately leads to a low diagnostic accuracy of the trained model. If you want to improve the diagnostic accuracy of the trained model, the existing method is often to conduct large-scale data search, but even if you search for fault data of fire pressure reducing valves in a very large range, the number of fault samples is far from enough to make the diagnostic accuracy of the trained model reach a good expectation. This is an inevitable result of the increasingly better quality of fire pressure reducing valves. Therefore, how to train a model with higher accuracy through less fire pressure reducing valve failure data has become a question worth thinking about.

[0020] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0021] It can be understood that the method provided in the present application can be applied to a processing device, which is a processing device that can obtain a normal data set of a sample pressure reducing valve in a normal state and a fault data set of the sample pressure reducing valve in a faulty state, for example, a terminal device or server that can reconstruct the basic data set in the time dimension and the space dimension based on the Cmor wavelet transform to obtain a reconstructed time-frequency diagram. The method provided in the present application can be executed independently by a terminal device or a server, and can also be applied to a network scenario in which a terminal device and a server communicate, and is executed by the cooperation of a terminal device and a server. Among them, the terminal device can be a computer, a mobile phone and other devices. The server can be understood as an application server or a Web server. In actual deployment, the server can be an independent server or a cluster server.

[0022] Figure 1 A flow chart of a fire pressure reducing valve fault diagnosis method based on a convolutional neural network provided in this application, the method comprising the following steps: S101: Acquire a normal data set of a sample pressure reducing valve in a normal state and a fault data set of the sample pressure reducing valve in a fault state.

[0023] The processing device obtains a normal data set of the sample pressure reducing valve in a normal state and a fault data set of the sample pressure reducing valve in a fault state, wherein the normal data includes normal vibration data and normal pressure data, and the fault data set includes abnormal vibration data and abnormal pressure data; In actual applications, fire pressure reducing valves may fail due to various reasons, such as frequent activation of the pressure stabilizing device, no signal from the water flow indicator, etc. These failures will not only affect the normal operation of the fire protection system, but may also pose a threat to people's lives and property safety.

[0024] Pressure reducing valves may have various faults such as membrane damage, valve wall damage, spring fatigue, and loose bolts, which are reflected in the outlet pressure as three faults: surge, low outlet pressure, and high lock pressure. In a possible implementation, the processing equipment can simulate the above working conditions through the pressure reducing valve performance test data acquisition system, collect vibration data and pressure data of the fire-fighting pressure reducing valve in normal and faulty states, and determine the normal data set and fault data set based on this.

[0025] S102: Construct a basic data set based on the normal data set and the fault data set.

[0026] The processing device constructs a basic data set based on the normal data set and the fault data set. The data in the basic data set carries a state label, and the state label includes a normal state label and an abnormal state label.

[0027] S103: reconstructing the basic data set in the time dimension and the space dimension based on Cmor wavelet transform to obtain a reconstructed time-frequency diagram.

[0028] The processing device reconstructs the basic data set in the time dimension and the space dimension based on the Cmor wavelet transform to obtain a reconstructed time-frequency diagram.

[0029] In a possible implementation, the basic data set is reconstructed in the time dimension and the space dimension based on the Cmor wavelet transform to obtain the reconstructed time-frequency diagram, which can be performed by the following formula: ; Among them, W(t) represents the time-frequency domain of the time-frequency diagram, t represents the time parameter, b represents the space parameter, exp represents the exponential function, and f represents the frequency.

[0030] In order to improve the accuracy of data in feature extraction and fault classification, the processing equipment can use Cmor wavelet transform to reconstruct and disassemble the data in the basic data set.

[0031] S104: Determine a training set based on the reconstructed time-frequency graph.

[0032] The processing device may determine a training set based on the reconstructed frequency spectrum diagram.

[0033] In a possible implementation, the processing device can divide the reconstructed time-frequency diagram into data according to 6:4 to obtain a training set and a test set. It should be noted that the ratio of 6:4 is only one possible implementation, and of course it can also be 5:5 or 7:3, or any other ratio.

[0034] S105: Perform feature extraction on the data in the training set based on the residual structure graph neural network to obtain a first feature extraction result.

[0035] The processing device performs feature extraction on the data in the training set based on the residual structure graph neural network to obtain a first feature extraction result. The residual structure graph neural network is a residual structure graph neural network after feature learning is performed on the first feature extraction result corresponding to the abnormal state label in the first feature extraction result.

[0036] In a possible implementation, the processing device can use a multi-way parallel network of a residual structure graph neural network to extract features from the data in the training set to obtain a first feature extraction result, wherein the multi-way parallel network includes at least three feature extraction modules, and the convolution kernel of each feature extraction module is one of the three scales of 1×1, 3×3 or 5×5. Small convolution kernels can optimize the number of parameters and the amount of calculation. At the same time, it can improve the generalization of the model and accelerate convergence.

[0037] In a possible implementation, the feature extraction of the data in the training set based on the residual structure graph neural network to obtain the first feature extraction result can be performed by the following formula: ; Among them, T(1) represents the feature extraction result of the first type of abnormal state, T(2) represents the feature extraction result of the second type of abnormal state, and T(3) represents the feature extraction result of the third type of abnormal state. represents the activation function, D represents the degree matrix, X(1) represents the weight of the first type of abnormal state, X(2) represents the weight of the second type of abnormal state, X(3) represents the weight of the third type of abnormal state, S and J represent the input array and the unit matrix respectively, and (S+J) represents the adjacent matrix with self-loop.

[0038] S106: Obtain a basic fire-fighting pressure reducing valve fault diagnosis model.

[0039] The processing device obtains a basic fire pressure reducing valve fault diagnosis model. The model type of the basic fire pressure reducing valve fault diagnosis model is a densely connected network model. The loss function corresponding to the basic fire pressure reducing valve fault diagnosis model is a cross entropy loss function.

[0040] In a possible implementation, the basic fire-fighting pressure reducing valve fault diagnosis model can use the pressure reducing valve outlet pressure in different fault states as input. A single sample contains 2048 data points, which are converted into a two-dimensional image with a size of 64×64.

[0041] Secondly, the basic fire pressure reducing valve fault diagnosis model is built with three transition layers cascaded with three Dense Blocks. The cascaded network structure enables the network to support feature reuse and improve parameter utilization while ensuring the uniformity of feature map dimensions.

[0042] Then, the output of the third transition layer is flattened and input into two fully connected layers; finally, after feature extraction, an N-dimensional array is output through the Softmax layer. Each dimension in the array corresponds to a different fault category, that is, an N-classification problem, which can be expressed as: ; Among them, Out is the final output of the network, P(·) is the matching probability between the output result and the corresponding category, n is the different fault categories, Wn and bn are the weight matrix and bias of each category respectively, and exp(·) is the exponential function.

[0043] S107: Optimizing the cross entropy loss function based on the weights of the normal state and the abnormal state of the pressure reducing valve to obtain an optimized cross entropy loss function.

[0044] The processing device optimizes the cross entropy loss function based on the weights of the normal state and the abnormal state of the pressure reducing valve to obtain an optimized cross entropy loss function.

[0045] In order to solve the problem of difficulty in classifying fault categories of unbalanced samples, this application is based on the weighted cross entropy loss function. On the basis of the original loss function, corresponding penalty coefficients are added to the error losses of different categories, thereby achieving the weighted average of error losses of different categories.

[0046] The cross entropy loss function is optimized based on the weights of the normal state and the abnormal state of the pressure reducing valve to obtain the optimized cross entropy loss function which can be obtained by the following formula: ; ; in, represents the penalty coefficient, Indicates the number of samples of different categories, represents the category with the largest number of N samples to be classified, Q represents the total number of samples, l(·) represents the judgment function, represents the actual output of the basic fire pressure reducing valve fault diagnosis model, and n represents the nth category.

[0047] The optimized cross entropy loss function can be expressed by the following formula: ; Among them, L represents the optimized cross entropy loss function, Q represents the total number of samples, and N represents N samples to be classified. represents the number of samples of different categories, l(·) represents the judgment function, represents the actual output of the basic fire pressure reducing valve fault diagnosis model, n represents the nth category, represents the weight matrix of the i-th category, represents the weight matrix of the nth category, represents the bias of the i-th category, represents the bias of the nth category, and exp(·) represents the exponential function.

[0048] S108: Continuously train the basic fire pressure reducing valve fault diagnosis model using the optimized cross entropy loss function using the data in the training set, and adjust the parameters of the basic fire pressure reducing valve fault diagnosis model based on the result of the optimized cross entropy loss function until the training cutoff condition is met to obtain the fire pressure reducing valve fault diagnosis model.

[0049] The processing device continuously trains the basic fire pressure reducing valve fault diagnosis model using the optimized cross entropy loss function using the data in the training set, and adjusts the parameters of the basic fire pressure reducing valve fault diagnosis model based on the result of the optimized cross entropy loss function until the training cutoff condition is met to obtain the fire pressure reducing valve fault diagnosis model.

[0050] S109: Diagnose the fault of the pressure reducing valve based on the fire pressure reducing valve fault diagnosis model.

[0051] The processing equipment diagnoses the fault of the pressure reducing valve based on the fault diagnosis model of the fire pressure reducing valve.

[0052] In a possible implementation, the processing device can input the diagnostic data of the fire pressure reducing valve to be diagnosed into the fire pressure reducing valve fault diagnosis model to obtain diagnostic data. Then, based on the diagnostic data, it is determined whether the fire pressure reducing valve to be diagnosed is faulty, and if so, the fault type can be further diagnosed.

[0053] The fire pressure reducing valve fault diagnosis method based on convolutional neural network provided by itself reconstructs the data in time and frequency by combining Cmor wavelet transform. In actual working conditions, most of the fire pressure reducing valves are in normal state, resulting in a scarcity of fault data. Traditional models are often difficult to accurately identify the fault type when the data is unbalanced. This application can expand the data used for training through time-frequency reconstruction, and use convolutional neural network to extract more refined features, which can better capture the differences in the time and space dimensions of fault data, thereby effectively improving the accuracy of fault diagnosis. This application adopts a residual structure graph neural network, which can perform deep learning in the feature extraction process, especially feature learning for abnormal states, effectively avoiding the problem of insufficient abnormal data processing in traditional methods. Through the optimization of the residual network, the model can adaptively adjust the network structure, extract more effective feature information, and improve the robustness and diagnostic ability of the model. This application combines Cmor wavelet transform with convolutional neural network for the first time to perform time-frequency reconstruction and feature extraction, so that the model can simultaneously utilize information in time and space dimensions. Through the combination of time-frequency analysis and deep learning, the operating status of the pressure reducing valve can be analyzed more comprehensively, potential fault characteristics can be discovered, and the shortcomings of traditional methods in multi-dimensional data processing can be made up. This application proposes a training method based on optimizing the cross-entropy loss function. By optimizing the weights of normal and abnormal states, the model can better cope with complex fault diagnosis scenarios. During the training process, the model parameters are adjusted based on the continuously optimized loss function to ensure accurate identification of different fault states during the model training process.

[0054] The present application also provides a Figure 2 The structural diagram of the fire pressure reducing valve fault diagnosis device based on the convolutional neural network is shown, and the fire pressure reducing valve fault diagnosis device based on the convolutional neural network 200 includes: A first acquisition module 201 is used to acquire a normal data set of a sample pressure reducing valve in a normal state and a fault data set of the sample pressure reducing valve in a fault state, wherein the normal data includes normal vibration data and normal pressure data, and the fault data set includes abnormal vibration data and abnormal pressure data; A construction module 202 is used to construct a basic data set based on the normal data set and the fault data set, wherein the data in the basic data set carries a state label, and the state label includes a normal state label and an abnormal state label; A reconstruction module 203 is used to reconstruct the basic data set in the time dimension and the space dimension based on Cmor wavelet transform to obtain a reconstructed time-frequency diagram; A determination module 204, configured to determine a training set based on the reconstructed time-frequency graph; A feature extraction module 205 is used to extract features from the data in the training set based on a residual structure graph neural network to obtain a first feature extraction result, wherein the residual structure graph neural network is a residual structure graph neural network obtained by performing feature learning on a first feature extraction result corresponding to an abnormal state label in the first feature extraction result; The second acquisition module 206 is used to acquire a basic fire pressure reducing valve fault diagnosis model, wherein the model type of the basic fire pressure reducing valve fault diagnosis model is a densely connected network model, and the loss function corresponding to the basic fire pressure reducing valve fault diagnosis model is a cross entropy loss function; An optimization module 207, configured to optimize the cross entropy loss function based on the weights of the normal state and the abnormal state of the pressure reducing valve to obtain an optimized cross entropy loss function; A training module 208 is used to continuously train the basic fire pressure reducing valve fault diagnosis model using the optimized cross entropy loss function using the data in the training set, and adjust the parameters of the basic fire pressure reducing valve fault diagnosis model based on the result of the optimized cross entropy loss function until the training cutoff condition is met to obtain the fire pressure reducing valve fault diagnosis model; The diagnosis module 209 is used to diagnose the fault of the pressure reducing valve based on the fire pressure reducing valve fault diagnosis model.

[0055] The fire pressure reducing valve fault diagnosis device based on convolutional neural network provided by itself reconstructs the data in time and frequency by combining Cmor wavelet transform. In actual working conditions, most of the fire pressure reducing valves are in normal state, resulting in a scarcity of fault data. Traditional models are often difficult to accurately identify the fault type when the data is unbalanced. This application can expand the data used for training through time-frequency reconstruction, and use convolutional neural network to extract more refined features, which can better capture the differences in the time and space dimensions of fault data, thereby effectively improving the accuracy of fault diagnosis. This application adopts a residual structure graph neural network, which can perform deep learning in the feature extraction process, especially for feature learning of abnormal states, effectively avoiding the problem of insufficient abnormal data processing in traditional methods. Through the optimization of the residual network, the model can adaptively adjust the network structure, extract more effective feature information, and improve the robustness and diagnostic ability of the model. This application combines Cmor wavelet transform with convolutional neural network for the first time to perform time-frequency reconstruction and feature extraction, so that the model can simultaneously utilize information in time and space dimensions. Through the combination of time-frequency analysis and deep learning, the operating status of the pressure reducing valve can be analyzed more comprehensively, potential fault characteristics can be discovered, and the shortcomings of traditional methods in multi-dimensional data processing can be made up. This application proposes a training method based on optimizing the cross-entropy loss function. By optimizing the weights of normal and abnormal states, the model can better cope with complex fault diagnosis scenarios. During the training process, the model parameters are adjusted based on the continuously optimized loss function to ensure accurate identification of different fault states during the model training process.

[0056] An embodiment of the present application also provides a fire pressure reducing valve fault diagnosis device based on a convolutional neural network, wherein the device includes a memory and a processor, the memory is used to store instructions or codes, and the processor is used to execute the instructions or codes so that the device performs the steps of the fire pressure reducing valve fault diagnosis method based on a convolutional neural network described in any embodiment of the present application.

[0057] In practical applications, the computer-readable storage medium may be any combination of one or more computer-readable media, which may be a computer-readable signal medium or a computer-readable storage medium.

[0058] The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, device, or device.

[0059] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0060] The program code embodied on the computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0061] Computer program code for performing the operations of the present invention may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0062] It should be noted that each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. The device embodiment described above is merely schematic, in which the unit described as a separate component may or may not be physically separated, and the component prompted as a unit may or may not be a physical unit, that is, it may be located in one place, or it may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative work.

[0063] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A fire pressure reducing valve fault diagnosis method based on convolutional neural network, characterized in that: include: Acquire a normal data set of a sample pressure reducing valve in a normal state and a fault data set of the sample pressure reducing valve in a fault state, wherein the normal data includes normal vibration data and normal pressure data, and the fault data set includes abnormal vibration data and abnormal pressure data; Building a basic data set based on the normal data set and the fault data set, wherein the data in the basic data set carries a state label, and the state label includes a normal state label and an abnormal state label; Based on Cmor wavelet transform, the basic data set is reconstructed in time dimension and space dimension to obtain a reconstructed time-frequency diagram; Determining a training set based on the reconstructed time-frequency graph; Performing feature extraction on the data in the training set based on a residual structure graph neural network to obtain a first feature extraction result, wherein the residual structure graph neural network is a residual structure graph neural network after feature learning is performed on a first feature extraction result corresponding to an abnormal state label in the first feature extraction result; Obtain a basic fire pressure reducing valve fault diagnosis model, wherein the model type of the basic fire pressure reducing valve fault diagnosis model is a densely connected network model, and the loss function corresponding to the basic fire pressure reducing valve fault diagnosis model is a cross entropy loss function; The cross entropy loss function is optimized based on the weights of the normal state and the abnormal state of the pressure reducing valve to obtain an optimized cross entropy loss function; Continuously training the basic fire pressure reducing valve fault diagnosis model using the optimized cross entropy loss function using the data in the training set, and adjusting the parameters of the basic fire pressure reducing valve fault diagnosis model based on the result of the optimized cross entropy loss function until the training cutoff condition is reached to obtain the fire pressure reducing valve fault diagnosis model; The fault of the pressure reducing valve is diagnosed based on the fire-fighting pressure reducing valve fault diagnosis model.

2. The method according to claim 1, characterized in that The reconstructing the basic data set in the time dimension and the space dimension to obtain the reconstructed time-frequency diagram based on the Cmor wavelet transform is performed by the following formula: ； Among them, W(t) represents the time-frequency domain of the time-frequency diagram, t represents the time parameter, b represents the space parameter, exp represents the exponential function, and f represents the frequency.

3. The method according to claim 1, characterized in that: The extracting features of the data in the training set based on the residual structure graph neural network to obtain the first feature extraction result includes: A multi-way parallel network of a residual structure graph neural network is used to perform feature extraction on the data in the training set to obtain a first feature extraction result, wherein the multi-way parallel network includes at least three feature extraction modules, and the convolution kernel of each feature extraction module is one of the three scales of 1×1, 3×3 or 5×5.

4. The method according to claim 1, characterized in that: The feature extraction of the data in the training set based on the residual structure graph neural network to obtain the first feature extraction result is performed by the following formula: ； Among them, T(1) represents the feature extraction result of the first type of abnormal state, T(2) represents the feature extraction result of the second type of abnormal state, and T(3) represents the feature extraction result of the third type of abnormal state. represents the activation function, D represents the degree matrix, X(1) represents the weight of the first type of abnormal state, X(2) represents the weight of the second type of abnormal state, X(3) represents the weight of the third type of abnormal state, S and J represent the input array and the unit matrix respectively, and (S+J) represents the adjacent matrix with self-loop.

5. The method according to claim 1, characterized in that The cross entropy loss function is optimized based on the weights of the normal state and the abnormal state of the pressure reducing valve to obtain the optimized cross entropy loss function through the following formula: ； ； in, represents the penalty coefficient, Indicates the number of samples of different categories, represents the category with the largest number of N samples to be classified, Q represents the total number of samples, l(·) represents the judgment function, represents the actual output of the basic fire pressure reducing valve fault diagnosis model, and n represents the nth category; The optimized cross entropy loss function can be expressed by the following formula: ； Among them, L represents the optimized cross entropy loss function, Q represents the total number of samples, and N represents N samples to be classified. represents the number of samples of different categories, l(·) represents the judgment function, represents the actual output of the basic fire pressure reducing valve fault diagnosis model, n represents the nth category, represents the weight matrix of the i-th category, represents the weight matrix of the nth category, represents the bias of the i-th category, represents the bias of the nth category, and exp(·) represents the exponential function.

6. The method according to claim 1, characterized in that The diagnosing the fault of the pressure reducing valve based on the fire pressure reducing valve fault diagnosis model includes: Inputting the to-be-diagnosed data of the fire-fighting pressure-reducing valve to be diagnosed into the fire-fighting pressure-reducing valve fault diagnosis model to obtain the diagnosis data; Based on the diagnostic data, it is determined whether the fire-fighting pressure reducing valve to be diagnosed is faulty, and if a fault occurs, the fault type is diagnosed.

7. A fire pressure reducing valve fault diagnosis device based on convolutional neural network, characterized in that: include: A first acquisition module, used to acquire a normal data set of a sample pressure reducing valve in a normal state and a fault data set of the sample pressure reducing valve in a fault state, wherein the normal data includes normal vibration data and normal pressure data, and the fault data set includes abnormal vibration data and abnormal pressure data; A construction module, configured to construct a basic data set based on the normal data set and the fault data set, wherein the data in the basic data set carries a state label, and the state label includes a normal state label and an abnormal state label; A reconstruction module, used for reconstructing the basic data set in the time dimension and the space dimension respectively based on Cmor wavelet transform to obtain a reconstructed time-frequency diagram; A determination module, used for determining a training set based on the reconstructed time-frequency graph; A feature extraction module, used for performing feature extraction on the data in the training set based on a residual structure graph neural network to obtain a first feature extraction result, wherein the residual structure graph neural network is a residual structure graph neural network after feature learning is performed on a first feature extraction result corresponding to an abnormal state label in the first feature extraction result; A second acquisition module is used to acquire a basic fire pressure reducing valve fault diagnosis model, wherein the model type of the basic fire pressure reducing valve fault diagnosis model is a densely connected network model, and the loss function corresponding to the basic fire pressure reducing valve fault diagnosis model is a cross entropy loss function; An optimization module, used for optimizing the cross entropy loss function based on the weights of the normal state and the abnormal state of the pressure reducing valve to obtain an optimized cross entropy loss function; A training module, used to continuously train the basic fire pressure reducing valve fault diagnosis model using the optimized cross entropy loss function using the data in the training set, and adjust the parameters of the basic fire pressure reducing valve fault diagnosis model based on the result of the optimized cross entropy loss function until the training cutoff condition is met to obtain the fire pressure reducing valve fault diagnosis model; The diagnosis module is used to diagnose the fault of the pressure reducing valve based on the fire pressure reducing valve fault diagnosis model.

8. The device according to claim 7, characterized in that The reconstruction module is specifically used to reconstruct through the following formula: ； Among them, W(t) represents the time-frequency domain of the time-frequency diagram, t represents the time parameter, b represents the space parameter, exp represents the exponential function, and f represents the frequency.

9. An electronic device, characterized in that: comprising a memory and a processor, wherein: The memory is used to store the computer program; The processor is used to execute the computer program to implement the fire pressure reducing valve fault diagnosis method based on convolutional neural network as described in any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that: Used to store a computer program, wherein when the computer program is executed by a processor, the fire pressure reducing valve fault diagnosis method based on a convolutional neural network as described in any one of claims 1 to 6 is implemented.

Images