A method and system for diagnosing faults of an air compression system based on a CNN neural network

Abstract

This invention discloses a method and system for fault diagnosis of air compression systems based on a CNN neural network. The method includes: acquiring normal variable data and fault variable data associated with the target system; preprocessing the normal variable data and fault variable data; inputting the preprocessed normal variable data and fault variable data into a pre-trained CNN neural network fault diagnosis model for diagnosis, thereby obtaining the fault diagnosis result of the air compression system. The CNN neural network fault diagnosis model is trained using SDP image data corresponding to the normal variable data and fault variable data of the air compression system as input and the corresponding fault diagnosis result as output. This invention more accurately reflects fault information and locates faults, deeply explores the hidden features of fault data in time and space, thereby improving the accuracy of fault diagnosis under different working environments.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis technology, and more specifically, to a fault diagnosis method and system for air compression systems based on CNN neural networks. Background Technology

[0002] An air compression system refers to the compression of air at atmospheric pressure and its delivery to a pneumatic system at a higher pressure. A typical air compression system consists of the following components: an air compressor (which may have a cooler), an air tank, a dryer (refrigerated or adsorption type), filters (including oil-water separators, oil removal filters, deodorizing filters, sterilizing filters, etc.), a pressure-stabilizing air tank, an automatic drain and sewage discharge device, air delivery pipelines, pipeline valves, control instruments, pneumatic tools, pneumatic power machinery, and the process flow using compressed air. Currently, to ensure the stable operation of equipment, it is frequently necessary to check whether any components in the equipment have malfunctioned. With the development of modern industry, air compression systems have become an indispensable fourth energy system. Therefore, air compression systems are widely used. At the same time, this system is susceptible to malfunctions due to external environmental influences. Therefore, timely and accurate fault detection of air compression systems is a very necessary task.

[0003] The information disclosed in the background section is only intended to enhance the understanding of the background of this application, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0004] This invention provides a method and system for fault diagnosis of air compression systems based on CNN neural networks, which solves the technical problem of low accuracy of fault detection results in air compression systems.

[0005] In a first aspect, one technical solution provided in this embodiment of the invention is a fault diagnosis method for an air compression system based on a CNN neural network, comprising:

[0006] Acquire normal and fault variable data associated with the target system;

[0007] Preprocess normal variable data and fault variable data;

[0008] The preprocessed normal variable data and fault variable data are input into a pre-trained CNN neural network fault diagnosis model for diagnosis, and the fault diagnosis results of the air compression system are obtained.

[0009] The CNN neural network fault diagnosis model is trained by taking the SDP image data corresponding to the normal variable data and fault variable data of the air compression system as input and the fault diagnosis results corresponding to the normal variable data and fault variable data of the air compression system as output.

[0010] The CNN neural network fault diagnosis model includes an input layer, a backbone network, a hybrid attention module, a feature fusion module, and an output layer. The backbone network extracts features from the SDP image to obtain low-semantic feature maps and high-semantic feature maps. The hybrid attention module compresses the high-semantic feature map to obtain an enhanced feature map that fuses the channel attention feature map and the spatial attention feature map. The feature fusion module performs feature fusion operations on the low-semantic feature map and the enhanced feature map to obtain a fused feature map.

[0011] Secondly, an embodiment of the present invention also provides a technical solution: a fault diagnosis system for an air compression system based on a CNN neural network, comprising:

[0012] The data acquisition module obtains normal variable data and fault variable data associated with the target system;

[0013] The preprocessing module preprocesses normal variable data and fault variable data;

[0014] The diagnostic module inputs the pre-processed normal variable data and fault variable data into a pre-trained CNN neural network fault diagnosis model for diagnosis, and obtains the fault diagnosis results of the air compression system.

[0015] The CNN neural network fault diagnosis model is trained by taking the SDP image data corresponding to the normal variable data and fault variable data of the air compression system as input and the fault diagnosis results corresponding to the normal variable data and fault variable data of the air compression system as output.

[0016] The CNN neural network fault diagnosis model includes an input layer, a backbone network, a hybrid attention module, a feature fusion module, and an output layer. The backbone network extracts features from the SDP image to obtain low-semantic feature maps and high-semantic feature maps. The hybrid attention module compresses the high-semantic feature map to obtain an enhanced feature map that fuses the channel attention feature map and the spatial attention feature map. The feature fusion module performs feature fusion operations on the low-semantic feature map and the enhanced feature map to obtain a fused feature map.

[0017] Thirdly, one technical solution provided in this embodiment of the invention is a fault diagnosis manager for an air compression system based on a CNN neural network. The fault diagnosis manager includes a processor and a memory. The memory stores at least one instruction, which is loaded and executed by the processor to perform the operations performed by the above-described fault diagnosis method for an air compression system based on a CNN neural network.

[0018] Fourthly, one technical solution provided in this embodiment of the invention is a computer-readable storage medium, wherein the storage medium stores at least one instruction, which is loaded and executed by a processor to perform the operations performed by the above-described air compression system fault diagnosis method based on CNN neural network.

[0019] The beneficial effects of this invention are:

[0020] Various variables of the air compression system (including normal and fault variable data) are used as raw data. After SDP transformation, they are used as input to a pre-trained CNN neural network fault diagnosis model. The output of the pre-trained CNN neural network fault diagnosis model is the fault type related to the variable, such as abnormal pressure leading to excessive pressure. The robustness is improved by using the pre-trained CNN neural network fault diagnosis model. Because the CNN neural network fault diagnosis model has the ability to collect multi-source data and extract latent features from a large amount of data, and can process multiple variable data simultaneously, it can more accurately reflect fault information and locate faults more precisely, thereby improving the accuracy of fault diagnosis in different working environments.

[0021] The above description of the invention is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0022] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings. The drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings.

[0023] Figure 1 A flowchart of a fault diagnosis method for an air compression system based on a CNN neural network provided in Embodiment 1 of the present invention;

[0024] Figure 2 This is a schematic diagram of SDP conversion for the air compression system fault diagnosis method based on CNN neural network provided in Embodiment 1 of the present invention;

[0025] Figure 3 The diagram shows the CNN model structure of the air compression system fault diagnosis method based on CNN neural network provided in Embodiment 1 of the present invention.

[0026] Figure 4 The experimental results on the motor bearing dataset of the air compression system fault diagnosis method based on CNN neural network provided in Embodiment 1 of the present invention are shown.

[0027] Figure 5 This is a schematic diagram of the structure of the air compression system fault diagnosis system based on CNN neural network provided in Embodiment 2 of the present invention. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only one preferred embodiment of this invention and are only used to explain this invention. They do not limit the scope of protection of this invention. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0029] Before discussing the exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe operations (or steps) as being processed sequentially, many of these operations (or steps) may be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations may be rearranged. A process may terminate when its operation is completed, but it may also have additional steps not included in the figures; a process may correspond to a method, function, procedure, subroutine, subprogram, etc.

[0030] The following is a brief description of the technical field involved in the solution disclosed herein:

[0031] Convolutional Neural Networks (CNNs) are deep feedforward neural networks characterized by local connectivity and weight sharing. They are one of the representative algorithms of deep learning, excelling at handling image processing, particularly image recognition and related machine learning problems. They show significant improvements in various visual tasks such as image classification, object detection, and image segmentation, and are among the most widely used models today. The basic structure of a CNN generally includes: convolutional layers, activation functions, pooling layers, fully connected layers, and an output layer.

[0032] Example 1

[0033] Figure 1The flowchart illustrates a fault diagnosis method for an air compression system based on a CNN neural network, according to an embodiment of the present invention. The method includes:

[0034] Step S101: Obtain normal variable data and fault variable data associated with the target system;

[0035] It should be noted that the difference between normal variable data and fault variable data lies in the different collection conditions, although the types of data collected are the same.

[0036] Variable data includes: compressed air generation variables and compressed air process variables; compressed air generation variables include: compressed air flow rate, pressure, and temperature; compressed air process variables include: standard volumetric flow rate, cooling temperature, drying temperature, and pressurized tank pressure.

[0037] Among them, the normal variable data are the variable data of the industrial process under normal operating conditions, including: variables such as the flow rate, pressure and temperature of compressed air, as well as variables such as the standard volumetric flow rate, cooling temperature, drying temperature and pressure stabilizing tank pressure during the compression process;

[0038] Among them, the fault variable data are the variable data of the industrial process under fault conditions, including: variables such as compressed air flow rate, pressure and temperature, as well as variables such as standard volume flow rate, cooling temperature, drying temperature and pressure tank pressure during the compression process;

[0039] Step S102: Preprocess the normal variable data and the fault variable data;

[0040] The preprocessing includes normalizing the normal and fault variable data and performing SDP image conversion to obtain SDP image data; including:

[0041] The data is normalized to obtain time-domain data; specifically, the original data is set as R0(t), and this data segment is first normalized to obtain the corresponding time-domain data, as shown in the following formula:

[0042]

[0043] In the above formula (1), R(i) is used to represent the normalized value of the i-th sampling point of the current time domain data;

[0044] R0(i) is used to represent the value of R0(t) at the i-th sampling point;

[0045] t is used to represent time;

[0046] R 0mean Used to represent the mean of R0(t);

[0047] R0std Used to represent the standard deviation of R0(t).

[0048] SDP image transformation is performed on the time-domain data to obtain SDP image data. Specifically, let the i-th sampling point R(i) and the sampling point R(i+j) that is j time intervals away from the i-th sampling point be transformed into polar coordinate space S(r(i),Θ(i),Φ(i)) according to the SDP principle. An example is shown below. Figure 2 As shown, the specific formula is as follows:

[0049]

[0050]

[0051]

[0052] In the above formulas (2)-(4):

[0053] R(i) is used to represent the normalized value of the i-th sample point of the current time-domain data;

[0054] R(i+j) is used to represent the normalized value of the sample point data that is j time intervals away from the i-th sample point data of the current time domain data;

[0055] r(i) is the polar radius of the point;

[0056] Θ(i) is the angle by which the point rotates clockwise about the plane of mirror symmetry;

[0057] Φ(i) is the angle by which the point is rotated counterclockwise about the mirror-symmetric plane;

[0058] θ is the rotation angle of the mirror-symmetric plane (θ = 360m / n, m = 1, 2, ..., n), and n is the number of mirror-symmetric planes;

[0059] ξ is the gain of the drawing angle (ξ≤θ);

[0060] j is the time interval factor (1≤j≤10);

[0061] Max(·) is the function that takes the largest value;

[0062] Min(·) is the function that takes the smallest value.

[0063] In the above embodiments, when converting normal variable data under normal operating conditions and fault variable data under fault conditions into SDP images, the adjustable parameters are θ, ξ, and j. The SDP image obtained by adjusting the parameters has a relatively simple graphic structure, is not too compact, and has relatively smooth edges, which is beneficial for fault diagnosis.

[0064] In the above embodiment, fault diagnosis is achieved by converting one-dimensional time-domain variable data into a two-dimensional mirror-symmetric snowflake image in polar coordinates and inputting the two-dimensional mirror-symmetric snowflake image into a pre-trained CNN neural network fault diagnosis model.

[0065] In the above embodiment, after step S102, the method further includes: performing image annotation on the preprocessed normal variable data and fault variable data.

[0066] Preferably, image annotation should include the following steps: annotating the converted SDP images according to the corresponding operating conditions, including normal operating conditions and fault conditions; and using the labelme tool to annotate the data based on the faults corresponding to abnormal variable data.

[0067] Step S103: Input the preprocessed normal variable data and fault variable data into the pre-trained CNN neural network fault diagnosis model for diagnosis, and obtain the fault diagnosis results of the air compression system.

[0068] The CNN neural network fault diagnosis model is trained by taking the SDP image data corresponding to the normal variable data and fault variable data of the air compression system as input and the fault diagnosis results corresponding to the normal variable data and fault variable data of the air compression system as output.

[0069] The CNN neural network fault diagnosis model includes an input layer, a backbone network, a hybrid attention module, a feature fusion module, and an output layer. The backbone network extracts features from the SDP image to obtain low-semantic feature maps and high-semantic feature maps. The hybrid attention module compresses the high-semantic feature map to obtain an enhanced feature map that fuses the channel attention feature map and the spatial attention feature map. The feature fusion module performs feature fusion operations on the low-semantic feature map and the enhanced feature map to obtain a fused feature map.

[0070] In the above embodiments, the CNN neural network fault diagnosis model is as follows: Figure 3 As shown, it includes an input layer, a backbone network, a hybrid attention module, a feature fusion module, two fully connected layers, and an output layer;

[0071] The input layer has 3 channels.

[0072] Backbone network: The network structure adopts 5 convolutional modules. The convolutional modules reconstruct the fault features and reduce the dimensionality of the feature vectors. Each module contains a convolutional layer, normalization processing, activation function and pooling. The first module receives the input and starts to extract low-level features. For example, the convolutional module may include: 3*3 convolution, batch normalization, activation function and average pooling.

[0073] The backbone network is the core of the model, which mainly extracts features from the image, extracts and compresses detailed information such as color, edge and shape to obtain abstract semantic information. The more times the convolution operation is performed, the less detailed information the features contain and the richer the semantic information.

[0074] It's important to note that activation functions are used to add non-linearity, because convolution methods are designed for linear operations, i.e., assigning weights to each pixel. Linear models lack sufficient expressive power, hence the introduction of activation functions. The ReLU function can be used here.

[0075] As a preferred embodiment, such as Figure 3 As shown, the feature sizes output by each convolutional layer in the backbone network from top to bottom are 320*320, 160*160, 80*80, 40*40 and 20*20, respectively, with 16, 32, 32, 64, 128 and 128 channels, respectively.

[0076] Hybrid Attention Module (CBAM): This module comprises a spatial attention module and a channel attention module. The CBAM obtains a fused enhanced feature map by compressing the high semantic feature map, resulting in a fusion of the channel and spatial attention feature maps. Specifically, it includes:

[0077] The high semantic feature map is input into the channel attention module, and the channel attention feature map is output.

[0078] The channel feature map is input into the spatial attention module. Based on the channel attention feature map, spatial information is further enhanced to generate an enhanced feature map that integrates the channel attention feature map and the spatial attention feature map.

[0079] The hybrid attention module, comprising spatial and channel attention, employs a focusing mechanism similar to focusing, highlighting features of interest while suppressing irrelevant features. Since SDP images exhibit varying complexity due to data differences, subtle variations may exist, such as jagged edges. To improve fault diagnosis accuracy, a hybrid attention model is used to focus on key regions, enabling the identification of subtle differences and thus enhancing diagnostic accuracy.

[0080] Feature fusion module (Fuse): Performs feature fusion operations on the low semantic feature map and the enhanced feature map to obtain a fused feature map, specifically including:

[0081] By scaling the low-semantic feature map and the enhanced feature map to the same size and superimposing them in the channel dimension, feature fusion is achieved.

[0082] Because features lose detailed information during convolution operations, making it difficult to distinguish subtle differences, a feature fusion module is used to fuse low-level features (features that undergo fewer convolution operations) with high-level features (features that undergo more convolution operations). This combines the detailed information of low-level features with the semantic information of high-level features, thereby improving the discrimination rate of subtle differences. For example, the duration of different faults varies, resulting in significant differences in the representation of SDP images, such as shape. The feature fusion module can fuse feature information from different levels, thereby improving the accuracy of fault diagnosis.

[0083] In a preferred embodiment, the feature map output by the feature fusion module has a size of 40*40 and 128 channels.

[0084] Two fully connected layers are used to integrate the extracted features;

[0085] The output layer outputs the fault diagnosis results.

[0086] As a preferred embodiment, the fault diagnosis results include: normal condition, component failure, etc.

[0087] In a preferred embodiment, the backbone network includes multiple convolutional modules, each module containing a convolutional layer, normalization processing, activation function and pooling. The first convolutional module receives input and begins to extract low-level features. At least one convolutional module is connected to a feature fusion module, and at least one convolutional module is connected to a hybrid attention module. The hybrid attention module is connected to the feature fusion module.

[0088] As a preferred embodiment, the backbone network may include five convolutional modules, wherein the first convolutional module receives input and begins to extract low-level features;

[0089] The second convolutional module is connected to the feature fusion module, at which point the feature fusion module can receive the low semantic feature map output by the second convolutional module;

[0090] The third to fifth convolutional modules are connected to the hybrid attention module. At this point, the hybrid attention module compresses the high semantic feature map to obtain an enhanced feature map that fuses the channel attention feature map and the spatial attention feature map.

[0091] The hybrid attention module is connected to the feature fusion module. At this time, the feature fusion module can receive the enhanced feature map output by the hybrid attention module.

[0092] For example, a CNN neural network fault diagnosis model, such as Figure 3As shown, it includes an input layer, a backbone network conv1-conv5, a hybrid attention module CBAM, a feature fusion module Fuse, a first fully connected layer Fc1, a second fully connected layer Fc2, and an output layer.

[0093] The backbone network consists of five convolutional modules, conv1-conv5, and three hybrid attention modules, referred to as the first hybrid attention module, the second hybrid attention module, and the third hybrid attention module, respectively.

[0094] The input layer is connected to the conv1 convolutional module;

[0095] The conv1 convolutional module is connected to the conv2 convolutional module, the conv2 convolutional module is connected to the feature fusion module, the conv2 convolutional module is connected to the conv3 convolutional module, the conv3 convolutional module is connected to the first hybrid attention module, and the first hybrid attention module is connected to the feature fusion module. At this time, the feature fusion module can receive the enhanced feature map output by the first hybrid attention module.

[0096] The conv3 convolutional module is connected to the conv4 convolutional module, the conv4 convolutional module is connected to the second hybrid attention module, and the second hybrid attention module is connected to the feature fusion module. At this time, the feature fusion module can receive the enhanced feature map output by the second hybrid attention module.

[0097] The conv4 convolutional module is connected to the conv5 convolutional module, the conv5 convolutional module is connected to the third hybrid attention module, and the third hybrid attention module is connected to the feature fusion module. At this time, the feature fusion module can receive the enhanced feature map output by the third hybrid attention module.

[0098] The feature fusion module is connected to the first fully connected layer, the first fully connected layer is connected to the second fully connected layer, and the second fully connected layer is connected to the output layer.

[0099] In the above implementation, for example, component failures include motor bearing failures. This dataset comes from Case Western Reserve University (CWRU), and the experiment was conducted using Python. Motor bearing failures further include three failure types, each with three different damage sizes, for a total of ten motor bearing conditions: nine failure conditions and one normal condition (NO). The three failure types are roller failure (RF), outer ring failure (OF), and inner ring failure (IF), which can be further categorized according to damage size as 0.18 mm, 0.36 mm, and 0.54 mm, respectively. The result of SDP transformation of the variable data is as follows: Figure 4 As shown, where ξ = 35 and j = 2.

[0100] As a preferred embodiment, the training process of the CNN neural network fault diagnosis model includes:

[0101] Obtain normal variable data, fault characteristic data, and fault type data associated with the variable data that are related to the target system.

[0102] Preprocess the normal variable data and fault characteristic data of the variable data;

[0103] After the preprocessed normal variable data and fault feature data are labeled according to the fault type data, they are randomly divided into training set, validation set and test set.

[0104] The CNN neural network fault diagnosis model is trained using variable data from the training set as input and fault type data from the training set as output. The model is evaluated using a validation set every few training cycles, and the diagnostic performance of the model is observed in real time. After training is completed, the trained CNN neural network fault diagnosis model is tested using a test set.

[0105] In the above implementation, when training the CNN neural network fault diagnosis model, the dataset is divided into a training set, a validation set, and a test set in a ratio of 8:1:1. The training set is used for training the model, the validation set is used to evaluate the model's performance, and finally, the final model parameters are determined based on the test results of the test set.

[0106] As a preferred implementation, a validation set is used to evaluate the model in real time, and a test set is used to test the trained CNN neural network fault diagnosis model, including:

[0107] The data in the validation set is input into the initial CNN neural network fault diagnosis model to obtain the fault type data output by the initial CNN neural network fault diagnosis model.

[0108] The data in the test set is input into the trained CNN neural network fault diagnosis model to obtain the fault type data output by the CNN neural network fault diagnosis model.

[0109] The accuracy of the CNN neural network fault diagnosis model after training is judged based on the fault type data in the test set. If the accuracy of the CNN neural network fault diagnosis model is greater than the preset test accuracy threshold, the CNN neural network fault diagnosis model that has passed the evaluation will be used as the fault diagnosis model of the air compression system; otherwise, the model parameters will be adjusted and retrained.

[0110] In the above embodiments, various variables of the air compression system (including normal variable data and fault variable data) are used as raw data. After SDP conversion, they are used as input to a pre-trained CNN neural network fault diagnosis model. The output of the pre-trained CNN neural network fault diagnosis model is the fault type related to the variable, such as abnormal pressure leading to excessive pressure. The robustness is improved by using the pre-trained CNN neural network fault diagnosis model. Since the CNN neural network fault diagnosis model has the ability to collect multi-source data and extract potential features from a large amount of data, and can process multiple variable data at the same time, it can more accurately reflect fault information and locate faults more accurately, thereby improving the accuracy of fault diagnosis in different working environments.

[0111] In the above embodiment, the time-domain data of various variables of the air compression system are first converted to SDP, labeled, and used as training samples to train a CNN neural network fault diagnosis model. The trained CNN neural network fault diagnosis model is then used to achieve fault diagnosis. Through SDP conversion, the frequency and amplitude changes of the time series can be intuitively reflected using the polar space SDP image, thereby enabling deeper exploration of the hidden features in the time and space of the fault data and further improving the accuracy of fault diagnosis.

[0112] Example 2

[0113] Figure 5 This is a structural diagram of a fault diagnosis system for an air compression system based on a CNN neural network, according to an embodiment of this application. Figure 5 As shown, the system includes:

[0114] The data acquisition module obtains normal variable data and fault variable data associated with the target system;

[0115] The preprocessing module preprocesses normal variable data and fault variable data;

[0116] The diagnostic module inputs the pre-processed normal variable data and fault variable data into a pre-trained CNN neural network fault diagnosis model for diagnosis, and obtains the fault diagnosis results of the air compression system.

[0117] The CNN neural network fault diagnosis model is trained by taking the SDP image data corresponding to the normal variable data and fault variable data of the air compression system as input and the fault diagnosis results corresponding to the normal variable data and fault variable data of the air compression system as output.

[0118] The CNN neural network fault diagnosis model includes an input layer, a backbone network, a hybrid attention module, a feature fusion module, and an output layer. The backbone network extracts features from the SDP image to obtain low-semantic feature maps and high-semantic feature maps. The hybrid attention module compresses the high-semantic feature map to obtain an enhanced feature map that fuses the channel attention feature map and the spatial attention feature map. The feature fusion module performs feature fusion operations on the low-semantic feature map and the enhanced feature map to obtain a fused feature map.

[0119] In the above embodiments, various variables of the air compression system (including normal variable data and fault variable data) are used as raw data. After SDP conversion, they are used as input to a pre-trained CNN neural network fault diagnosis model. The output of the pre-trained CNN neural network fault diagnosis model is the fault type related to the variable, such as abnormal pressure leading to excessive pressure. The robustness is improved by using the pre-trained CNN neural network fault diagnosis model. Since the CNN neural network fault diagnosis model has the ability to collect multi-source data and extract potential features from a large amount of data, and can process multiple variable data at the same time, it can more accurately reflect fault information and locate faults more accurately, thereby improving the accuracy of fault diagnosis in different working environments.

[0120] In the above embodiment, the time-domain data of various variables of the air compression system are first converted to SDP, labeled, and used as training samples to train a CNN neural network fault diagnosis model. The trained CNN neural network fault diagnosis model is then used to achieve fault diagnosis. Through SDP conversion, the frequency and amplitude changes of the time series can be intuitively reflected using the polar space SDP image, thereby enabling deeper exploration of the hidden features in the time and space of the fault data and further improving the accuracy of fault diagnosis.

[0121] Example 3

[0122] To implement the above embodiments, this disclosure also proposes an air compression system fault diagnosis manager based on a CNN neural network, wherein the fault diagnosis manager includes a processor and a memory, the memory storing at least one instruction, which is loaded and executed by the processor to perform the operation as described in Embodiment 1.

[0123] Example 4

[0124] To implement the above embodiments, this disclosure also proposes an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, it implements the method described in Embodiment 1.

[0125] Example 5

[0126] To implement the above embodiments, this disclosure also proposes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in Embodiment 1.

[0127] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0128] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0129] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A fault diagnosis method for an air compression system based on a CNN neural network, characterized in that, include: Acquire normal and fault variable data associated with the target system; The normal variable data and the fault variable data are preprocessed; The preprocessed normal variable data and fault variable data are input into a pre-trained CNN neural network fault diagnosis model for diagnosis, and the fault diagnosis results of the air compression system are obtained. The CNN neural network fault diagnosis model is trained by taking the SDP image data corresponding to the normal variable data and fault variable data of the air compression system as input and the fault diagnosis results corresponding to the normal variable data and fault variable data of the air compression system as output. The CNN neural network fault diagnosis model includes an input layer (input), a backbone network (conv1-conv5), a hybrid attention module (CBAM), a feature fusion module (Fuse), a first fully connected layer (Fc1), a second fully connected layer (Fc2), and an output layer (output). The backbone network consists of five convolutional modules, conv1-conv5, and three hybrid attention modules, referred to as the first hybrid attention module, the second hybrid attention module, and the third hybrid attention module, respectively. The input layer is connected to the conv1 convolutional module; The conv1 convolutional module is connected to the conv2 convolutional module, the conv2 convolutional module is connected to the feature fusion module, the conv2 convolutional module is connected to the conv3 convolutional module, the conv3 convolutional module is connected to the first hybrid attention module, and the first hybrid attention module is connected to the feature fusion module. At this time, the feature fusion module can receive the enhanced feature map output by the first hybrid attention module. The conv3 convolutional module is connected to the conv4 convolutional module, the conv4 convolutional module is connected to the second hybrid attention module, and the second hybrid attention module is connected to the feature fusion module. At this time, the feature fusion module can receive the enhanced feature map output by the second hybrid attention module. The conv4 convolutional module is connected to the conv5 convolutional module, the conv5 convolutional module is connected to the third hybrid attention module, and the third hybrid attention module is connected to the feature fusion module. At this time, the feature fusion module can receive the enhanced feature map output by the third hybrid attention module. The feature fusion module is connected to the first fully connected layer, the first fully connected layer is connected to the second fully connected layer, and the second fully connected layer is connected to the output layer. The backbone network extracts features from the SDP image to obtain low semantic feature maps and high semantic feature maps; the hybrid attention module compresses the high semantic feature map to obtain an enhanced feature map that fuses the channel attention feature map and the spatial attention feature map; the feature fusion module performs feature fusion operation on the low semantic feature map and the enhanced feature map to obtain a fused feature map.

2. The air compression system fault diagnosis method based on CNN neural network as described in claim 1, characterized in that, The training process of the CNN neural network fault diagnosis model includes: Obtain normal variable data associated with the target system, fault characteristic data corresponding to the variable data, and fault type data corresponding to the fault characteristic data; The normal variable data and the fault characteristic data of the variable data are preprocessed; After the preprocessed normal variable data and the fault feature data are labeled according to the fault type data, they are randomly divided into training set, validation set and test set; The CNN neural network fault diagnosis model is trained using variable data from the training set as input and fault type data from the training set as output. The model is evaluated using a validation set every few training cycles, and the diagnostic performance of the model is observed in real time. After training, the trained CNN neural network fault diagnosis model is tested using a test set.

3. The air compression system fault diagnosis method based on CNN neural network as described in claim 2, characterized in that, The process of evaluating the model using a validation set and testing the evaluated CNN neural network fault diagnosis model using a test set to obtain a trained CNN neural network fault diagnosis model includes: The data in the validation set is input into the initial CNN neural network fault diagnosis model to obtain the fault type data output by the initial CNN neural network fault diagnosis model. The data in the test set is input into the trained CNN neural network fault diagnosis model to obtain the fault type data output by the CNN neural network fault diagnosis model. The accuracy of the CNN neural network fault diagnosis model after training is judged based on the fault type data in the test set. If the accuracy of the CNN neural network fault diagnosis model is greater than the preset test accuracy threshold, the CNN neural network fault diagnosis model that has passed the evaluation is regarded as the trained CNN neural network fault diagnosis model; otherwise, the model parameters are adjusted and retraining is performed.

4. The air compression system fault diagnosis method based on CNN neural network as described in claim 1, characterized in that, The backbone network consists of multiple convolutional modules, each containing convolutional layers, normalization processing, activation functions, and pooling. The first convolutional module receives input and begins to extract low-level features. In this configuration, at least one convolutional module is connected to the feature fusion module, at least one convolutional module is connected to the hybrid attention module, and the hybrid attention module is connected to the feature fusion module.

5. The air compression system fault diagnosis method based on CNN neural network as described in claim 1, characterized in that, The preprocessing includes normalizing the normal variable data and the fault variable data, and performing SDP image conversion to obtain SDP image data; including: The data is normalized to obtain time-domain data, as shown in the following formula: ； In the above formula, R0(t) is used to represent the original data; R(i) is used to represent the normalized value of the i-th sample point of the current time-domain data; R0(i) is used to represent the value of R0(t) at the i-th sampling point; t is used to represent time; R 0mean Used to represent the mean of R0(t); R 0std Used to represent the standard deviation of R0(t); The time-domain data is converted to SDP image data as shown in the following formula: ； ； ； In the above formula: R(i) is used to represent the normalized value of the i-th sample point of the current time-domain data; R(i+j) is used to represent the normalized value of the sample point data that is j time intervals away from the i-th sample point data of the current time domain data; r(i) is used to represent the polar coordinate radius of a point; Θ(i) is used to represent the angle of clockwise rotation of a point about the plane of mirror symmetry; Φ(i) is used to represent the angle of counterclockwise rotation of a point about the mirror-symmetric plane; θ is used to represent the rotation angle of the mirror-symmetric plane θ = 360m / n, m = 1, 2, ..., n, where n is the number of mirror-symmetric planes; ξ is used to represent the gain of the drawing angle, where ξ ≤ θ; j is used to represent the time interval factor 1≤j≤10; Used to represent functions that take the largest value; Used to represent the function that takes the smaller value.

6. The air compression system fault diagnosis method based on CNN neural network as described in claim 1, characterized in that, The hybrid attention module includes: a spatial attention module and a channel attention module; The hybrid attention module obtains an enhanced feature map by fusing channel attention feature maps and spatial attention feature maps through feature compression of the high semantic feature map. Specifically, it includes: The high semantic feature map is input into the channel attention module, and the channel attention feature map is output. The channel attention feature map is input into the spatial attention module, and spatial information is further enhanced based on the channel attention feature map to generate an enhanced feature map that integrates the channel attention feature map and the spatial attention feature map.

7. The air compression system fault diagnosis method based on CNN neural network as described in claim 1, characterized in that, Variable data, including: compressed air generation variables and compressed air process variables; The variables generated by compressed air include: the flow rate, pressure, and temperature of the compressed air. Compressed air process variables include: standard volumetric flow rate, cooling temperature, drying temperature, and pressurized tank pressure.

8. A fault diagnosis system for an air compression system based on a CNN neural network, characterized in that, include: The data acquisition module obtains normal variable data and fault variable data associated with the target system; The preprocessing module preprocesses the normal variable data and the fault variable data; The diagnostic module inputs the preprocessed normal variable data and the fault variable data into a pre-trained CNN neural network fault diagnosis model for diagnosis, and obtains the fault diagnosis results of the air compression system. The CNN neural network fault diagnosis model is trained by taking the SDP image data corresponding to the normal variable data and fault variable data of the air compression system as input and the fault diagnosis results corresponding to the normal variable data and fault variable data of the air compression system as output. The CNN neural network fault diagnosis model includes an input layer (input), a backbone network (conv1-conv5), a hybrid attention module (CBAM), a feature fusion module (Fuse), a first fully connected layer (Fc1), a second fully connected layer (Fc2), and an output layer (output). The backbone network consists of five convolutional modules, conv1-conv5, and three hybrid attention modules, referred to as the first hybrid attention module, the second hybrid attention module, and the third hybrid attention module, respectively. The input layer is connected to the conv1 convolutional module; The conv1 convolutional module is connected to the conv2 convolutional module, the conv2 convolutional module is connected to the feature fusion module, the conv2 convolutional module is connected to the conv3 convolutional module, the conv3 convolutional module is connected to the first hybrid attention module, and the first hybrid attention module is connected to the feature fusion module. At this time, the feature fusion module can receive the enhanced feature map output by the first hybrid attention module. The conv3 convolutional module is connected to the conv4 convolutional module, the conv4 convolutional module is connected to the second hybrid attention module, and the second hybrid attention module is connected to the feature fusion module. At this time, the feature fusion module can receive the enhanced feature map output by the second hybrid attention module. The conv4 convolutional module is connected to the conv5 convolutional module, the conv5 convolutional module is connected to the third hybrid attention module, and the third hybrid attention module is connected to the feature fusion module. At this time, the feature fusion module can receive the enhanced feature map output by the third hybrid attention module. The feature fusion module is connected to the first fully connected layer, the first fully connected layer is connected to the second fully connected layer, and the second fully connected layer is connected to the output layer. The backbone network extracts features from the SDP image to obtain low semantic feature maps and high semantic feature maps; the hybrid attention module compresses the high semantic feature map to obtain an enhanced feature map that fuses the channel attention feature map and the spatial attention feature map; the feature fusion module performs feature fusion operation on the low semantic feature map and the enhanced feature map to obtain a fused feature map.

9. A fault diagnosis manager for an air compression system based on a CNN neural network, characterized in that, The fault diagnosis manager includes a processor and a memory, the memory storing at least one instruction, which is loaded and executed by the processor to perform the operations of the air compression system fault diagnosis method based on any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The storage medium stores at least one instruction, which is loaded and executed by a processor to perform the operation of the air compression system fault diagnosis method based on any one of claims 1-7.

Images