Industrial equipment state monitoring method, system, medium and equipment

Abstract

The invention discloses an industrial equipment state monitoring method and system, a medium and equipment, and relates to the technical field of state monitoring, and the method comprises the steps: collecting monitoring signals of industrial equipment in various operation states through a sensor, and carrying out the preprocessing; performing time-frequency domain conversion on the preprocessed monitoring signal to obtain a monitoring feature data set, and dividing the monitoring feature data set into a training set and a verification set according to different proportions; constructing a residual multi-scale state monitoring model, and training and verifying the residual multi-scale state monitoring network model by using the training set and the verification set to obtain a final state monitoring model; the state monitoring model comprises a pre-processing layer, a plurality of feature extraction layers and an operation state discrimination layer which are sequentially connected in series, and the feature extraction layers are composed of a plurality of Bottle2check modules; and performing state monitoring on a to-be-monitored signal by using the final state monitoring model. According to the invention, the operation state of the industrial equipment is accurately judged.

Description

Technical Field

[0001] The present invention relates to the technical field of condition monitoring, and in particular to an industrial equipment condition monitoring method, system, medium and equipment. Background Art

[0002] As the core of the manufacturing industry, industrial equipment has always been the focus of research and development in the manufacturing industry. With the widespread application of intelligent manufacturing and automation technologies in the industrial field, the operational reliability and production efficiency of industrial equipment have become the focus of enterprises. However, the components of industrial equipment will produce problems such as wear and aging during long-term work. This will not only increase maintenance costs, but may also cause serious production accidents and economic losses. Therefore, in order to ensure the safe and reliable production of industrial equipment, various operating status monitoring methods have begun to be continuously applied in the field of industrial equipment.

[0003] Common methods for monitoring the condition of industrial equipment include rule-based and model-based monitoring methods. These methods rely on expert systems and modeling to achieve real-time monitoring of operating conditions, but these methods have poor generalization capabilities and are difficult to adapt to the increasingly diverse market environment. On the other hand, the rapid development of machine learning technology has spawned data-driven equipment condition monitoring methods. Data-driven intelligent condition monitoring methods no longer rely on expert knowledge and have strong data adaptability, becoming the preferred method for monitoring the condition of industrial machinery.

[0004] As the operating conditions of industrial equipment become more complex, the collected operating condition data also becomes more complicated. When extracting features from the operating status data generated under different operating conditions, the features extracted under different operating conditions will differ in expression scale. Current data-driven industrial equipment status monitoring methods often ignore these differences in the scale of extracted features and cannot fully utilize feature information at different scales, resulting in a decrease in the quality of the collected feature information, leading to a decrease in the accuracy of operating condition monitoring, which restricts the development of data-driven industrial equipment status monitoring methods.

[0005] In summary, existing data-driven intelligent condition monitoring methods cannot fully utilize feature information at different scales. Summary of the invention

[0006] The purpose of the present invention is to solve the problem that the existing state monitoring method cannot fully utilize the characteristic information at different scales, and propose an industrial equipment state monitoring method, comprising the following steps:

[0007] S1. Collect monitoring signals of industrial equipment under various operating conditions through sensors and perform preprocessing;

[0008] S2. Perform time-frequency domain conversion on the preprocessed monitoring signal to obtain a monitoring feature data set, and divide the monitoring feature data set into a training set and a validation set according to different proportions;

[0009] S3, construct a residual multi-scale state monitoring model, use the training set and the validation set to train and validate the residual multi-scale state monitoring network model, and obtain the final state monitoring model;

[0010] The state monitoring model comprises a pre-processing layer, a plurality of feature extraction layers and an operation state discrimination layer connected in series, wherein the feature extraction layer is composed of a plurality of Bottle2neck modules;

[0011] S4. Use the final state monitoring model to perform state monitoring on the signal to be monitored.

[0012] Furthermore, the preprocessing method includes: truncating the monitoring signal segments during the startup and shutdown phases of the industrial equipment, and removing abnormal values ​​of the monitoring signal by using an isolation forest method.

[0013] Furthermore, the one-dimensional monitoring signal is converted into a two-dimensional feature map through short-time Fourier transform to obtain a monitoring feature data set.

[0014] Furthermore, the pre-processing layer includes: a convolution layer, a normalization layer, a ReLU activation layer, and a maximum pooling layer; the pre-processing layer completes the preliminary extraction of monitoring features.

[0015] Furthermore, a 7×7 convolution kernel is set in the convolution layer of the pre-processing layer.

[0016] Further, the condition monitoring model includes four feature extraction layers, a first feature extraction layer has three Bottle2neck modules connected in series, a second feature extraction layer has four Bottle2neck modules connected in series, a third feature extraction layer has six Bottle2neck modules connected in series, and a fourth feature extraction layer has three Bottle2neck modules connected in series;

[0017] The Bottle2neck module includes a dimensionality reduction convolution layer, a feature segmentation layer, a multi-scale feature extraction layer, a feature merging layer, and a channel recovery layer, which are connected in series. The input of the dimensionality reduction convolution layer is connected to the output of the channel recovery layer through a residual connection.

[0018] The input feature map x0 is convolved with a 1×1 convolution operation through the dimensionality reduction convolution layer to obtain the feature map x. The feature segmentation layer evenly divides the output feature map x of the dimensionality reduction convolution layer into s feature map subsets. The feature map subsets are denoted by x. i Represents, where i∈{1,2,...,s};

[0019] x iInput the multi-scale feature extraction layer to get y i , expressed as:

[0020]

[0021] G1=σ(W1*x1+b1)

[0022] G2=σ(W2*K2+b2)

[0023] Among them, ⊙ represents element-by-element multiplication, which is used to apply the gated map to filter the information in the feature map; G1 and G2 represent the gated maps of different inputs, respectively, and K i () represents the i-th convolution operator, W1 and W2 are learnable convolution kernels, b1 and b2 are bias terms, and σ is the sigmoid activation function;

[0024] y i After the feature merging layer performs channel splicing to obtain y, y outputs the operating status of the industrial equipment through the operating status discrimination layer.

[0025] Furthermore, the running state discrimination layer consists of an average pooling.

[0026] The present invention also provides an industrial equipment status monitoring system, comprising:

[0027] The data acquisition module is used to collect monitoring signals of industrial equipment under various operating conditions through sensors and perform preprocessing;

[0028] A data processing module is used to perform time-frequency domain conversion on the preprocessed monitoring signal, obtain a monitoring feature data set, and divide the monitoring feature data set into a training set and a validation set according to different proportions;

[0029] The model building module is used to build a residual multi-scale condition monitoring model, and use the training set and validation set to train and validate the residual multi-scale condition monitoring network model to obtain the final condition monitoring model;

[0030] The state monitoring model comprises: a pre-processing layer, a plurality of feature extraction layers and an operation state discrimination layer connected in series in sequence, wherein the feature extraction layer is composed of a plurality of Bottle2neck modules;

[0031] The condition monitoring module is used to perform condition monitoring on the signal to be monitored using the final condition monitoring model.

[0032] The present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned industrial equipment status monitoring method is implemented.

[0033] The present invention also proposes an electronic device, including a processor, a memory, a user interface and a network interface, wherein the memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device executes the above-mentioned industrial equipment status monitoring method.

[0034] The beneficial effects brought by the technical solution provided by the present invention are:

[0035] The monitoring model of the present invention is provided with multiple feature extraction layers for extracting monitoring signal features, each feature extraction layer includes multiple Bottle2neck modules, and feature extraction is performed on feature map subsets through multiple parallel feature extraction convolution layers, and channel splicing operations are performed, thereby improving the model's ability to extract feature expression of monitoring signals at different scales, so that the model can make full use of feature information at different scales to achieve accurate judgment of the operating status of industrial equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 is a flow chart of an industrial equipment status monitoring method according to an embodiment of the present invention;

[0037] Figure 2 is a structural diagram of a state monitoring model according to an embodiment of the present invention;

[0038] Figure 3 This is the Bottle2neck module structure diagram;

[0039] Figure 4 is an accuracy curve diagram of the operating status monitoring method according to an embodiment of the present invention;

[0040] Figure 5 It is a block diagram of an electronic device in an exemplary embodiment of the present invention. DETAILED DESCRIPTION

[0041] To make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0042] Example 1: A flowchart of the industrial equipment status monitoring method according to an embodiment of the present invention is as follows Figure 1 , specifically including the following steps:

[0043] S1. Collect monitoring signals of industrial equipment under various operating conditions through sensors and perform preprocessing. Different sensors are required for different industrial equipment. For example, vibration sensors can measure the vibration of objects and are used to monitor the operating status of mechanical equipment, such as gear boxes, motors, pumps, compressors, etc. By monitoring vibration signals, gear meshing problems and bearing failures can be discovered in time. Temperature sensors measure temperature changes and output electrical signals. They are used to monitor the operating temperature of equipment to prevent overheating that affects product quality or equipment life, such as temperature monitoring of motors, pumps, compressors, bearings and other equipment. Pressure sensors detect the pressure of gas or liquid and convert it into electrical signals. They are widely used in hydraulic and pneumatic system control to detect the working status of pumps and compressors. Flow sensors measure the flow of liquid or gas and are used to monitor the flow of fluids in pipelines, channels or systems to ensure the normal operation of equipment, such as flow monitoring of pumps, compressors and other equipment.

[0044] The preprocessing method includes: (1) truncating the unstable monitoring signal segments during the startup and shutdown stages of industrial equipment; and (2) removing outliers from the monitoring signal through the isolation forest method. The isolation forest is an unsupervised learning algorithm for anomaly monitoring that can effectively remove outliers from the monitoring signal.

[0045] S2. Perform time-frequency domain conversion on the preprocessed monitoring signal to obtain a monitoring feature data set, and divide the monitoring feature data set into a training set and a validation set according to different proportions.

[0046] The one-dimensional monitoring signal is converted into a two-dimensional feature graph through short-time Fourier transform to obtain a monitoring feature data set. Short-time Fourier transform (STFT) is a method for converting a one-dimensional time domain signal into a two-dimensional time-frequency domain representation. The spectrum results of each frame are arranged in chronological order, where the horizontal axis represents time and the vertical axis represents frequency. Through STFT, the changes in the frequency components of the monitoring signal over time can be visualized as a two-dimensional feature graph, which makes it easier to analyze the time-frequency characteristics of the signal.

[0047] S3. Construct a residual multi-scale state monitoring model, use the training set and validation set to train and validate the residual multi-scale state monitoring network model, and obtain the final state monitoring model.

[0048] The condition monitoring model includes: a pre-processing layer, multiple feature extraction layers and an operation status discrimination layer connected in series, wherein the feature extraction layer is composed of multiple Bottle2neck modules. Figure 2 , Figure 2 It is a structural diagram of a state monitoring model according to an embodiment of the present invention.

[0049] The pre-processing layer includes: a convolution layer, a normalization layer, a ReLU activation layer, and a maximum pooling layer; the pre-processing layer completes the preliminary extraction of monitoring features. The convolution layer of the pre-processing layer sets a 7×7 convolution kernel to capture a large range of local information in the feature map, and then connects to the normalization layer for accelerating training and improving model stability. The features after the normalization layer are then connected to the ReLU activation function layer to improve the nonlinearity of the model.

[0050] The condition monitoring model includes four feature extraction layers, the first feature extraction layer includes three Bottle2neck modules, the second feature extraction layer includes four Bottle2neck modules, the third feature extraction layer includes six Bottle2neck modules, and the fourth feature extraction layer includes three Bottle2neck modules.

[0051] The running state discrimination layer consists of an average pooling.

[0052] The Bottle2neck module includes a dimensionality reduction convolution layer, a feature segmentation layer, a multi-scale feature extraction layer, a feature merging layer, and a channel restoration layer, which are connected in series. The input of the dimensionality reduction convolution layer is connected to the output of the channel restoration layer through a residual connection. The dimensionality reduction convolution layer and the channel restoration layer are set to 1×1 convolution.

[0053] The input feature map x0 is convolved with a 1×1 convolution operation through the dimensionality reduction convolution layer to obtain the feature map x. The feature segmentation layer evenly divides the output feature map x of the dimensionality reduction convolution layer into s feature map subsets. The feature map subsets are denoted by x. i Represents, where i∈{1,2,...,s}. Each feature subset x i It has the same spatial size as the feature map x0, but the number of channels is 1 / s of the feature map x0.

[0054] x i Input the multi-scale feature extraction layer to get y i In the multi-scale feature extraction layer, except for the first feature map subset x1, each feature map subset x i There are corresponding 3×3 convolutions, using K i Indicates. Use y i K i () output. When i is greater than 3, the feature subset x i With K i-1 The output of () is added as K i(). In order to reduce the parameters while increasing s, the 3×3 convolution of x1 is omitted. At the same time, in order to improve the feature expression ability of the first feature map subset x1, a gated feature fusion module (Gated Fully Fusion, GFF) is set up to fuse the input from the first feature subset x1 and K2(). Therefore, y i can be written as:

[0055]

[0056] G1=σ(W1*x1+b1)

[0057] G2=σ(W2*K2+b2)

[0058] Where ⊙ represents element-by-element multiplication, which is used to apply the gated map to filter the information in the feature map. G1 and G2 represent the gated maps of different inputs, respectively, and K i () represents the i-th convolution operator, W1 and W2 are learnable convolution kernels, b1 and b2 are bias terms, and σ is the sigmoid activation function.

[0059] Each 3×3 convolution operator K i () can be obtained from all feature subsets {x j ,j≤i} receives feature information. Feature subset x j After the 3×3 convolution operator, the output result has a ratio of x j Larger receptive fields. The output of the Bottle2neck module contains different numbers and combinations of receptive fields.

[0060] y i ,i∈{1,2,...,s} is spliced ​​through the feature merging layer, and multiple small feature maps y after parallel processing of the multi-scale feature extraction layer are i Splice in the dimension direction to increase the dimension of the feature.

[0061] Finally, the operating status of the industrial equipment is output through the operating status judgment layer. Figure 3 , Figure 3 This is the Bottle2neck module structure diagram. Figure 3 The s in is 4.

[0062] S4. Use the final state monitoring model to perform state monitoring on the signal to be monitored.

[0063] Embodiment 2: The present invention also provides an industrial equipment status monitoring system, comprising:

[0064] The data acquisition module is used to collect monitoring signals of industrial equipment under various operating conditions through sensors and perform preprocessing;

[0065] A data processing module is used to perform time-frequency domain conversion on the preprocessed monitoring signal, obtain a monitoring feature data set, and divide the monitoring feature data set into a training set and a validation set according to different proportions;

[0066] The model building module is used to build a residual multi-scale condition monitoring model, and use the training set and validation set to train and validate the residual multi-scale condition monitoring network model to obtain the final condition monitoring model;

[0067] The state monitoring model comprises: a pre-processing layer, a plurality of feature extraction layers and an operation state discrimination layer connected in series in sequence, wherein the feature extraction layer is composed of a plurality of Bottle2neck modules;

[0068] The condition monitoring module is used to perform condition monitoring on the signal to be monitored using the final condition monitoring model.

[0069] In order to prove the effectiveness of this method, the following gearbox experiment was carried out. This test bench consists of two motors, a gearbox, a flywheel, a data acquisition board and a computer. The two motor models used in the experiment are ABBmv1008-225 (1.2kW). One of them is used as a prime mover to drive the multi-stage gearbox, and the other is used as an asynchronous generator to simulate various resistance torques. All components are connected to the drive shaft through a coupling. During the test, the input speed of the equipment was set to 1400rpm, and the speeds of the two meshing gears in the gearbox were 1184rpm and 840rpm respectively. Seven types of operating states of the gearbox were simulated and tested to verify this method.

[0070] The data set contains vibration signals collected under normal operating conditions and six abnormal operating conditions. They are collected by two vibration sensors, the collection direction is perpendicular to the axis direction, and the sampling frequency is 10240Hz. The sampling period is set to 100s, that is, 4096000 data points are obtained under each condition. The signal is divided into 1400 samples, each containing 4000 data points. For each condition, the signal samples are divided into a training set and a validation set with a ratio of 800:200.

[0071] As shown in Table 1, the operation status monitoring of the confusion matrix visualization on the data set is shown, with labels C0-healthy, C1-broken teeth, C2-tooth pieces falling off, C3-tooth pieces cracked, C4-bearing outer ring cracked, C6-loose gear fixing plate, C7-axial imbalance, where the numbers represent the number of samples, for example, the number of samples with the predicted label C0 and the true label C0 is 207; the number of samples with the predicted label C0 and the true label C1 is 1, indicating that this sample is misclassified by the model. It can be seen from Table 1 that the labels all converge on the diagonal of the confusion matrix, and most of the true labels of various operation states can correspond one-to-one with the predicted labels. It can be seen that the operation status monitoring method of the present invention can achieve accurate classification of various operation states, indicating that the operation status monitoring method of the present invention can well complete the operation status monitoring task.

[0072] Table 1

[0073]

[0074]

[0075] As shown in Table 2 or Figure 4 As shown in FIG. 1 , four running status monitoring methods, namely VGG16, Resnet50, Resnet18 and GoogLeNet, are used to compare with the running status monitoring method of the present invention. Table 2 shows the results of the running status monitoring method of the present invention and the other four running status monitoring methods. It can be seen that the accuracy and F1 score of the running status monitoring method of the present invention are higher than those of the other four methods as a whole. Figure 4 The accuracy curve of the operating status monitoring method of the present invention is shown in the figure. In this experiment, the accuracy increased rapidly in the first few tests, and then fluctuated at a high accuracy, further verifying the superiority of the method of the present invention for operating status monitoring of industrial equipment.

[0076] Table 2

[0077] Model Prediction accuracy (%) Model of the present invention 99.319 VGG16 99.150 Resnet50 98.231 Resnet18 98.521 GoogLeNet 95.248

[0078] Embodiment 3: In an exemplary embodiment, a computer-readable storage medium is included, the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned industrial equipment status monitoring method is implemented.

[0079] Example 4: Please refer to Figure 5 In an exemplary embodiment, an electronic device 500 is also included, including at least one processor 501 , at least one network interface 504 , a user interface 503 , a memory 505 , and at least one communication bus 502 .

[0080] The communication bus 502 is used to realize the connection and communication between these components.

[0081] The processor 501 is used to execute instructions stored in the memory 505 so that the electronic device 500 executes the above-mentioned industrial equipment status monitoring method.

[0082] The user interface 503 may include a display screen, and the optional user interface 503 may also include a standard wired interface or a wireless interface.

[0083] The network interface 504 may optionally include a standard wired interface or a wireless interface (such as a WI-FI interface).

[0084] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for monitoring the state of industrial equipment, characterized in that: The following steps are involved: S1. Collect monitoring signals of industrial equipment under various operating conditions through sensors and perform preprocessing; S2. Perform time-frequency domain conversion on the preprocessed monitoring signal to obtain a monitoring feature data set, and divide the monitoring feature data set into a training set and a validation set according to different proportions; S3, construct a residual multi-scale state monitoring model, use the training set and the validation set to train and validate the residual multi-scale state monitoring network model, and obtain the final state monitoring model; The state monitoring model comprises a pre-processing layer, a plurality of feature extraction layers and an operation state discrimination layer connected in series, wherein the feature extraction layer is composed of a plurality of Bottle2neck modules; S4. Use the final state monitoring model to perform state monitoring on the signal to be monitored.

2. The method for monitoring the state of industrial equipment according to claim 1, characterized in that: The preprocessing method includes: truncating the monitoring signal segments during the startup and shutdown phases of industrial equipment, and removing abnormal values ​​of the monitoring signal through the isolation forest method.

3. The industrial equipment status monitoring method according to claim 1, characterized in that: The one-dimensional monitoring signal is converted into a two-dimensional feature map through short-time Fourier transform to obtain a monitoring feature data set.

4. The method for monitoring the state of industrial equipment according to claim 1, characterized in that: The pre-processing layer includes: a convolution layer, a normalization layer, a ReLU activation layer, and a maximum pooling layer; the pre-processing layer completes the preliminary extraction of monitoring features.

5. The industrial equipment status monitoring method according to claim 4, characterized in that: The convolution layer of the pre-processing layer is set with a 7×7 convolution kernel.

6. The method for monitoring the state of industrial equipment according to claim 1, characterized in that: The condition monitoring model includes four feature extraction layers, the first feature extraction layer has three Bottle2neck modules in series, the second feature extraction layer has four Bottle2neck modules in series, the third feature extraction layer has six Bottle2neck modules in series, and the fourth feature extraction layer has three Bottle2neck modules in series; The Bottle2neck module includes a dimensionality reduction convolution layer, a feature segmentation layer, a multi-scale feature extraction layer, a feature merging layer, and a channel recovery layer, which are connected in series. The input of the dimensionality reduction convolution layer is connected to the output of the channel recovery layer through a residual connection. The input feature map x0 is convolved with a 1×1 convolution operation through the dimensionality reduction convolution layer to obtain the feature map x. The feature segmentation layer evenly divides the output feature map x of the dimensionality reduction convolution layer into s feature map subsets. The feature map subsets are denoted by x. i Represents, where i∈{1,2,...,s}; Feature map subset x i Input the multi-scale feature extraction layer to obtain feature y i , expressed as: G1=σ(W1*x1+b1) G2=σ(W2*K2+b2) Among them, ⊙ represents element-by-element multiplication, which is used to apply the gated map to filter the information in the feature map; G1 and G2 represent the gated maps of different inputs, respectively, and K i () represents the i-th convolution operator, W1 and W2 are learnable convolution kernels, b1 and b2 are bias terms, and σ is the sigmoid activation function; y i After the feature merging layer performs channel splicing to obtain y, y outputs the operating status of the industrial equipment through the operating status discrimination layer.

7. The industrial equipment status monitoring method according to claim 1, characterized in that: The running state discrimination layer consists of an average pooling.

8. An industrial equipment status monitoring system, characterized in that: include: The data acquisition module is used to collect monitoring signals of industrial equipment under various operating conditions through sensors and perform preprocessing; A data processing module is used to perform time-frequency domain conversion on the preprocessed monitoring signal, obtain a monitoring feature data set, and divide the monitoring feature data set into a training set and a validation set according to different proportions; The model building module is used to build a residual multi-scale condition monitoring model, and use the training set and validation set to train and validate the residual multi-scale condition monitoring network model to obtain the final condition monitoring model; The state monitoring model comprises: a pre-processing layer, a plurality of feature extraction layers and an operation state discrimination layer connected in series in sequence, wherein the feature extraction layer is composed of a plurality of Bottle2neck modules; The condition monitoring module is used to perform condition monitoring on the signal to be monitored using the final condition monitoring model.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

10. An electronic device, characterized in that: The electronic device (500) comprises a processor (501), a memory (505), a user interface (503) and a network interface (504), wherein the memory (505) is used to store instructions, the user interface (503) and the network interface (504) are used for communication, and the processor (501) is used to execute the instructions stored in the memory (505) so that the electronic device (500) executes the method according to any one of claims 1 to 7.

Images