Industrial equipment start-stop state judgment method and system based on vibration waveform characteristic matrix, equipment and medium

Abstract

The invention relates to the technical field of intelligent operation and maintenance of industrial equipment, in particular to an industrial equipment start-stop state judgment method and system based on a vibration waveform characteristic matrix, equipment and a medium. And a deep learning model with a specific hidden layer structure (8-16-8 neurons) is adopted for training and classification, so that high-precision identification of four states of shutdown, startup, shutdown-to-startup and startup-to-shutdown is realized. According to the method, the problem of misjudgment caused by equipment difference and environment interference in a traditional method is effectively solved by continuously optimizing the model through a dynamic updating mechanism, and the reliability of industrial equipment fault diagnosis is remarkably improved. Meanwhile, the invention further provides a corresponding system, electronic equipment and a computer readable storage medium, and practical application deployment is facilitated.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent operation and maintenance of industrial equipment, and particularly relates to an industrial equipment start-stop state judgment method, system, device and medium based on a vibration waveform feature matrix. BACKGROUND

[0002] Collecting high-frequency vibration waveforms for data analysis is one of the most common methods in the field of equipment fault diagnosis. In the implementation process of the equipment intelligent operation and maintenance system, a vibration acceleration sensor needs to be installed at the corresponding measuring point of the equipment. The sensor sends the collected vibration waveforms to the server end according to a certain communication method. The server end receives and stores the vibration waveform data for subsequent data analysis. The industrial equipment start-stop state is divided into four types: stop state, start state, stop-to-start process state, and start-to-stop process state. Except for the start state, the vibration waveforms collected in the other three states cannot be used in the equipment fault diagnosis process and will interfere with the diagnosis results, so the stop data needs to be excluded.

[0003] The traditional method cannot accurately distinguish the four start-stop states (stop, start, stop-to-start process, and start-to-stop process) of the industrial equipment due to the dependence on fixed threshold values, especially the dynamic vibration features in the transition state. At the same time, the vibration level fluctuations caused by the size, installation environment, and process differences of the equipment will significantly reduce the judgment accuracy, and the stop and transition state data that have not been effectively excluded will introduce noise interference in the fault diagnosis. SUMMARY

[0004] The present application aims to provide an industrial equipment start-stop state judgment method, system, device and medium based on a vibration waveform feature matrix to solve the limitations of the traditional fixed threshold method in the prior art.

[0005] The present application achieves the above-mentioned purpose through the following technical solutions: In a first aspect, the present application provides an industrial equipment start-stop state judgment method based on a vibration waveform feature matrix, which comprises: S1. Collecting vibration waveform data sets under different start-stop states of the equipment and adding start-stop state labels, wherein the start-stop states include stop state, start state, stop-to-start process state, and start-to-stop process state; S2. Extracting time domain feature values and frequency domain feature values of each vibration signal in the vibration waveform data set; S3. Combining the time domain feature values and frequency domain feature values of each vibration signal to form a feature value matrix containing the start-stop state labels; S4. Construct a judgment model based on a deep learning algorithm comprising an input layer, an output layer and at least three hidden layers, and train the judgment model using the feature value matrix and the corresponding start-stop state labels; S5. Input the vibration waveform data of the device to be tested into the trained judgment model after converting the vibration waveform data into the feature value matrix, and output the current start-stop state of the device.

[0006] Further, the vibration waveform data set of the acquisition device under different start-stop states specifically comprises: S11. Collect the vibration waveform data set A of the device under different start-stop states as follows: is the mth start-stop state, is the nth vibration signal waveform under the mth start-stop state; S12. Set the number of vibration waveforms under four start-stop states in the vibration waveform data set A as 1:1:1:1.

[0007] Further, in step S2, the time domain feature values include a minimum value min, a maximum value max, an average value rms, a mean square root value mean, a square root amplitude value root, an average amplitude value ave, a waveform index sha, a peak value index cre, a kurtosis index kur, a skewness index ske, a pulse index imp and a margin index cle. The frequency domain feature values are calculated by taking a Fourier transform of the vibration signal and intercepting a set spectrum part, and include a frequency domain minimum value fmin, a maximum value fmax, an average value frms, a mean square root value fmean, a square root amplitude value froot, an average amplitude value fave, a waveform index fsha, a peak value index fcre, a kurtosis index fkur, a skewness index fske, a pulse index fimp and a margin index fcle.

[0008] Further, step S3 comprises: S31. The feature value matrix comprises combining each time domain feature value and frequency domain feature value into a 6x4 matrix B as follows: S32. Combine the n x m groups of feature values into a feature value matrix B, and add the start-stop state labels.

[0009] Further, step S4 specifically comprises: S41. Set the number of neurons in the at least three hidden layers in the judgment model as 8, 16 and 8 respectively, and the number of output layer neurons as 4; S42. Output a prediction result based on the pre-set state judgment rule of the output layer; ​​S43. The eigenvalue matrix constructed from the n x m set of eigenvalue matrices B and the corresponding start-stop state labels are divided into a training set and a test set, the judgment model is repeatedly trained, and the judgment model prediction result is output; The state judgment rule is: The output layer is R=[r1, r2, r3, r4], each item in R is between 0 and 1, and r1+r2+r3+r4=1, the index of the maximum value in R is taken, if r1 is the maximum, the prediction result is 0; if r2 is the maximum, the prediction result is 1; if r3 is the maximum, the prediction result is 2; if r4 is the maximum, the prediction result is 3; wherein 0 represents the stop state, 1 represents the start state, 2 represents the stop-to-start state, and 3 represents the start-to-stop state.

[0010] Further, the method further comprises: Comparing the prediction result of the judgment model with the start-stop state label of the eigenvalue matrix to obtain the accuracy of model training; When the accuracy is < the first threshold value, repeatedly collecting equipment vibration waveform data and forming a new eigenvalue matrix; The new eigenvalue matrix is incorporated into the eigenvalue matrix containing the start-stop state label, and step S43 is repeated until the accuracy is ≥ the first threshold value, and a trained judgment model is obtained.

[0011] Further, the trained judgment model is used for start-stop state judgment, and the method further comprises: converting real-time collected vibration waveform data of the equipment into an eigenvalue matrix, obtaining a judgment result based on the trained judgment model, and if a judgment error occurs, combining the vibration waveform data with the real label of the judgment error and the training set, and repeating step S43 to update the judgment model.

[0012] In a second aspect, the present application provides an industrial equipment start-stop state judgment system based on a vibration waveform feature matrix, which is applied to execute the judgment method as described above, and the system comprises: A data acquisition module is configured to acquire a vibration waveform data set under different start-stop states of the equipment, and add a start-stop state label, wherein the start-stop state includes a stop state, a start state, a stop-to-start process state, and a start-to-stop process state. A feature extraction module is configured to extract time domain eigenvalues and frequency domain eigenvalues of each vibration signal in the vibration waveform data set. A matrix construction module is configured to combine the time domain eigenvalues and the frequency domain eigenvalues of each vibration signal to form an eigenvalue matrix containing the start-stop state label. A model training module is used to build a judgment model based on a deep learning algorithm including an input layer, an output layer and at least three hidden layers, and train the judgment model using the eigenvalue matrix and the corresponding start-stop state marks; The judgment module is used to convert the vibration waveform data of the device to be tested into the eigenvalue matrix and input the converted data into the trained judgment model to output the current start and stop status of the device.

[0013] In a third aspect, the present invention provides a computer device, comprising: a processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the above-mentioned determination method.

[0014] In a fourth aspect, the present invention proposes a computer-readable storage medium. When the instructions in the computer-readable storage medium are executed by a processor of a computer device, the computer device can execute the judgment method as described above.

[0015] The beneficial effects of the present invention are: 1. The present invention realizes intelligent and accurate judgment of the start-up and shutdown status of industrial equipment by constructing a vibration waveform feature matrix and combining it with a deep learning model. Compared with the traditional judgment method that relies on a fixed threshold, the present invention can fully capture the multi-dimensional characteristics of the vibration signal in the time and frequency domain, and effectively solve the problem of vibration level fluctuations caused by differences in size, installation environment and working conditions of different equipment. Through the adaptive learning ability of the deep learning model, it can not only accurately identify steady states such as startup and shutdown, but also accurately capture the dynamic vibration characteristics of transition states such as shutdown to startup, startup to shutdown, etc., significantly improving the comprehensiveness and accuracy of status judgment. At the same time, the method supports online updating and optimization of the model, can continuously adapt to changes in the operating status of the equipment, and has long-term stability and reliability.

[0016] 2. Through intelligent state judgment and automatic invalid data elimination, a high-quality data foundation is provided for subsequent equipment fault diagnosis. Traditional methods fail to effectively eliminate downtime and transition state data, which often interferes with fault diagnosis. However, this invention uses high-precision state classification to automatically filter out invalid data, ensuring that the data used for fault diagnosis is valid equipment operating status data. This not only improves the accuracy of fault diagnosis but also reduces the false alarm rate, providing reliable technical support for predictive maintenance of industrial equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A flow chart of a method for determining the start-up and shutdown status of industrial equipment based on a vibration waveform feature matrix in one embodiment of the present invention; Figure 2A technical roadmap of the industrial equipment start-stop state judgment method based on a vibration waveform feature matrix in an embodiment of the present application is provided. Figure 3 A structure diagram of a judgment model in the industrial equipment start-stop state judgment method based on a vibration waveform feature matrix in an embodiment of the present application is provided. Figure 4 A system block diagram of the industrial equipment start-stop state judgment system based on a vibration waveform feature matrix in an embodiment of the present application is provided. Figure 5 A structure diagram of an electronic device provided in an embodiment of the present application is provided. DETAILED DESCRIPTION

[0018] It is necessary to point out here that the following detailed description is only used to further illustrate the present application and cannot be understood as limiting the protection scope of the present application, and the skilled in the art can make some non-essential improvements and adjustments to the present application according to the above application content.

[0019] As shown in Figure 1 and Figure 2 , in a specific embodiment, the present application provides an industrial equipment start-stop state judgment method based on a vibration waveform feature matrix, which includes the following steps: S1. A high-precision vibration acceleration sensor is installed at a key measurement point of the equipment, vibration waveform raw data under different start-stop states are collected, and the collected data are standardized and preprocessed, and a corresponding start-stop state label is labeled for each piece of data, including a stop state, a start state, and a stop-to-start process state and a start-to-stop process state reflecting the change of the equipment running state; S2. Time-frequency domain features are extracted from the preprocessed vibration signal, the time domain features include 12 statistical features such as minimum value, maximum value, and root mean square value, and the frequency domain features are obtained by cutting off the frequency band below 1000 Hz after Fourier transform of the signal; S3. The extracted time domain and frequency domain features are combined according to a preset matrix structure to construct a 6x4-dimensional feature value matrix, which can not only retain the time domain statistical characteristics of the vibration signal, but also reflect the frequency energy distribution characteristics; S4. A deep neural network model with three hidden layers (the number of neurons is 8, 16, and 8 respectively) is designed (in the present application, the judgment model is specifically constructed based on a deep learning algorithm, refer to Figure 3 ), the feature value matrix dataset is labeled, the model is trained, and the model parameters are optimized; S5. In actual application, the real-time collected equipment vibration signal is converted into a standard feature value matrix and input into the trained model, and the accurate running state of the current equipment can be output, and the system has online learning ability and can continuously optimize the model performance according to the newly collected data.

[0020] In a preferred embodiment, the vibration waveform data set of the collecting device in different start-stop states specifically includes: S11. Under actual working conditions in the industrial field, the original vibration waveform data set A of the collecting device in different start-stop states is collected by a three-axis vibration acceleration sensor, as follows: , ,, is the mth start-stop state, is the nth group of vibration signal waveforms in the mth start-stop state (unit: m / s²), and the sampling frequency is f (unit: Hz); where m has four different start-stop states (1 - stop state, 2 - start state, 3 - stop-to-start transition state, 4 - start-to-stop transition state). represents the number of vibration signal groups collected in each state, and each group of signals contains x, y, and z axial vibration data, with a sampling frequency of 10 kHz to ensure complete capture of high-frequency vibration characteristics during the start-stop process of the device.

[0021] S12. To ensure the balance and generalization ability of model training, the proportion of sample numbers in the four start-stop states in the vibration waveform data set A is specially set to 1:1:1:1, and no less than 250 groups of effective data are collected in each state. At the same time, different load conditions, environmental temperature, and other actual influencing factors are considered during data collection, making the data set have better engineering representativeness and practicality.

[0022] In a preferred embodiment, in step S2, the time domain characteristic values include the minimum value min, the maximum value max, the average value rms, the mean square root value mean, the square root amplitude value root, the average amplitude value ave, the waveform index sha, the peak value index cre, the kurtosis index kur, the skewness index ske, the pulse index imp, and the margin index cle. The frequency domain characteristic values are calculated by Fourier transform of the vibration signal and then intercepting a set spectrum part (such as the part below 1000 Hz), including the frequency domain minimum value fmin, the maximum value fmax, the average value frms, the mean square root value fmean, the square root amplitude value froot, the average amplitude value fave, the waveform index fsha, the peak value index fcre, the kurtosis index fkur, the skewness index fske, the pulse index fimp, and the margin index fcle. This joint extraction method of time and frequency domain characteristics not only retains the time domain dynamic characteristics of the vibration signal, but also fully excavates the frequency energy distribution information, providing comprehensive and complementary feature representation for subsequent state recognition.

[0023] It can be understood that, firstly, the application firstly constructs an engineering representative vibration waveform data set through multi-sensor cooperative collection, and adopts an equalization sampling strategy to ensure data quality; secondly, an innovative time-frequency domain multi-dimensional feature joint extraction method is proposed, and a feature matrix capable of comprehensively representing the equipment running state is constructed; finally, a deep neural network model is used to realize high-precision classification of four start-stop states. It is particularly worth pointing out that the scheme not only considers steady-state recognition, but also focuses on solving the transition state judgment problem that traditional methods are difficult to handle. The entire technical scheme fully considers the actual needs of the industrial field from data collection, feature extraction to model construction, and has significant engineering practical value.

[0024] In a preferred embodiment, step S3 comprises: S31. The feature value matrix comprises combining each time domain feature value with a frequency domain feature value into a 6x4 matrix B, as follows: S32. The matrix B combined with n x m groups of feature values is constructed into a feature value matrix, and a start-stop machine state marker is added.

[0025] In a preferred embodiment, in combination with Figure 3 As shown in the figure, step S4 specifically comprises: S41. The deep neural network model constructed adopts a three-hidden-layer architecture of "8-16-8", wherein: The first hidden layer (8 neurons) is responsible for primary feature extraction, and uses a ReLU activation function to process the input feature matrix; The second hidden layer (16 neurons) performs advanced feature abstraction, and uses a ReLU activation function with Dropout (0.2) to prevent overfitting; The third hidden layer (8 neurons) realizes feature compression, and uses a LeakyReLU (a=0.1) activation function; The output layer (4 neurons) uses a Softmax activation function to output the probability distribution of four states; S42. The output prediction result is output based on the preset state judgment rule of the output layer, and more specifically, the state judgment rule is: The output layer is R=[r1, r2, r3, r4], each item in R is between 0 and 1, and r1+r2+r3+r4=1, take the index of the maximum value in R, if r1 is the maximum, the prediction result is 0; if r2 is the maximum, the prediction result is 1; if r3 is the maximum, the prediction result is 2; if r4 is the maximum, the prediction result is 3; wherein, 0 represents the stop machine state, 1 represents the start machine state, 2 represents the stop-to-start machine state, and 3 represents the start-to-stop machine state.

[0026] S43. The eigenvalue matrix constructed from the n x m set of eigenvalue matrices B and the corresponding start-stop status labels are divided into a training set and a test set, the training and determination model is repeated, and the prediction result of the determination model is output.

[0027] In a preferred embodiment, the method further comprises: comparing and analyzing the prediction result of the determination model with the start-stop status label of the eigenvalue matrix to obtain the accuracy rate of model training; when the accuracy rate < the first threshold value, repeatedly collecting the equipment vibration waveform data and forming a newly added eigenvalue matrix; incorporating the newly added eigenvalue matrix into the eigenvalue matrix containing the start-stop status label, and repeating step S43 until the accuracy rate ≥ the first threshold value, to obtain the trained determination model.

[0028] For example, if the accuracy rate < 0.9, new data is repeatedly collected and merged with old data, and step S43 is repeated. After the accuracy rate C ≥ 0.9, the trained model is deployed.

[0029] More specifically, the method of the present application further comprises a model performance optimization and online learning mechanism, which is implemented as follows: first, a closed-loop optimization process is established during model training, the prediction result of the determination model is compared and analyzed with the start-stop status label of the eigenvalue matrix, and the comprehensive classification accuracy rate of the model on the test set is calculated; when the accuracy rate is lower than the first threshold value (typical value is 90%, wherein the steady state judgment requirement is ≥ 95%, and the transition state is ≥ 85%), the data enhancement process is automatically triggered, i.e. the vibration waveform data of the equipment under different working conditions is re-collected, the newly added eigenvalue matrix is formed after feature extraction, and the matrix is re-divided with the original training set according to the ratio of 7:2:1, and the model training process (step S43) is repeated until the accuracy rate continuously reaches or exceeds the threshold requirement.

[0030] In a preferred embodiment, the trained determination model is used to determine the start-stop status, and the method further comprises: converting the real-time collected vibration waveform data of the equipment into an eigenvalue matrix, obtaining the determination result based on the trained determination model, if an error occurs, combining the vibration waveform data with the real label of the determination error with the training set, and repeating step S43 to update the determination model.

[0031] More specifically, when the trained judgment model is used for online monitoring of the start-stop state of industrial equipment, the system performs the following intelligent closed-loop workflow: first, the vibration waveform data of the equipment is collected in real time by a distributed vibration acceleration sensor network (sampling rate ≥ 10 kHz), the original signal is segmented by using a sliding time window (window length 1 s, overlap rate 50%), and then the time-frequency domain features are extracted and a standardized 6x4 feature value matrix is constructed according to the above method; the matrix is input into the deployed judgment model, and the output layer generates a four-dimensional probability vector [a0, a1, a2, a3], which outputs the equipment state classification result. When the system detects an error in the classification result, the misclassified sample and its true label are stored in the initial data set with a timestamp; a new round of model fine-tuning is started every 50 new data sets, and the updated model must pass the accuracy check on the preserved validation set before it can replace the old model.

[0032] In a specific embodiment, in combination with Figure 4 The present application proposes an industrial equipment start-stop state judgment system based on a vibration waveform feature matrix, which is applied to execute the judgment method as described above, and the system comprises: A data acquisition module 10 is configured to collect vibration waveform data sets under different start-stop states of the equipment and add start-stop state labels, wherein the start-stop states include a stop state, a start state, a stop-to-start process state, and a start-to-stop process state.

[0033] A feature extraction module 11 is configured to extract time domain feature values and frequency domain feature values of each vibration signal in the vibration waveform data set.

[0034] A matrix construction module 12 is configured to combine the time domain feature values and the frequency domain feature values of each vibration signal to form a feature value matrix containing start-stop state labels.

[0035] A model training module 13 is configured to construct a judgment model based on a deep learning algorithm comprising an input layer, an output layer, and at least three hidden layers, and train the judgment model using the feature value matrix and the corresponding start-stop state labels.

[0036] A judgment module 14 is configured to input the vibration waveform data of the equipment to be tested into the trained judgment model after converting the vibration waveform data into a feature value matrix, and output the current start-stop state of the equipment.

[0037] A first updating module 15 is configured to compare and analyze the prediction results of the judgment model with the start-stop state labels of the feature value matrix to obtain the accuracy of the model training; when the accuracy is less than a first threshold, the vibration waveform data of the equipment is repeatedly collected to form new feature value matrices; the new feature value matrices are incorporated into the feature value matrix containing the start-stop state labels, and step S43 is repeated until the accuracy is greater than or equal to the first threshold, thereby obtaining the trained judgment model.

[0038] The second updating module 16 is configured to convert the vibration waveform data collected in real time by the equipment into a feature value matrix, obtain a judgment result based on the trained judgment model, and if a judgment error occurs, combine the vibration waveform data with a real label of the judgment error and the training set, repeat step S43, and update the judgment model.

[0039] The specific limitations of the industrial equipment start-stop state judgment system based on the vibration waveform feature matrix can refer to the limitations of the industrial equipment start-stop state judgment method based on the vibration waveform feature matrix described above, and will not be repeated here. It should be noted that the modules in the above judgment system correspond to steps S1 to S5 in the implementation of the above judgment method, and the instances and application scenarios implemented by the modules and the corresponding steps are the same, but are not limited to the content disclosed in embodiment 1.

[0040] In one embodiment, as shown in Figure 5 A computer device 21 is provided, which can be a terminal or a server. The computer device 21 includes a processor 23, a memory 24 and a network interface 25 connected by a system bus 22. The processor 23 of the computer device 21 is configured to provide computing and control capabilities. The memory of the computer device 24 includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The network interface 26 of the computer device is configured to communicate with external terminals through network connection. The computer program is executed by the processor to implement the steps of the judgment method in the above embodiments.

[0041] In one embodiment, a computer readable storage medium is provided. When the instructions in the computer readable storage medium are executed by the processor 23 of the computer device 21, the computer device 21 can execute the steps of the judgment method in any of the above embodiments.

[0042] The above embodiments only express several embodiments of the present application, and the description is more specific and detailed, but it cannot be understood as a limitation on the scope of the patent of the present application. It should be noted that for ordinary skilled persons in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are within the scope of protection of the present application.

Claims

1. A method for judging the start-up and shutdown status of industrial equipment based on a vibration waveform feature matrix, characterized in that: The method comprises: S1. Collect vibration waveform data sets under different start-stop states of the device and add start-stop state marks. The start-stop states include the stop state, the start state, the stop-to-start process state, and the start-to-stop process state; S2. Extract the time domain eigenvalues ​​and frequency domain eigenvalues ​​of each vibration signal in the vibration waveform dataset; S3. The time domain eigenvalues ​​and frequency domain eigenvalues ​​of each vibration signal are combined to form an eigenvalue matrix containing the start / stop state marker; S4. Construct a judgment model based on a deep learning algorithm including an input layer, an output layer, and at least three hidden layers, and train the judgment model using the eigenvalue matrix and the corresponding start / stop status mark; S5. Convert the vibration waveform data of the device to be tested into the eigenvalue matrix and input the matrix into the trained judgment model to output the current start / stop status of the device.

2. The method for judging the start-up and shutdown status of industrial equipment based on the vibration waveform characteristic matrix according to claim 1, characterized in that: The vibration waveform data set of the acquisition device under different start and stop states specifically includes: S11. Collect the vibration waveform data set A of the equipment under different start and stop states, as follows: , , is the mth start-stop state, The nth group of vibration signal waveforms in the mth start-stop state; S12. Set the ratio of the number of vibration waveforms in the four start-stop states in the vibration waveform data set A to 1:1:1:

1.

3. The method for judging the start-up and shutdown status of industrial equipment based on the vibration waveform characteristic matrix according to claim 1, characterized in that: In step S2, the time domain characteristic values ​​include the time domain minimum value min, maximum value max, average value rms, root mean square value mean, root square amplitude root, average amplitude ave, waveform index sha, peak index cre, kurtosis index kur, skewness index ske, impulse index imp and margin index cle; The frequency domain eigenvalues ​​are calculated by performing Fourier transform on the vibration signal and then intercepting a set spectrum portion, and include: frequency domain minimum value fmin, maximum value fmax, average value frms, root mean square value fmean, root square amplitude froot, average amplitude fave, waveform index fsha, peak index fcre, kurtosis index fkur, skewness index fske, pulse index fimp and margin index fcle.

4. The method for judging the start-up and shutdown status of industrial equipment based on the vibration waveform characteristic matrix according to claim 3 is characterized in that: Step S3 includes: S31. The eigenvalue matrix includes combining each time domain eigenvalue and frequency domain eigenvalue into a 6×4 matrix B, as shown in the following formula: S32. Construct the matrix B of n×m groups of eigenvalue combinations into an eigenvalue matrix, and add the start / stop state mark.

5. The method for judging the start-up and shutdown status of industrial equipment based on the vibration waveform characteristic matrix according to claim 3 is characterized in that: Step S4 specifically includes: S41. Set the number of neurons in at least three hidden layers of the judgment model to 8, 16, and 8 respectively, and the number of neurons in the output layer to 4; S42. Output prediction results based on the state judgment rules preset by the output layer; S43. Divide the eigenvalue matrix constructed by the n×m groups of eigenvalue matrices B and the corresponding start / stop state marks into a training set and a test set, repeatedly train the judgment model, and output the judgment model prediction result; The state judgment rule is: The output layer is R=[r1,r2,r3,r4], where each item in R is between 0 and 1, and r1+r2+r3+r4=1. The index of the maximum value in R is taken. If r1 is the largest, the prediction result is 0; if r2 is the largest, the prediction result is 1; if r3 is the largest, the prediction result is 2; if r4 is the largest, the prediction result is 3; among them, 0 represents the shutdown state, 1 represents the power-on state, 2 represents the shutdown to power-on state, and 3 represents the power-on to shutdown state.

6. The method for judging the start-up and shutdown status of industrial equipment based on the vibration waveform characteristic matrix according to claim 5, characterized in that: The method further comprises: Comparing and analyzing the prediction results of the judgment model with the start-stop state marks of the eigenvalue matrix to obtain the accuracy of the model training; When the accuracy is less than the first threshold, repeatedly collecting the device vibration waveform data and forming the newly added eigenvalue matrix; The newly added eigenvalue matrix is ​​incorporated into the eigenvalue matrix containing the start-stop state mark, and step S43 is repeated until the accuracy rate is greater than or equal to the first threshold, thereby obtaining a judgment model that has completed training.

7. The method for judging the start-up and shutdown status of industrial equipment based on the vibration waveform characteristic matrix according to claim 5, characterized in that: The method further includes: converting the vibration waveform data collected by the device in real time into an eigenvalue matrix, and obtaining a judgment result based on the judgment model that has been trained. If a judgment error occurs, the vibration waveform data with the true label and the training set that have been judged incorrectly are merged, and step S43 is repeated to update the judgment model.

8. A system for judging the start and stop status of industrial equipment based on a vibration waveform feature matrix, applied to execute the judgment method according to any one of claims 1 to 7, characterized in that: The system comprises: The data acquisition module is used to collect vibration waveform data sets under different start and stop states of the equipment and add start and stop state marks. The start and stop states include stop state, start state, stop-to-start process state, and start-to-stop process state; A feature extraction module is used to extract the time domain eigenvalue and frequency domain eigenvalue of each vibration signal in the vibration waveform data set; A matrix construction module, configured to combine the time domain eigenvalues ​​and frequency domain eigenvalues ​​of each vibration signal to form an eigenvalue matrix containing the start / stop state mark; A model training module is used to build a judgment model based on a deep learning algorithm including an input layer, an output layer and at least three hidden layers, and train the judgment model using the eigenvalue matrix and the corresponding start-stop state marks; The judgment module is used to convert the vibration waveform data of the device to be tested into the eigenvalue matrix and input the converted data into the trained judgment model to output the current start and stop status of the device.

9. A computer device, characterized in that: include: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the determination method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that When the instructions in the computer-readable storage medium are executed by a processor of a computer device, the computer device is enabled to execute the judgment method according to any one of claims 1 to 7.

Images