Industrial equipment operation state analysis method and system based on machine learning

Abstract

The invention discloses an industrial equipment operation state analysis method and system based on machine learning, and relates to the technical field of equipment detection. The method comprises the following steps: obtaining a plurality of types of detection items and abnormal behaviors, respectively combining the detection items and the abnormal behaviors, respectively inputting corresponding data in historical detection records and historical abnormal records into different deep networks for training, and screening out a plurality of detection items sensitive to each abnormal behavior; obtaining the type of a to-be-monitored abnormal behavior of the target; acquiring parameters of the target to-be-monitored detection item; and according to the type of the target to-be-monitored abnormal behavior and the plurality of detection items sensitive to each abnormal behavior, identifying and analyzing parameters of the target to-be-monitored detection items to obtain state parameters of the target to-be-monitored abnormal behavior of the equipment. According to the method, the data dimension of the detection item in the analysis process is reduced, so that the equipment abnormal behavior analysis speed is increased.

Description

Technical Field

[0001] The present invention belongs to the technical field of equipment detection, and in particular relates to an industrial equipment operating status analysis method and system based on machine learning. Background Art

[0002] With the rapid development of industrial automation and intelligence, the operation and management of industrial equipment has become particularly important. Monitoring and analyzing the operating status of equipment is of great significance to ensure production efficiency and reduce maintenance costs. Traditional equipment operation monitoring methods mainly rely on regular physical inspections and scheduled maintenance procedures, which are often time-consuming and labor-intensive, and cannot reflect the immediate status of the equipment in real time, which limits the ability to prevent faults and respond quickly. Summary of the invention

[0003] The purpose of the present invention is to provide an industrial equipment operation status analysis method and system based on machine learning, which improves the speed of equipment abnormal behavior analysis by modeling and analyzing equipment detection items and abnormal behaviors.

[0004] In order to solve the above technical problems, the present invention is achieved through the following technical solutions: The present invention provides an industrial equipment operation status analysis method based on machine learning, which is characterized by comprising: Obtaining historical test records including test parameters of multiple types of test items of the equipment in a working state at multiple collection times; Obtaining historical anomaly records including state parameters of multiple types of abnormal behaviors of the device in a working state at multiple collection moments; Combining the detection items and abnormal behaviors of multiple types respectively, and inputting the corresponding data in the historical detection records and the historical abnormal records into different deep networks for training respectively, and screening out several detection items that are sensitive to each abnormal behavior; Obtain the type of abnormal behavior of the target to be monitored; Obtaining parameters of the target detection item to be monitored; According to the type of the target abnormal behavior to be monitored and the number of detection items to which each abnormal behavior is sensitive, the state parameters of the target abnormal behavior to be monitored of the device are obtained by identifying and analyzing the parameters of the detection items to be monitored.

[0005] The present invention also discloses a method for analyzing the operating status of industrial equipment based on machine learning, comprising: Enter the type of abnormal behavior to be monitored; Send the type of abnormal behavior to be monitored by the target; Receive and display the status parameters of the target abnormal behavior to be monitored by the device.

[0006] The present invention also discloses an industrial equipment operation status analysis system based on machine learning, comprising: A memory, used for storing historical detection records and historical abnormality records; Multiple types of sensors are used to collect detection parameters of multiple types of detection items; An analysis unit, used to obtain historical detection records including detection parameters of multiple types of detection items of the device in a working state at multiple collection moments; Obtaining historical anomaly records including state parameters of multiple types of abnormal behaviors of the device in a working state at multiple collection moments; Combining the detection items and abnormal behaviors of multiple types respectively, and inputting the corresponding data in the historical detection records and the historical abnormal records into different deep networks for training respectively, and screening out several detection items that are sensitive to each abnormal behavior; Obtain the type of abnormal behavior of the target to be monitored; Obtaining parameters of the target detection item to be monitored; According to the type of abnormal behavior to be monitored and the number of detection items to which each abnormal behavior is sensitive, the state parameters of the abnormal behavior to be monitored of the device are obtained by identifying and analyzing the parameters of the detection items to be monitored; and A user terminal is used to input the type of abnormal behavior to be monitored; Send the type of abnormal behavior to be monitored by the target; Receive and display the status parameters of the target abnormal behavior to be monitored by the device.

[0007] The present invention analyzes the detection parameters of different types of detection items and the state parameters of abnormal behaviors through an analysis unit, thereby screening out detection items that are sensitive to each abnormal behavior, reducing the data dimension of equipment operation status analysis, and being able to quickly analyze various abnormal behaviors including equipment failures by collecting detection parameters of some detection items.

[0008] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0010] Figure 1A schematic diagram of functional units and information interaction of an industrial equipment operation status analysis system based on machine learning in one embodiment of the present invention; Figure 2 A schematic diagram of a process flow of an analysis unit according to an embodiment of the present invention; Figure 3 A schematic diagram of a process flow of a user terminal according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the step flow of step S3 in one embodiment of the present invention; Figure 5 This is a schematic diagram of the step flow of step S31 in one embodiment of the present invention; Figure 6 This is a schematic diagram of a step flow chart of step S313 in an embodiment of the present invention; Figure 7 This is a schematic diagram of the step flow of step S33 in one embodiment of the present invention; Figure 8 A schematic diagram of a step flow chart of step S34 in an embodiment of the present invention; Fig. 9 The step S6 of the present invention is a schematic diagram of a step flow in an embodiment. Figure 1 ; Fig.10 The step S6 of the present invention is a schematic diagram of a step flow in an embodiment. Figure 2 ; In the accompanying drawings, the components represented by the reference numerals are listed as follows: 1- memory, 2- sensor, 3- analysis unit, 4- user terminal. DETAILED DESCRIPTION

[0011] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0012] It should be noted that the terms "first", "second", etc. in this application are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0013] In the process of monitoring and diagnosing industrial equipment, many types of detection items are collected, but not all detection items are necessary for the abnormal behavior that needs to be monitored. Unnecessary detection items will slow down the speed of equipment operation status analysis. In order to improve the analysis speed, the present invention provides the following solution.

[0014] See also Figures 1 to 3 As shown, the present invention provides an industrial equipment operation status analysis system based on machine learning, which can include a memory 1, a sensor 2, an analysis unit 3 and a user terminal 4 in terms of functional units.

[0015] The memory 1 stores sample data for training the deep network. The sensor 2 can collect real-time detection data of the device, and collect parameters such as vibration, temperature, current, voltage, pressure, etc. of the device through the sensor.

[0016] The analysis unit 3 can establish a deep network for device operation analysis and determine the abnormal state of the device in real time. The user terminal 4 can allow the user to input the type of abnormal behavior to be monitored and receive and display the analysis result calculated by the analysis unit 3.

[0017] In the specific implementation process, the memory 1 stores historical detection records and historical abnormality records. The types of sensors 2 can be increased or decreased according to demand, and step S021 can be executed to collect detection parameters of multiple types of detection items.

[0018] Since there are many detection items for the equipment, the analysis unit 3 needs to screen out detection items with high correlation for each type of abnormal behavior before conducting formal equipment operation analysis, that is, sensitive detection items in subsequent schemes. Specifically, steps S1 and S2 can be first executed to obtain historical detection records containing detection parameters of multiple types of detection items of the equipment in the working state at multiple acquisition moments, and historical abnormal records containing state parameters of multiple types of abnormal behaviors of the equipment in the working state at multiple acquisition moments. Next, step S3 can be executed to combine multiple types of detection items and abnormal behaviors respectively, and input the corresponding data in the historical detection records and historical abnormal records into different deep networks for training, and screen out several detection items that are sensitive to each abnormal behavior. The algorithm extracts key features through feature selection methods to reduce the dimension of the data set. Compared with traditional dimensionality reduction schemes, this scheme directly analyzes the practical significance of different detection items, more effectively realizes data dimensionality reduction, and improves the efficiency and accuracy of the algorithm.

[0019] The deep network used in this solution is usually called Deep Neural Networks (DNNs), which is an artificial neural network characterized by a multi-layer network structure that can achieve complex feature abstraction and data representation. This network contains multiple hidden layers, each of which is composed of multiple neurons, and the neurons are connected by weights.

[0020] After the staff uses the user terminal 4 to execute step S041 to input the type of abnormal behavior to be monitored, and executes step S042 to send the type of abnormal behavior to be monitored, the analysis unit 3 continues to execute step S4 to obtain the type of abnormal behavior to be monitored, and then step S5 can be executed to obtain the parameters of the detection items to be monitored. Next, step S6 can be executed to identify and analyze the parameters of the detection items to be monitored according to the type of abnormal behavior to be monitored and the number of detection items sensitive to each abnormal behavior, and obtain the state parameters of the abnormal behavior to be monitored by the device.

[0021] Finally, the user terminal 4 executes step S043 to receive and display the state parameters of the target abnormal behavior to be monitored of the device. Of course, the state parameters of the target abnormal behavior to be monitored of the device can also be displayed by other display devices. In this case, the display device and the hardware operated by the staff together constitute the functional unit of the user terminal 4 in this solution.

[0022] This solution can also convert the raw data of detection items and abnormal behaviors into more meaningful features, enhance the algorithm's ability to understand the equipment status, and thus realize equipment operation status monitoring under conditions of high dimensionality and high complexity of data.

[0023] See also Figure 4 As shown, during the deep network training process, the greater the correlation between the input layer and the output layer, the higher the effect of the iterative training will be. Based on this principle, each detection item that is sensitive to abnormal behavior can be screened out. Specifically, step S31 can be first executed to combine multiple types of detection items and then establish an input-output relationship between each group of detection items and each abnormal behavior. Next, step S32 can be executed to use the combination of detection items in each pair of input-output relationships as the input layer, and the corresponding abnormal behavior as the output layer, to train multiple deep networks. Next, step S33 can be executed to continuously obtain the loss function value in each deep network training process, and calculate the degree of correlation between the input layer and the output layer of the deep network.

[0024] During the deep learning training process, the behavior and monitoring of the loss function value is crucial. At the beginning of training, the loss value is usually high because the model parameters are randomly initialized. As the training progresses, the loss value should gradually decrease, indicating that the model is learning the patterns in the data. In the later stages of training, the loss value tends to stabilize and remain at a low level, indicating that the model parameters have reached a relatively stable state. Finally, step S34 can be executed to obtain several detection items sensitive to each abnormal behavior based on the degree of correlation between the input layer and the output layer of each deep network.

[0025] Specifically, see Figure 5 As shown, in the process of combing the input layer and output layer of each deep network, step S311 can be first performed to combine multiple types of detection items to obtain multiple groups of detection item combinations. Next, step S312 can be performed to form multiple pairs of input-output relationships between each group of detection item combinations and each abnormal behavior.

[0026] In the above process, especially in step S31, due to the large number of combinations of detection items, if each combination of detection items and the deep network of each abnormal behavior are trained, the amount of calculation is too large, so pre-screening is required. Therefore, step S313 can be executed at the end to filter out some input-output relationships based on historical detection records and historical abnormal records. Specifically, for each input-output relationship, please refer to Figure 6 As shown, first, step S3131 can be executed to determine whether the state parameters of abnormal behavior at each acquisition moment within the set time period remain unchanged. If so, it means that the time period has no training significance, so next step S3132 can be executed to change the set time period and continue to determine whether the state parameters of abnormal behavior at each acquisition moment within the set time period remain unchanged, if not, then step S3133 can be executed to retain the set time period. Next, step S3134 can be executed to determine whether there is any type of detection item in the combination of each group of detection items within the set time period. The detection parameters remain unchanged at each acquisition moment within the set time period. If so, it means that the input-output relationship contains data that has no training sample significance, so next step S3135 can be executed that the input-output relationship should be screened out, if not, next step S3136 can be executed to retain the input-output relationship.

[0027] See also Figure 7As shown, since the training convergence effects of deep networks with different correlations between the input layer and the output layer are different, and at the same time, the training convergence effects of deep networks with different complexities are also different, in order to try to eliminate the interference caused by the complexity of the deep network on the judgment of the degree of correlation, step S331 can be first executed for each deep network, and the same preliminary set number of times are trained, and the difference between the initial training and the last training loss function values ​​of each deep network is obtained as the initial convergence degree. Next, step S332 can be executed to eliminate the deep networks whose corresponding initial convergence degrees are less than the mean. Next, step S333 can be executed to continue training all the remaining deep networks after elimination for the same subsequent set number of times, and the difference between the initial training of each deep network and the last training loss function value of the subsequent set number of times is obtained as the overall convergence degree. Finally, step S334 can be executed to use the inverse of the overall convergence degree of each deep network as the degree of correlation between the input layer and the output layer of the deep network.

[0028] In order to supplement the implementation process of the above steps S331 to S334, the source code of some functional modules is provided, and the explanation is compared in the comment section. In order to avoid the leakage of data involving commercial secrets, some data that does not affect the implementation of the solution are desensitized, the same below.

[0029] #include <iostream> #include <vector> #include <numeric> #include <algorithm> #include <cmath> using namespace std; class DeepLearningNetwork { public: double initialLoss; double finalLoss; / / Constructor, initial and final losses are randomly generated to simulate different network behaviors DeepLearningNetwork() { initialLoss = rand() % 1000 / 100.0 + 10; / / Initial loss, between 10 and 20 finalLoss = initialLoss - (rand() % 1000 / 200.0); / / Final loss, smaller than initial loss } / / Simulation training process void train(int epochs) { double decrementPerEpoch = (initialLoss - finalLoss) / epochs; initialLoss = finalLoss; finalLoss = finalLoss - epochs * decrementPerEpoch / 2; / / Example: Subsequent training loss reduction rate is halved } / / Get the loss function value difference of the current network double getLossDifference() const { return initialLoss - finalLoss; } }; int main() { int initialEpochs = 10; / / Preliminary setting of training times int additionalEpochs = 5; / / Subsequent training times vector <deeplearningnetwork>networks(10); / / Initialize 10 networks / / Initial training of all networks for (auto&net : networks) { net.train(initialEpochs); } / / Calculate the average initial convergence degree double averageInitialConvergence = accumulate( networks.begin(), networks.end(), 0.0, [](double sum, const DeepLearningNetwork&net) { return sum + net.getLossDifference(); }) / networks.size(); / / Eliminate networks whose initial convergence degree is less than the mean networks.erase( remove_if(networks.begin(), networks.end(), [averageInitialConvergence](const DeepLearningNetwork&net) { return net.getLossDifference() <averageInitialConvergence; }), networks.end() ); / / Continue training the remaining networks for (auto&net : networks) { net.train(additionalEpochs); } / / Calculate and output the overall convergence and correlation of each network for (const auto&net : networks) { double totalConvergence = net.getLossDifference(); double associationDegree = 1.0 / totalConvergence; / / The degree of association is calculated as the inverse of the degree of convergence cout<<"Network Convergence: "< <totalConvergence<<", AssociationDegree: "<<associationDegree<<endl; } return 0; } This program first creates a series of deep learning networks, randomly generates initial and final loss values ​​to simulate different training effects. Each network is first trained a certain number of times, and then the degree of convergence is calculated based on the degree of decrease in the loss function value. Networks whose loss reduction is less than the average reduction will be eliminated. The remaining networks continue to train, and then the degree of association of each network is calculated based on the final loss reduction. Through such simulations, the effects and associations of different network structures on specific tasks can be evaluated, helping to optimize and select the most suitable network configuration.

[0030] See also Figure 8 As shown, in order to select the deep network with high correlation between the input layer and the output layer, so as to obtain several detection items sensitive to each abnormal behavior, it is necessary to compare and analyze the correlation degree of each deep network. Specifically, first, step S341 can be executed to arrange the correlation degree of the input layer and the output layer of each deep network according to the numerical value to obtain a correlation value list. Next, step S342 can be executed to take the maximum value in the correlation value list as the most sensitive correlation degree. Next, step S343 can be executed to calculate the mean of the difference between each correlation degree and other adjacent correlation degrees in the correlation value list as the scale value of the correlation value list. Next, step S344 can be executed to start from the most sensitive correlation degree in the correlation value list, and calculate in sequence from large to small whether the difference between two adjacent correlation degrees is less than the scale value. If so, it means that all deep networks that meet the high correlation requirements have not yet been included, so step S344 can be executed to continue the calculation. If not, it means that all deep networks that meet the high correlation requirements have been included, so step S345 can be executed next to use all deep networks corresponding to the correlation degrees in the correlation value list that participate in the sequential calculation as applicable deep networks. Finally, step S346 can be executed to use the output layer and input layer of the applicable deep network as abnormal behavior and several detection items sensitive to abnormal behavior, respectively.

[0031] In order to provide additional explanation for the implementation process of the above-mentioned steps S341 to S346, source code of some functional modules is provided, and comparative explanations are provided in the comment section.

[0032] #include <iostream> #include <vector> #include <algorithm> #include <numeric> using namespace std; / / Example deep network structure class DeepLearningNetwork { public: vector <double>inputLayer; / / Input layer data (the example is the data of the detection project) string outputBehavior; / / Output layer data (example is abnormal behavior) double associationDegree; / / The degree of association between the input layer and the output layer / / Constructor DeepLearningNetwork(vector <double>inputs, string output, double degree) : inputLayer(inputs), outputBehavior(output), associationDegree(degree) {} }; / / Main function int main() { / / Example network list vector <deeplearningnetwork>networks = { { {1.0, 2.0}, "Overheat", 0.9}, { {1.5, 2.5}, "Leakage", 0.85}, { {1.1, 2.1}, "Overheat", 0.95}, { {1.6, 2.6}, "Fracture", 0.8} }; / / Sort the networks in descending order of relevance sort(networks.begin(), networks.end(), [](const DeepLearningNetwork&a, const DeepLearningNetwork&b) { return a.associationDegree>b.associationDegree; }); / / Find the most sensitive correlation double maxSensitivity = networks.front().associationDegree; / / Calculate the mean of the correlation difference vector <double>differences; for (size_t i = 1; i <networks.size(); ++i) { differences.push_back(networks[i - 1].associationDegree - networks[i].associationDegree); } double avgDifference = accumulate(differences.begin(),differences.end(), 0.0) / differences.size(); / / Find the correlation breakpoint below the scale value vector <deeplearningnetwork>applicableNetworks; for (size_t i = 0; i < networks.size() - 1; ++i) { if (networks[i].associationDegree - networks[i + 1].associationDegree < avgDifference) { applicableNetworks.push_back(networks[i]); } else { applicableNetworks.push_back(networks[i]); break; } } / / Output applicable deep networks cout << "Applicable Deep Networks for Sensing Specific Behaviors:" << endl; for (const auto& net : applicableNetworks) { cout << "Behavior: " << net.outputBehavior << ", Inputs: { "; for (auto input : net.inputLayer) { cout << input << " "; } cout << "}, Association Degree: " << net.associationDegree << endl; } return 0; } This code first defines an example deep network class, including input layer, output layer and correlation degree. The program initializes a series of deep network instances and sorts them by correlation degree. By comparing the difference between correlation degrees and the average difference, the program can identify the most sensitive networks. These networks are considered to be particularly sensitive to specific abnormal behaviors because they show a small decrease in correlation degree. Finally, these networks are output as a list of detection items that are sensitive to the corresponding abnormal behaviors. This program simulates how to select the most suitable network configuration by analyzing the correlation degree of the network to optimize the detection ability of specific abnormal behaviors.

[0033] See also Fig. 9 As shown, in actual use, there may not be enough time to accurately analyze the operating status of the device, and there is no need to continuously analyze the operating status of the device. At this time, step S611 can be executed to obtain the deep network corresponding to the detection items of each target to be detected that has been trained to convergence. Then step S612 can be executed to input the parameters of several detection items sensitive to the abnormal behavior of each target to be monitored into the corresponding deep network to obtain the state parameters of the abnormal behavior of each target to be monitored. That is, the operating status analysis results of the output device are quickly responded to by directly calling the existing deep network.

[0034] See also Fig.10 As shown, in the process of formally analyzing the operating status of the device, a new target deep network can be directly constructed. Specifically, step S621 can be first executed to merge several detection items that are sensitive to abnormal behaviors of each type of target to be monitored to obtain the type of detection item of the target to be detected. Next, step S622 can be executed to use the detection item of the target to be detected as the input layer, the abnormal behavior of the target to be monitored as the output layer, and the corresponding data in the historical detection records and historical abnormal records are respectively input into the same deep network for training until convergence to obtain the target deep network. Finally, step S623 can be executed to input the parameters of the detection item of the target to be monitored into the corresponding target deep network to obtain the state parameters of the abnormal behavior of the target to be monitored of the device. Compared with the previous solution, although this solution requires time and computing power to train the target deep network, once the target deep network training converges, the analysis results of the device can be output at high speed, which is suitable for long-term continuous operating status monitoring of the device.

[0035] The flow chart and block diagram in the accompanying drawings show the possible architecture, function and operation of the device, system, method and computer program product according to multiple embodiments of the present application. In this regard, each square frame in the flow chart or block diagram can represent a part of a module, program segment or instruction, and a part of the module, program segment or instruction includes one or more executable instructions for realizing the logical function of the specification. In some alternative implementations, the functions marked in the square frame can also occur in a sequence different from that marked in the accompanying drawings. For example, two continuous square frames can actually be executed substantially in parallel, and they can also be executed in reverse order sometimes, depending on the functions involved.

[0036] It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart, can be implemented by hardware that performs the corresponding function or action, such as a circuit or ASIC (Application Specific Integrated Circuit), or can be implemented by a combination of hardware and software, such as firmware.

[0037] Although the present invention is described herein in conjunction with various embodiments, in the process of implementing the claimed invention, those skilled in the art may understand and implement other variations of the disclosed embodiments by viewing the drawings, the disclosure, and the appended claims. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude multiple situations. A single processor or other unit may implement several functions listed in the claims. Certain measures are recorded in different dependent claims, but this does not mean that these measures cannot be combined to produce good results.

[0038] The embodiments of the present application have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes are obvious to those of ordinary skill in the art without departing from the scope of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.< / deeplearningnetwork> < / double> < / deeplearningnetwork> < / double> < / double> < / numeric> < / algorithm> < / vector> < / iostream> < / deeplearningnetwork> < / cmath> < / algorithm> < / numeric> < / vector> < / iostream>

Claims

1. A method for analyzing the operating status of industrial equipment based on machine learning, characterized in that: include, Obtaining historical test records including test parameters of multiple types of test items of the equipment in a working state at multiple collection times; Obtaining historical anomaly records including state parameters of multiple types of abnormal behaviors of the device in a working state at multiple collection moments; Combining the detection items and abnormal behaviors of multiple types respectively, and inputting the corresponding data in the historical detection records and the historical abnormal records into different deep networks for training respectively, and screening out several detection items that are sensitive to each abnormal behavior; Obtain the type of abnormal behavior of the target to be monitored; Obtaining parameters of the target detection item to be monitored; According to the type of the target abnormal behavior to be monitored and the number of detection items to which each abnormal behavior is sensitive, the state parameters of the target abnormal behavior to be monitored of the device are obtained by identifying and analyzing the parameters of the detection items to be monitored.

2. The method according to claim 1, characterized in that The steps of combining the detection items and abnormal behaviors of multiple types respectively, and inputting the corresponding data in the historical detection records and the historical abnormal records into different deep networks for training respectively, and screening out several detection items that are sensitive to each abnormal behavior, include: After combining the detection items of multiple types, respectively establish an input-output relationship between each combination of the detection items and each abnormal behavior to obtain an input-output relationship between each combination of the detection items and each abnormal behavior; Taking each combination of the detection items in each pair of the input-output relationship as an input layer, and taking the corresponding abnormal behavior as an output layer, training multiple deep networks; Continuously obtaining the loss function value during each deep network training process, and calculating the correlation degree between the input layer and the output layer of the deep network; According to the degree of association between the input layer and the output layer of each of the deep networks, a plurality of detection items sensitive to each of the abnormal behaviors are obtained.

3. The method according to claim 2, characterized in that The step of combining the detection items of multiple types and establishing an input-output relationship between each combination of the detection items and each abnormal behavior respectively includes: Combining multiple types of the detection items to obtain multiple groups of the detection item combinations; Combining each group of the detection items with each abnormal behavior to form a plurality of pairs of input-output relationships; Part of the input-output relationship is filtered out according to the historical detection records and the historical abnormal records.

4. The method according to claim 3, characterized in that The step of filtering out part of the input-output relationship according to the historical detection records and the historical abnormal records, include, For each of the input-output relationships, Determine whether the state parameters of the abnormal behavior at each collection moment within the set time period remain unchanged, If yes, then after changing the set time period, continue to determine whether the state parameters of the abnormal behavior at each collection moment within the set time period remain unchanged. If not, the set time period is retained. Determine whether there is any type of the detection item in each group of the detection items within the set time period, and the detection parameters of the detection items remain unchanged at each collection moment within the set time period, If so, the input-output relationship should be screened out. If not, the input-output relationship is retained.

5. The method according to claim 2, characterized in that: The step of continuously obtaining the loss function value in each deep network training process and calculating the degree of association between the input layer and the output layer of the deep network, include, For each of the deep networks, the same initially set number of trainings are performed, and the difference between the initial training loss function value and the final training loss function value of each of the deep networks is obtained as the initial convergence degree; Eliminate the corresponding deep network whose initial convergence degree is less than the mean value; Continue training all the remaining deep networks after the elimination for the same subsequent set number of times, and obtain the difference in loss function value of each deep network from the initial training of the initial set number of times to the final training of the subsequent set number of times as the overall convergence degree; The inverse of the overall convergence degree of each of the deep networks is used as the degree of association between the input layer and the output layer of the deep network.

6. The method according to claim 2, characterized in that The step of obtaining a plurality of detection items sensitive to each abnormal behavior according to the degree of association between the input layer and the output layer of each of the deep networks, include, Arranging the correlation degree between the input layer and the output layer of each of the deep networks according to the numerical value to obtain a correlation numerical value list; Taking the maximum value in the association value list as the most sensitive association degree; Calculate and obtain the average of the differences between each of the correlation degrees in the correlation value list and other adjacent correlation degrees as the scale value of the correlation value list; In the correlation value list, starting from the most sensitive correlation degree, sequentially calculating from large to small whether the difference between two adjacent correlation degrees is less than the scale value; If yes, continue the calculation; If not, the deep networks corresponding to all the correlation degrees in the correlation value list that are sequentially calculated are used as the applicable deep networks; The output layer and the input layer of the applicable deep network are respectively used as the abnormal behavior and a plurality of the detection items that are sensitive to the abnormal behavior.

7. The method according to any one of claims 1 to 6, characterized in that: The step of obtaining the state parameters of the abnormal behavior of the target to be monitored by the device by analyzing the parameters of the detection items to be monitored according to the type of the abnormal behavior to be monitored and the number of detection items to which each abnormal behavior is sensitive, include, Obtaining the deep network corresponding to each detection item of the target to be detected that has been trained to convergence; The parameters of several detection items that are sensitive to the abnormal behavior of each target to be monitored are input into the corresponding deep network to obtain the state parameters of the abnormal behavior of each target to be monitored.

8. The method according to claim 1, characterized in that The step of obtaining the state parameters of the abnormal behavior of the target to be monitored by analyzing the parameters of the detection items to be monitored according to the type of the abnormal behavior to be monitored and the number of detection items to which each abnormal behavior is sensitive, further includes: Combining a plurality of detection items that are sensitive to abnormal behaviors of each type of target to be monitored to obtain the type of detection item of the target to be detected; The target detection item to be detected is used as the input layer, the target abnormal behavior to be monitored is used as the output layer, and the corresponding data in the historical detection record and the historical abnormal record are respectively input into the same deep network for training until convergence, so as to obtain the target deep network; The parameters of the target detection items to be monitored are input into the corresponding target deep network to obtain the state parameters of the target abnormal behavior of the device to be monitored.

9. A method for analyzing the operating status of industrial equipment based on machine learning, characterized in that: include, Enter the type of abnormal behavior to be monitored; Send the type of abnormal behavior to be monitored by the target; Receive and display the state parameters of the target abnormal behavior to be monitored of the equipment in the industrial equipment operation status analysis method based on machine learning as described in any one of claims 1 to 8.

10. An industrial equipment operation status analysis system based on machine learning, characterized in that: include, A memory, used for storing historical detection records and historical abnormality records; Multiple types of sensors are used to collect detection parameters of multiple types of detection items; An analysis unit, used to obtain historical detection records including detection parameters of multiple types of detection items of the device in a working state at multiple collection moments; Obtaining historical anomaly records including state parameters of multiple types of abnormal behaviors of the device in a working state at multiple collection moments; Combining the detection items and abnormal behaviors of multiple types respectively, and inputting the corresponding data in the historical detection records and the historical abnormal records into different deep networks for training respectively, and screening out several detection items that are sensitive to each abnormal behavior; Obtain the type of abnormal behavior of the target to be monitored; Obtaining parameters of the target detection item to be monitored; According to the type of the target abnormal behavior to be monitored and the number of detection items to which each abnormal behavior is sensitive, the state parameters of the target abnormal behavior to be monitored of the device are obtained by parameter identification and analysis of the detection items to be monitored; as well as, A user terminal is used to input the type of abnormal behavior to be monitored; Send the type of abnormal behavior to be monitored by the target; Receive and display the status parameters of the target abnormal behavior to be monitored by the device.

Images