Predictive maintenance method and system for industrial equipment

Abstract

The invention relates to the field of equipment operation state monitoring, in particular to an industrial equipment predictive maintenance method and system.The method comprises the steps that historical equipment parameters of industrial equipment operation are processed and labeled, and labels comprise normal operation, to-be-maintained and faults; calculating the delay time between the label and the historical operation parameters, and calculating the importance of the equipment parameters to the equipment operation label under each delay time; inputting real-time equipment parameters of the industrial equipment into a preset classification model, and outputting predicted values of the real-time operation parameters corresponding to the label probabilities; calculating weight coefficients according to importance, wherein each type of labels corresponds to one weight coefficient; and taking the product of the prediction value and the corresponding weight coefficient as a prediction result, and generating maintenance information according to the prediction result of each tag. The method has the effect of carrying out predictive maintenance according to the actual operation condition of the equipment.

Description

Technical Field

[0001] This application relates to the field of equipment operation status monitoring, and particularly to a predictive maintenance method and system for industrial equipment. Background Art

[0002] With the continuous acceleration of the industrialization process, modern manufacturing industries have an increasingly high dependence on production equipment. Equipment failures or downtimes have a huge impact on production efficiency and enterprise operations, and may even lead to the shutdown of production lines, causing huge economic losses. Therefore, how to improve the reliability of equipment, reduce the frequency of failures and downtime has become a crucial issue in industrial production.

[0003] Traditional equipment maintenance methods include corrective maintenance and preventive maintenance. Corrective maintenance is to repair the equipment when a failure occurs. Although this method has low costs, it may lead to long equipment downtime and even affect the overall production schedule. Preventive maintenance is a strategy of performing regular inspections and maintenance at fixed time intervals. For example, a method, system, equipment and medium for power quality monitoring in an industrial park disclosed in a Chinese patent application document with publication number CN114966269A includes: deploying Internet of Things acquisition terminals and Internet of Things gateways for power quality monitoring in the distribution rooms and large power-consuming equipment of different users in the industrial park, and then self-organizing these Internet of Things terminal devices through the gateway, and periodically receiving grid data and environmental data of multiple users in the industrial park collected by the Internet of Things acquisition terminals through the Internet of Things network, realizing the monitoring of the power quality of multiple users in the industrial park.

[0004] The above-mentioned existing technology is a strategy of performing regular inspections and maintenance according to fixed time intervals. Although it can reduce the occurrence of failures, since the inspections are not carried out according to the actual operating conditions of the equipment, there are problems of over-maintenance and resource waste. Summary of the Invention

[0005] In order to solve the above technical problem of how to perform predictive maintenance according to the actual operating conditions of the equipment, this application provides a predictive maintenance method and system for industrial equipment.

[0006] In the first aspect, this application provides a predictive maintenance method for industrial equipment, adopting the following technical solution: A predictive maintenance method for industrial equipment includes the steps of: processing and tagging the historical equipment parameters of the operation of industrial equipment, where the tags include normal operation, to be repaired, and failure; calculating the delay time between the tags and the historical operation parameters, and calculating the importance of the equipment parameters to the equipment operation tags at each delay time; where the formula for importance is: ; In the formula, represents at a delay time of The importance of device parameters for the type of label, indicating the degree of confusion between the type of label and device parameters at a delay time of ; indicating the similarity between the type of label and the distribution of device parameters of other labels at a delay time of ; Input the real-time device parameters of industrial equipment into a preset classification model, and output the predicted values of the probabilities of the real-time operation parameters corresponding to each label; calculate the weight coefficient according to the importance, and each type of label corresponds to a weight coefficient; take the product of the predicted value and the corresponding weight coefficient as the prediction result, and generate maintenance information according to the prediction results of each type of label.

[0007] The beneficial effects are as follows: This application can monitor the operating status of the device and predict the fault trend, and take timely maintenance measures. This application is not based on time, but on the actual health status of the device. By real-time monitoring, data analysis and model prediction of the device, it is judged whether the device needs repair or component replacement, so as to reduce the device failure rate and improve the device utilization rate and production efficiency.

[0008] By introducing the degree of confusion and similarity to quantify the importance of device parameters for device operation labels at each delay time, this quantification method enables the model to dynamically adjust the attention degree to different parameters, thereby improving the accuracy and reliability of prediction.

[0009] Optionally, the expression of the degree of confusion is: ; In the formula, represents the degree of confusion between the type of label and device parameters at a delay time of ; represents the probability that the th device parameter appears under the type of label at a delay time of ; represents the logarithmic function.

[0010] The beneficial effects are as follows: Using the form of information entropy to measure the degree of confusion can effectively reflect the uncertainty of device parameters in different operating states. A higher degree of confusion (i.e., a higher entropy value) indicates that the distribution of device parameters is more uniform in this state and it is difficult to distinguish; a lower degree of confusion indicates that certain parameter values tend to specific states more, which helps to improve the discrimination ability of prediction. The greater the degree of confusion between a certain type of label and device parameters, it means that at this delay time, the distribution of device parameters under the same label is chaotic, and it is more difficult to judge the device label through device parameters.

[0011] Optionally, the expression of the degree of confusion is: ; where, represents the degree of chaos of the th class labels and device parameters at the delay time of represents the probability that the th type of device parameter appears under the th class label at the delay time of represents the value of the th device parameter, represents the average value of all device parameters, represents the variance of the device parameters under the th class label at the delay time of

[0012] The beneficial effect is that by combining the mean and variance of the device parameters, this expression can more intuitively measure the degree to which the device parameters deviate from the normal range. This standardized method of calculating the degree of chaos can effectively eliminate the influence of different parameter dimensions, making the degrees of chaos between different device parameters comparable.

[0013] Optionally, the calculation expression for the similarity of the parameter distribution is: ; where, represents the similarity of the device parameter distributions of the th class label and other labels at the delay time of represents the total number of category labels, represents the th class label and the th divergence of the device parameters under the

[0014] The beneficial effect is that by using the divergence to measure the differences between the device parameter distributions in different operating states and adopting an exponential decay form to convert these differences into similarities, the purpose of quantifying the relative differences between the device parameter distributions is achieved.

[0015] Optionally, the calculation expression for the similarity of the parameter distribution is: ; where, represents the similarity of the device parameter distributions of the th class label and other labels at the delay time of represents the total number of category labels, represents the variance of the device parameters under the th class label at the delay time of Indicates the variance of the device parameters under the -th class label at the delay time.

[0016] The beneficial effect is that by calculating the variance of the device parameters and to measure the similarity under different operating states, it can intuitively reflect the fluctuation degree of the device parameters in different states. This similarity calculation method based on variance can effectively avoid misjudgment caused by the absolute value difference of parameter distributions.

[0017] Optionally, the expression of the prediction result is: ; where represents the prediction result of the -th class label of the device, represents the importance of the device parameters for the -th class label at the delay time, represents the prediction result of the classification model for the -th class label at the delay time, represents the prediction result of the classification model for the -th class label, represents the corresponding weight coefficient.

[0018] Optionally, the construction method of the classification model is: construct an initial model, train the initial model according to the labeled historical operation data to obtain a classification model, the loss function of model training is the cross-entropy loss function, and use the gradient descent algorithm to update the model parameters; when the model reaches the preset number of training times or the loss function of the model is less than the preset loss threshold, the model stops training to obtain a classification model.

[0019] In the second aspect, the present application provides an industrial equipment predictive maintenance system, adopting the following technical solution: An industrial equipment predictive maintenance system, including a processor and a memory, where the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned industrial equipment predictive maintenance method is implemented.

[0020] The beneficial effect is that the above-mentioned industrial equipment predictive maintenance method is generated into a computer program and stored in the memory to be loaded and executed by the processor. Thus, a system is made according to the memory and the processor, which is convenient to use.

[0021] The present application has the following technical effects: This application can monitor the operating status of the device and predict the fault trend, and take timely maintenance measures. This application is not based on time, but on the actual health status of the device. By real-time monitoring, data analysis and model prediction of the device, it is judged whether the device needs repair or component replacement, so as to reduce the device failure rate and improve the device utilization rate and production efficiency.

[0022] The corresponding relationship between the device operating status monitoring and the device parameters is determined by calculating the delay time of the device parameters and the device operating status. Calculate the chaos degree of the device parameters under each device operating status at different delay times, calculate the similarity of the device parameter distributions between different device operating states at different delay times, calculate the importance of the device parameters to each device operating state at different delay times according to the chaos degree and similarity, and integrate the device operation state prediction results at different delay times. By introducing the chaos degree and similarity to quantify the importance of the device parameters to the device operation label at each delay time, this quantification method enables the model to dynamically adjust the attention degree to different parameters, thereby improving the accuracy and reliability of the prediction. Brief Description of the Drawings

[0023] By referring to the following detailed description with reference to the accompanying drawings, the above and other objects, features and advantages of the exemplary embodiments of the present application will become readily understood. In the drawings, several embodiments of the present application are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts.

[0024] Figure 1 It is a flowchart of a method for predictive maintenance of industrial equipment in an embodiment of the present application. Detailed Embodiments

[0025] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present application.

[0026] It should be understood that when terms such as "first" and "second" are used in the claims, the description and drawings of the present application, they are only used to distinguish different objects, rather than to describe a specific order. The terms "including" and "comprising" used in the description and claims of the present application indicate the presence of the described features, wholes, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or their combinations.

[0027] An embodiment of the present application discloses a method for predictive maintenance of industrial equipment. Refer toFigure 1 , including steps S1 - S4, specifically as follows: S1: Process and label the historical equipment parameters during the operation of industrial equipment. The labels include normal operation, to be repaired, and malfunction.

[0028] Obtain the historical equipment parameters, such as voltage, current, power, etc., and preprocess the historical equipment parameters. After preprocessing, those skilled in the art assign a label to each sample according to the actual situation. The labels are normal operation of the equipment, to be repaired, or malfunction.

[0029] The method of preprocessing is to delete the samples containing missing values and the samples with abnormal collected parameter values. The missing values and abnormal values may be caused by sensor failures, data transmission errors, or abnormal equipment operation.

[0030] S2: Calculate the delay time between the label and the historical operation parameters, and calculate the importance of the equipment parameters for the equipment operation label at each delay time.

[0031] Predict the working state of the equipment through the operation parameters of the equipment, and determine whether the equipment needs maintenance. Before the equipment has problems, it often reflects in advance in the equipment operation parameters, but how long in advance it will be reflected may vary for different equipment problems. Therefore, calculate the delay time between each label and each parameter. The delay time reflects that the similarity between the equipment parameters and the equipment state is the largest within this time interval. Then, predicting the equipment state based on the equipment parameters at this delay time has a better effect.

[0032] Calculate the delay time between each label and each parameter. Calculate the confusion degree between the equipment operation state label and the equipment parameters at different delay times, and at the same time calculate the distribution similarity of the equipment parameters between different operation state labels. Calculate the importance of the equipment parameters for the equipment operation label at different delay times according to the confusion degree and the distribution similarity.

[0033] In one embodiment, the expression of the confusion degree is: ; where represents the confusion degree between the th type of label and the equipment parameters at the delay time of , represents the probability that the th type of equipment parameter appears under the th type of label at the delay time of , represents the logarithmic function.

[0034] As becomes more uniform, increases; when concentrates on a few parameters, Decrease. Using the form of information entropy to measure the degree of chaos can effectively reflect the uncertainty of device parameters in different operating states. A higher degree of chaos (i.e., a higher entropy value) indicates that the distribution of device parameters in this state is more uniform and difficult to distinguish; a lower degree of chaos indicates that certain parameter values tend to be in specific states, which helps to improve the discrimination ability of prediction. The greater the degree of chaos of a certain type of label and device parameters, the more chaotic the distribution of device parameters under the same label at this delay time, and the greater the difficulty of judging the device label through device parameters. This embodiment is more suitable for scenarios with a clear probability distribution and relatively uniform data, and can quickly quantify the uncertainty.

[0035] In one embodiment, the expression for the degree of chaos is: ; where represents the degree of chaos of the th type of label and device parameters at the delay time of , represents the probability that the th type of device parameter appears under the th type of label at the delay time of , represents the value of the rd type of device parameter, represents the average value of all device parameters, represents the variance of the device parameters under the th type of label at the delay time of .

[0036] When the gap between and increases, increases, indicating that the degree of deviation of the parameter from the normal range increases; when increases, decreases, indicating that the degree of dispersion of the parameter distribution increases and the degree of chaos decreases relatively. By combining the mean and variance of the device parameters, this expression can more intuitively measure the degree of deviation of the device parameters from the normal range. This standardized way of calculating the degree of chaos can effectively eliminate the influence of different parameter dimensions, making the degrees of chaos between different device parameters comparable. This embodiment is more suitable for scenarios where the actual numerical values of parameters need to be considered and the data dimensions are inconsistent, and can be closer to the actual physical meaning.

[0037] In one embodiment, the calculation expression for the similarity of the parameter distribution is: ; where represents the similarity of the device parameter distributions of the th type of label and other labels at the delay time of , Indicates the total number of class labels, Indicates at the delay time of the class label and the divergence of the device parameters under the class label, The smaller this value is, the more similar it indicates. The greater the similarity, the more difficult it is to predict the operating state of the device using the device parameters at this delay time.

[0038] When increases, decreases, indicating that the parameter distribution difference between different states increases and the similarity decreases; conversely, when decreases, increases, indicating that the similarity between parameter distributions increases. By using divergence to measure the difference between the device parameter distributions under different operating states and adopting an exponential decay form to convert these differences into similarity, the purpose of quantifying the relative differences between the device parameter distributions is achieved.

[0039] In one embodiment, the calculation expression of the similarity of the parameter distribution is: ; where represents the similarity between the device parameter distributions of the class label and other label devices at the delay time of , represents the total number of class labels, represents the variance of the device parameters under the class label at the delay time of , represents the variance of the device parameters under the class label at the delay time of .

[0040] When the variance relative to other states increases, decreases, indicating that the difference between the current state and other states increases; conversely, when is relatively small or close to the variance of other states, increases, indicating that the similarity increases.

[0041] By using the variance and of the device parameters to measure the similarity under different operating states, it can intuitively reflect the fluctuation degree of the device parameters in different states. This similarity calculation method based on variance can effectively avoid misjudgment caused by the absolute value difference of parameter distributions.

[0042] Among them, the calculation formula of importance is: ; wherein, represents the importance of the device parameters for the -th type of label at the delay time of ; represents the degree of confusion between the -th type of label and the device parameters at the delay time of ; represents the similarity of the distribution of the -th type of label and the device parameters of other labels at the delay time of ;

[0043] S3: Input the real-time device parameters of the industrial device into a preset classification model, and output the predicted values of the probabilities corresponding to each label of the real-time operating parameters.

[0044] Construct an initial model, and train the initial model according to the historical operating data with labels to obtain a classification model. Exemplarily, the initial model can be a BP (Backpropagation) model or an LSTM (Long Short-Term Memory) network model. The loss function for model training is the cross-entropy loss function, and the gradient descent algorithm is used to update the model parameters. When the model reaches the preset number of training times or the loss function of the model is less than the preset loss threshold, the model stops training to obtain a classification model. The preset number of training times can be 1000 times, and the loss threshold can be 0.01. The output is the predicted value of the operating state of the device at this delay time. The construction and training of the model are both prior arts and will not be elaborated here.

[0045] Construct classification models at different delay times, input the real-time collected device parameters into the corresponding classification models respectively, and obtain the predicted values of the operating states of the device at the same moment at different delay times.

[0046] S4: Calculate the weight coefficient according to the importance, and each type of label corresponds to a weight coefficient; use the product of the predicted value and the corresponding weight coefficient as the prediction result, and generate maintenance information according to the prediction results of each type of label.

[0047] The expression of the prediction result is: ; wherein, represents the prediction result of the -th type of label of the device, represents the importance of the device parameters for the -th type of label at the delay time of ; represents the prediction result of the classification model for the -th type of label at the delay time of ;

[0048] Indicates The corresponding weight coefficient, which can weightedly adjust the prediction result according to the importance of the device parameters at different delay times. This weighting method enables the model to more flexibly handle the device state changes at different time scales, thereby improving the accuracy and robustness of the prediction.

[0049] Calculate the probabilities of different operating states of the device according to the above method. The user can set different thresholds according to the importance of different labels. For example, the impact of device failure is relatively large. Therefore, when the probability of device failure is greater than a relatively small threshold, it is considered that there is a failure. The impact of the device to be repaired is relatively small compared to the device failure. Then, the threshold for the probability of the device to be repaired is set to be larger than that of the device failure. An alarm is issued when it is detected that the device is in a state of being to be repaired or has a failure. Exemplarily, if the failure probability > 0.3, it is recommended to immediately stop the machine for inspection. If the probability to be repaired > 0.5, it is recommended to arrange a maintenance plan.

[0050] The embodiment of the present application also discloses an industrial device predictive maintenance system, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the industrial device predictive maintenance method according to the present application.

[0051] The above system further includes a communication bus, a communication interface and other components well known to those skilled in the art. Their settings and functions are known in the art, so they will not be elaborated here.

[0052] In the present application, the aforementioned memory can be any tangible medium that contains or stores a program, which can be used by or in combination with an instruction execution system, apparatus or device. For example, the computer-readable storage medium can be any suitable magnetic storage medium or magneto-optical storage medium, such as, resistive random access memory RRAM (Resistive Random Access Memory), dynamic random access memory DRAM (Dynamic Random Access Memory), static random access memory SRAM (Static Random Access Memory), enhanced dynamic random access memory EDRAM (Enhanced Dynamic Random Access Memory), high bandwidth memory HBM (High Bandwidth Memory), hybrid memory cube HMC (Hybrid Memory Cube), etc., or any other medium that can be used to store the required information and can be accessed by an application program, a module or both. Any such computer storage medium can be part of the device or accessible or connectable to the device.

[0053] Although this specification has shown and described multiple embodiments of the present application, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art will think of many changes, alterations, and alternative ways without departing from the spirit and scope of the present application. It should be understood that various alternatives to the embodiments of the present application described herein may be employed in the practice of the present application.

[0054] The above are all preferred embodiments of the present application, and do not limit the protection scope of the present application accordingly. Therefore, all equivalent changes made according to the structure, shape, and principle of the present application shall be covered within the protection scope of the present application.

Claims

1. A method for predictive maintenance of industrial equipment, characterized in that: Includes steps: Process and label the historical equipment parameters of industrial equipment operation, including normal operation, pending maintenance, and failure; calculate the delay time between the label and the historical operation parameters, and calculate the importance of the equipment parameters to the equipment operation label at each delay time; The calculation formula of importance is: ; In the formula, Indicates that the delay time is When the device parameters are The importance of class labels, Indicates that the delay time is Time The level of confusion of class labels and device parameters, Indicates that the delay time is Time Similarity between class label and other label device parameter distributions; The real-time equipment parameters of industrial equipment are input into the preset classification model, and the predicted values ​​of the real-time operating parameters corresponding to the probability of each label are output; the weight coefficient is calculated according to the importance, and each type of label corresponds to a weight coefficient; the product of the predicted value and the corresponding weight coefficient is used as the prediction result, and maintenance information is generated according to the prediction results of each type of label.

2. The method for predictive maintenance of industrial equipment according to claim 1, characterized in that: The expression of the degree of chaos is: ; In the formula, Indicates that the delay time is Time The level of confusion of class labels and device parameters, Indicates that the delay time is Time Class label The probability of occurrence of a device parameter, Represents a logarithmic function.

3. The method for predictive maintenance of industrial equipment according to claim 1, characterized in that: The expression of the degree of chaos is: ; In the formula, Indicates that the delay time is Time The level of confusion of class labels and device parameters, Indicates that the delay time is Time Class label The probability of occurrence of a device parameter, Indicates The value of a device parameter. represents the average value of all device parameters, Indicates that the delay time is Time Variance of device parameters under class labels.

4. The method for predictive maintenance of industrial equipment according to claim 1, characterized in that: The calculation expression of the similarity of parameter distribution is: ; In the formula, Indicates that the delay time is Time Similarity between class label and other label device parameter distributions, represents the total number of category labels, Indicates that the delay time is Time Class labels and Device parameters under the class tag Divergence.

5. The method for predictive maintenance of industrial equipment according to claim 1, characterized in that: The calculation expression of the similarity of parameter distribution is: ; In the formula, Indicates that the delay time is Time Similarity between class label and other label device parameter distributions, represents the total number of category labels, Indicates that the delay time is Time The variance of device parameters under class labels, Indicates that the delay time is Time Variance of device parameters under class labels.

6. The method for predictive maintenance of industrial equipment according to claim 1, characterized in that: The expression of the prediction result is: ; In the formula, Indicates the device The prediction results of class labels, Indicates that the delay time is When the device parameters are The importance of class labels, Indicates that the delay time is The classification model at this time is The prediction results of class labels, express The corresponding weight coefficient.

7. The method for predictive maintenance of industrial equipment according to any one of claims 1 to 6, characterized in that: The construction method of the classification model is as follows: construct an initial model, train the initial model according to the historical running data after labeling to obtain a classification model, the loss function of the model training is the cross entropy loss function, and the gradient descent algorithm is used to update the model parameters; when the model reaches the preset number of training times or the loss function of the model is less than the preset loss threshold, the model stops training to obtain a classification model.

8. An industrial equipment predictive maintenance system, characterized in that: include: A processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the industrial equipment predictive maintenance method according to any one of claims 1 to 7 is implemented.

Images