Predictive Maintenance Control System for Embedded Computers Based on Digital Twin

Abstract

The present invention discloses a predictive maintenance control system for an embedded computer based on digital twin, which relates to the field of computer technology and includes: a data fusion and modeling module that uses a digital twin to build a high-precision digital model; a fault and anomaly detection module that is used to determine whether the junction temperature of a chip exceeds the early warning critical point, and if so, issues an early warning, extracts the vibration signal energy entropy value by using wavelet packet decomposition, and determines whether the energy entropy value exceeds a preset value. If so, it is determined that the embedded computer has an anomaly; a prediction and equipment maintenance module that dynamically adjusts the maintenance cycle by using reinforcement learning; a human-machine collaborative decision-making module that obtains a machine evaluation, obtains a human evaluation based on a high-precision digital model, and combines the machine evaluation and the human evaluation to obtain a human-machine collaborative evaluation. The advantages of the present invention are as follows: it can predict the problems of the embedded computer prospectively, dynamically adjust the maintenance cycle, and realize the predictive maintenance of the embedded computer.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and more particularly to an embedded computer predictive maintenance control system based on digital twin. Background Art

[0002] Industrial embedded computers come with an embedded operating system, featuring stability, no crashing, and immunity to viruses. They have an independent embedded system disk, do not require a hard disk as the system disk, have no virtual memory, and are not afraid of power-off damage to the operating system. They can be used immediately upon power-on. Such computers require a maintenance device during use.

[0003] If the usage cycle of the maintenance device is too short, the maintenance cost will be too high. If the usage cycle of the maintenance device is too long, the likelihood of the embedded computer malfunctioning will increase. The maintenance of the embedded computer is carried out on the premise that the embedded computer has not malfunctioned, and the reason for maintenance is also to reduce the likelihood of malfunction. If the embedded computer malfunctions, the maintenance loses its meaning. Summary of the Invention

[0004] To solve the above technical problems, an embedded computer predictive maintenance control system based on digital twin is provided. This technical solution solves the problems of difficult determination of the maintenance cycle and the occurrence of malfunctions in the embedded computer before maintenance as mentioned in the above background art.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] An embedded computer predictive maintenance control system based on digital twin, comprising:

[0007] A data fusion and modeling module, which sets sensors to monitor the temperature and vibration data of the embedded computer in real time, uploads the data to the cloud, and constructs a high-precision digital model using the digital twin.

[0008] A fault anomaly detection module, which evaluates the change in the junction temperature of the chip based on the high-precision digital model, sets a junction temperature warning critical point. If the junction temperature of the chip exceeds the warning critical point, it issues a warning. Based on the high-precision digital model, wavelet packet decomposition is used to extract the energy entropy value of the vibration signal, and it is judged whether the energy entropy value exceeds a preset value. If so, it is judged that the embedded computer has an anomaly.

[0009] A prediction device maintenance module, which dynamically adjusts the maintenance cycle using reinforcement learning based on the prediction of the remaining life of the device.

[0010] The human-machine collaborative decision-making module deploys a lightweight CNN model at the device end to achieve local real-time analysis of the vibration spectrum, obtain a machine evaluation, obtain a human evaluation based on a high-precision digital model, merge the machine evaluation and the human evaluation to obtain a human-machine collaborative evaluation, and perform maintenance control.

[0011] Preferably, the sensor is set to monitor the temperature and vibration data of the embedded computer in real time, upload the data to the cloud, and the construction of a high-precision digital model using the digital twin includes:

[0012] Obtain the external dimensions of the embedded computer and construct a three-dimensional solid model of the embedded computer;

[0013] Obtain the hardware component models, usage time, and software components of the embedded computer, and construct a preliminary high-precision digital model of the embedded computer based on the three-dimensional solid model of the embedded computer;

[0014] Obtain the temperature, vibration, and electrical parameter data of the embedded computer during operation, and record them as real parameters;

[0015] Use the preliminary high-precision digital model to simulate the normal operation of the embedded computer, and record the temperature, vibration, and electrical parameter data of the high-precision digital model, and record them as simulation parameters;

[0016] Obtain the normal fluctuation range of each parameter during the normal operation of the embedded computer, calculate the length of the normal fluctuation range of each parameter, and record half of the length of the normal fluctuation range of each parameter as the fluctuation preset value of each parameter;

[0017] Compare the real parameters with the simulation parameters, calculate the difference between the real parameters and the simulation parameters, list each parameter as a target parameter one by one, and judge whether the difference of the target parameter is greater than the fluctuation preset value of the target parameter. If so, record the target parameter as an influencing parameter;

[0018] Adjust the hardware model, usage time, and software components of the high-precision digital model, and record the changes in each parameter of the high-precision digital model;

[0019] Quantify the adjusted hardware model and usage time of the high-precision digital model into numbers according to the release year of the model and the length of the usage time, and record them as independent variable parameters. Quantify the degree of change of each parameter into numbers, and record them as dependent variable parameters;

[0020] Based on the changes in each parameter of the high-precision digital model, calculate the Pearson correlation coefficient between the independent variable parameters and the dependent variable parameters to obtain the correlation matrix between the independent variable parameters and the dependent variable parameters;

[0021] Compare the difference between the influencing parameter and the true parameter, and determine the adjustment degree of the independent variable parameter corresponding to the influencing parameter based on the product of the Pearson coefficient of the correlation matrix of the influencing parameter and the difference;

[0022] Summarize the adjustment degrees of the independent variable parameters corresponding to at least one influencing parameter to obtain the adjustment details of the hardware and software;

[0023] Based on the obtained adjustment details of the hardware and software, adjust the preliminary high-precision digital model to obtain a high-precision digital model;

[0024] The Pearson correlation coefficient is a statistic used to measure the linear correlation degree between two continuous variables, and its value range is from -1 to 1.

[0025] Preferably, for evaluating the change of the junction temperature of the evaluation chip and setting the junction temperature warning critical point, it includes:

[0026] Obtain the temperature and air flow of the environment where the embedded computer is located;

[0027] Use the data display in the high-precision digital model to obtain the average temperature at the chip node in the high-precision digital model, which is recorded as the junction temperature of the chip;

[0028] Obtain the usage time and load of the common software used by the user on the embedded computer;

[0029] Obtain the junction temperature when the user uses all common software simultaneously, which is recorded as the maximum junction temperature;

[0030] Evaluate the aging speed and performance of the chip when the maximum junction temperature is exceeded, and draw the change image of the junction temperature and the aging speed and performance of the chip;

[0031] Find a point with the smallest derivative on the image, and record the corresponding junction temperature as the warning critical point.

[0032] Preferably, for using wavelet packet decomposition to extract the energy entropy value of the vibration signal and determining whether the energy entropy value exceeds a preset value, if so, determining that the embedded computer has an abnormality, it includes:

[0033] Obtain the vibration signal of the high-precision digital model;

[0034] Use wavelet packet decomposition for the vibration signal to obtain at least one divided frequency band;

[0035] Determine whether the decomposition layer number of the wavelet packet decomposition of the vibration signal belongs to the standard decomposition layer number interval. If so, calculate the energy entropy value of the vibration signal using the entropy value calculation formula. If not, obtain the vibration signal of the high-precision digital model again and use wavelet packet decomposition for the vibration signal;

[0036] Determine whether the energy entropy value exceeds the preset value. If so, determine that the embedded computer has an abnormality;

[0037] The entropy value calculation formula is as follows: ,

[0038] where H is the entropy value energy, is the ratio of the number of the i-th frequency band to the total number of all frequency bands, and n is the number of frequency bands after wavelet packet decomposition of the vibration signal;

[0039] The wavelet packet decomposition is to let the signal pass through a series of filters with different center frequencies but the same bandwidth to obtain signals in at least one frequency band;

[0040] The steps for obtaining the preset value and the standard decomposition layer interval are as follows:

[0041] Obtain at least one vibration frequency when the embedded computer is running normally for at least one time period;

[0042] Use wavelet packet decomposition for at least one vibration frequency to obtain the decomposition layer number and energy entropy value of at least one vibration frequency;

[0043] Draw a bar chart of the decomposition layer number and energy entropy value of the vibration frequency;

[0044] Based on the bar chart, find out the decomposition layer interval corresponding to the concentrated distribution of the energy entropy value, denoted as the standard decomposition layer interval, and the average entropy value corresponding to the concentrated distribution of the energy entropy value is denoted as the preset value.

[0045] Preferably, the dynamic adjustment of the maintenance period based on the prediction of the remaining life of the device by using reinforcement learning includes:

[0046] Obtain the average daily usage time and average daily usage intensity of the user using the embedded computer;

[0047] Simulate the daily operation of the high-precision digital model in the virtual world to accelerate the time flow rate in the virtual time;

[0048] If the data simulation driver cannot use the software normally, it is determined that the data simulation driver is damaged;

[0049] Obtain the running time of the high-precision digital model from running to damage, denoted as the remaining life;

[0050] Obtain the daily usage situation of the user, update the average daily usage time and usage intensity of the user using the embedded computer every day, and obtain the updated remaining life of the embedded computer;

[0051] Based on the remaining life of the embedded computer, calculate the initial maintenance period of the embedded computer by using the period calculation formula;

[0052] During the initial maintenance period, determine whether the updated remaining life is greater than the updated remaining life of the previous day. If so, add one day to the initial maintenance period. If not, subtract one day from the initial maintenance period;

[0053] The formula for the period is as follows: ,

[0054] In the formula, T is the initial maintenance period, A is the remaining life of the embedded computer, and B is the average number of maintenance times of the embedded computer.

[0055] Preferably, the human-machine collaborative decision-making module deploys a lightweight CNN model at the device end to realize local real-time analysis of the vibration spectrum, and the machine evaluation obtained includes:

[0056] Deploy a CNN model on the embedded computing to obtain the vibration signal of the embedded computer;

[0057] Use the CNN model to analyze the vibration signal of the embedded computer in real time and convert it into an image, denoted as the real-time signal;

[0058] Use artificial intelligence to analyze and store the vibration images of at least one embedded computer running normally, complete machine learning, and obtain the normal signal;

[0059] Artificial intelligence analyzes and compares the real-time signal with the normal signal to obtain the similarity between the real-time signal and the normal signal, and the similarity is expressed as a percentage;

[0060] The difference between 1 and the similarity is denoted as the machine evaluation;

[0061] The output of the CNN model is the feature vector processed by the convolutional layer, pooling layer and fully connected layer.

[0062] Preferably, the human evaluation obtained based on the high-precision digital model includes:

[0063] Obtain the chip information, hardware information, and software operation status of the high-precision digital model;

[0064] Invite at least one expert to analyze the chip information, hardware information, and software operation status to obtain the evaluation of the embedded computer;

[0065] Analyze the authority and academic achievements of at least one expert, and use the analytic hierarchy process to obtain the analysis weight of each expert;

[0066] Calculate the product sum of the evaluations of at least one expert and the analysis weight to obtain the human evaluation.

[0067] Preferably, determine whether the machine evaluation is greater than the preset value. If so, calculate the human evaluation;

[0068] The initial weight of the machine evaluation is equal to the initial weight of the human evaluation;

[0069] Multiply the sum of the machine evaluation plus 1 by the initial weight of the machine evaluation to obtain the weight of the machine evaluation;

[0070] The difference between 1 and the weight of the machine evaluation is denoted as the weight of the human evaluation;

[0071] The sum of the product of the weight of the machine evaluation and the machine evaluation and the product of the weight of the human evaluation and the human evaluation is the human-machine collaborative evaluation;

[0072] Obtain the quantity and priority of the components maintained by the embedded computer;

[0073] Based on the quantity and priority of the components maintained by the embedded computer, divide them into several maintenance levels;

[0074] Divide the human-machine collaborative evaluation into several intervals, and each interval corresponds to a maintenance level from small to large;

[0075] Determine the interval to which the human-machine collaborative evaluation belongs and execute the corresponding maintenance level;

[0076] The steps for obtaining the preset value are as follows:

[0077] Count the number of times a failure occurs in at least one machine-evaluated embedded computer;

[0078] Select the smallest machine evaluation with the number of failures greater than 1 as the preset value.

[0079] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0080] The present invention proposes a predictive maintenance control system for an embedded computer based on digital twin, which uses digital twin to construct a high-precision digital model of the embedded computer, constructs a one-to-one restored model in the virtual world, helps to achieve cost-free simulation operation, improves the accuracy of prediction, and has a fault anomaly detection module that can timely detect the faults of the embedded computer and avoid the situation where the embedded computer fails during maintenance and cannot be maintained. The prediction equipment maintenance module uses the high-precision digital model to determine the predicted life of the embedded computer, thereby dynamically adjusting the maintenance cycle, solving the problems of increased probability of failure of embedded computers with too short maintenance cycles and too high costs for too long maintenance cycles. The human-machine collaborative decision-making module integrates the accurate prediction of the computer and the empirical prediction of humans, greatly improving the accuracy of fault probability prediction. Description of the Drawings

[0081] Figure 1 It is a schematic flow diagram of the predictive maintenance control system for an embedded computer based on digital twin of the present invention;

[0082] Figure 2 It is a schematic flow chart of setting sensors to monitor the temperature and vibration data of the embedded computer in real time in the present invention, uploading the data to the cloud, and constructing a high-precision digital model using digital twins;

[0083] Figure 3 It is a schematic flow chart of evaluating the change of the junction temperature of the chip and setting the critical point of junction temperature warning in the present invention;

[0084] Figure 4 It is a schematic flow chart of extracting the energy entropy value of the vibration signal by wavelet packet decomposition in the present invention, judging whether the energy entropy value exceeds a preset value, and if so, judging that the embedded computer has an abnormality;

[0085] Figure 5 It is a schematic flow chart of dynamically adjusting the maintenance period by reinforcement learning based on the prediction of the remaining life of the device in the present invention;

[0086] Figure 6 It is a schematic flow chart of deploying a lightweight CNN model at the device end by the human-machine collaborative decision-making module in the present invention to realize local real-time analysis of the vibration spectrum and obtain machine evaluation;

[0087] Figure 7 It is a schematic flow chart of obtaining human evaluation based on a high-precision digital model in the present invention;

[0088] Figure 8 It is a schematic flow chart of judging whether the machine evaluation is greater than a preset value, and if so, calculating the human evaluation in the present invention. Detailed implementation manners

[0089] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments in the following description are only examples, and those skilled in the art can think of other obvious variations.

[0090] Referring to Figure 1 As shown, the predictive maintenance control system for an embedded computer based on digital twins includes:

[0091] A data fusion and modeling module, which sets sensors to monitor the temperature and vibration data of the embedded computer in real time, uploads the data to the cloud, and constructs a high-precision digital model using digital twins;

[0092] A fault and abnormality detection module, which evaluates the change of the junction temperature of the chip based on the high-precision digital model, sets the critical point of junction temperature warning, and if the junction temperature of the chip exceeds the warning critical point, issues a warning. Based on the high-precision digital model, wavelet packet decomposition is used to extract the energy entropy value of the vibration signal, and it is judged whether the energy entropy value exceeds a preset value. If so, it is judged that the embedded computer has an abnormality;

[0093] A predictive device maintenance module that dynamically adjusts the maintenance cycle using reinforcement learning based on the predicted remaining life of the device.

[0094] A human-machine collaborative decision-making module that deploys a lightweight CNN model at the device end to achieve local real-time analysis of the vibration spectrum, obtain a machine evaluation, obtain a human evaluation based on a high-precision digital model, merge the machine evaluation and the human evaluation to obtain a human-machine collaborative evaluation, and perform maintenance control.

[0095] It can be explained that using digital twins to build a high-precision digital model of an embedded computer and constructing a one-to-one restored model in the virtual world helps to achieve cost-free simulation operation, improve the accuracy of prediction. The fault and anomaly detection module can timely detect the faults of the embedded computer and avoid the situation where the embedded computer fails during maintenance and cannot be maintained. The predictive device maintenance module uses the high-precision digital model to determine the predicted life of the embedded computer, thereby dynamically adjusting the maintenance cycle, solving the problems of increased failure probability of embedded computers with too short maintenance cycles and too high costs of too long maintenance cycles. The human-machine collaborative decision-making module combines the accurate prediction of the computer and the empirical prediction of humans, greatly improving the accuracy of fault probability prediction.

[0096] Refer to Figure 2 As shown, sensors are set to monitor the temperature and vibration data of the embedded computer in real time, and the data is uploaded to the cloud. Using digital twins to build a high-precision digital model includes:

[0097] Obtain the external dimensions of the embedded computer and construct a three-dimensional solid model of the embedded computer;

[0098] Obtain the hardware component model, usage time, and software composition of the embedded computer, and construct a preliminary high-precision digital model of the embedded computer based on the three-dimensional solid model of the embedded computer;

[0099] Obtain the temperature, vibration, and electrical parameter data of the embedded computer during operation, denoted as real parameters;

[0100] Use the preliminary high-precision digital model to simulate the normal operation of the embedded computer, and record the temperature, vibration, and electrical parameter data of the high-precision digital model, denoted as simulated parameters;

[0101] Obtain the normal fluctuation range of each parameter during the normal operation of the embedded computer, calculate the length of the normal fluctuation range of each parameter, and denote half of the length of the normal fluctuation range of each parameter as the fluctuation preset value of each parameter;

[0102] Compare the real parameters with the simulated parameters, calculate the difference between the real parameters and the simulated parameters, list each parameter as a target parameter one by one, and determine whether the difference of the target parameter is greater than the preset fluctuation value of the target parameter. If so, record the target parameter as an influencing parameter;

[0103] Adjust the hardware model, usage time, and software composition of the high-precision digital model, and record the changes in various parameters of the high-precision digital model;

[0104] Quantify the adjustment of the hardware model and usage time of the high-precision digital model into numbers according to the release year of the model and the length of the usage time, and record them as independent variable parameters. Quantify the degree of change of each parameter into numbers and record them as dependent variable parameters;

[0105] Based on the changes in various parameters of the high-precision digital model, calculate the Pearson correlation coefficient between the independent variable parameters and the dependent variable parameters to obtain the correlation matrix between the independent variable parameters and the dependent variable parameters;

[0106] Compare the difference between the influencing parameter and the real parameter, and determine the adjustment degree of the independent variable parameter corresponding to the influencing parameter based on the product of the Pearson coefficient of the correlation matrix of the influencing parameter and the difference;

[0107] Summarize the adjustment degrees of the independent variable parameters corresponding to at least one influencing parameter to obtain the adjustment details of the hardware and software;

[0108] Based on the obtained adjustment details of the hardware and software, adjust the preliminary high-precision digital model to obtain the high-precision digital model;

[0109] The Pearson correlation coefficient is a statistic used to measure the linear correlation degree between two continuous variables, and its value range is from -1 to 1.

[0110] It can be understood that a high-precision digital model is constructed using temperature, vibration, and electrical parameter data. The temperature, vibration, and electrical parameter data can reflect the operating conditions of the embedded computer, which helps to construct a high-precision digital model with higher accuracy. Based on the comparison of the high-precision digital model and the temperature, vibration, and electrical parameter data of the normal operation of the embedded computer, the hardware and software configurations of the high-precision digital model are adjusted to simulate the operation of the embedded computer with a more accurate high-precision digital model, improving the accuracy of predictive maintenance. Each element in each row of the correlation matrix is composed of the Pearson correlation coefficient between one variable in the hardware model and usage time according to the release year of the model and the length of the usage time and the degree of change of each parameter, reflecting the linear correlation degree between the two, providing a reference for adjusting the hardware model and usage time according to the release year of the model and the length of the usage time to make the temperature, vibration, and electrical parameter data meet the standards.

[0111] Refer to Figure 3As shown, to evaluate the change in the junction temperature of the evaluation chip, setting the junction temperature warning critical point includes:

[0112] Obtain the temperature and air flow of the environment where the embedded computer is located;

[0113] Use the data display in the high-precision digital model to obtain the average temperature at the chip node in the high-precision digital model, denoted as the junction temperature of the chip;

[0114] Obtain the usage time and load of the commonly used software by the user on the embedded computer;

[0115] Obtain the junction temperature when the user uses all commonly used software simultaneously, denoted as the maximum junction temperature;

[0116] Evaluate the aging speed and performance of the chip when the maximum junction temperature is exceeded, and draw the change images of the junction temperature and the aging speed and performance of the chip;

[0117] Find a point with the minimum derivative on the image, and denote the corresponding junction temperature as the warning critical point.

[0118] It can be understood that the junction temperature obtained by using the high-precision digital model constructed by digital twin is simpler and more accurate than other methods. Maintaining the chip junction temperature within the normal range is an important way to ensure the normal power operation of the device and improve the service life of the device. If the junction temperature warning critical point is too high, it will cause accelerated device aging and affect the operation rate of the device. If the junction temperature warning critical point is too low, it is not conducive to the normal operation of the device. Therefore, it is very important to set a suitable chip junction temperature warning critical point.

[0119] Refer to Figure 4 As shown, adopt wavelet packet decomposition to extract the energy entropy value of the vibration signal, and judge whether the energy entropy value exceeds the preset value. If so, judge that the embedded computer has an abnormality, including:

[0120] Obtain the vibration signal of the high-precision digital model;

[0121] Use wavelet packet decomposition on the vibration signal to obtain at least one divided frequency band;

[0122] Judge whether the decomposition layer number of the wavelet packet decomposition of the vibration signal belongs to the standard decomposition layer number interval. If so, calculate the energy entropy value of the vibration signal using the entropy value calculation formula. If not, obtain the vibration signal of the high-precision digital model again and use wavelet packet decomposition on the vibration signal;

[0123] Judge whether the energy entropy value exceeds the preset value. If so, judge that the embedded computer has an abnormality;

[0124] The entropy value calculation formula is: ,

[0125] In the formula, H is the entropy value energy, is the ratio of the number of the i-th frequency band to the total number of all frequency bands, and n is the number of frequency bands after wavelet packet decomposition of the vibration signal;

[0126] The wavelet packet decomposition is to let the signal pass through a series of filters with different center frequencies but the same bandwidth to obtain signals in at least one frequency band;

[0127] The steps for obtaining the preset value and the standard decomposition layer range are as follows:

[0128] Obtain at least one vibration frequency when the embedded computer is running normally in at least one time period;

[0129] Use wavelet packet decomposition for at least one vibration frequency to obtain the decomposition layer number and energy entropy value of at least one vibration frequency;

[0130] Draw a bar chart of the decomposition layer number and energy entropy value of the vibration frequency;

[0131] Based on the bar chart, find out the decomposition layer range corresponding to the concentrated distribution of the energy entropy value, denoted as the standard decomposition layer range, and the average entropy value corresponding to the concentrated distribution of the energy entropy value is denoted as the preset value.

[0132] It can be understood that by using wavelet packet decomposition for the vibration signal of the embedded computer, vibration frequencies in different frequency bands are obtained, and then using the entropy value calculation formula, the energy entropy values of different decomposed vibration signals are calculated. Since the vibration signal when the embedded computer is running normally is regular, if the increase in the energy entropy of the obtained vibration signal is too large, it indicates that the embedded computer may have a fault. If the decomposition layer number of the wavelet packet decomposition of the vibration signal is too large or too small, it will affect the accuracy of the energy entropy value. Therefore, it is necessary to determine the decomposition layer range.

[0133] Refer to Figure 5 As shown, based on the prediction of the remaining life of the device, dynamically adjusting the maintenance period by using reinforcement learning includes:

[0134] Obtain the average daily usage time and average daily usage intensity of the user using the embedded computer;

[0135] Simulate the daily operation of the high-precision digital model in the virtual world and accelerate the time flow rate in the virtual time;

[0136] If the data simulation driver cannot use the software normally, it is determined that the data simulation driver is damaged;

[0137] Obtain the running time of the high-precision digital model from running to damage, denoted as the remaining life;

[0138] Obtain the daily usage of users, update the average daily usage time and usage intensity of users using the embedded computer every day, and obtain the remaining updated life of the embedded computer;

[0139] Based on the remaining life of the embedded computer, calculate the initial maintenance cycle of the embedded computer using the cycle calculation formula;

[0140] During the initial maintenance cycle, determine whether the updated remaining life is greater than the updated remaining life of the previous day. If so, add one day to the initial maintenance cycle. If not, subtract one day from the initial maintenance cycle;

[0141] The cycle calculation formula is: ,

[0142] In the formula, T is the initial maintenance cycle, A is the remaining life of the embedded computer, and B is the average maintenance times of the embedded computer.

[0143] It can be explained that the remaining life of the embedded computer is obtained based on a high-precision digital model, and the maintenance cycle is adjusted in real time based on the remaining life of the embedded computer, which solves the problems of increased probability of failure of embedded computers with too short maintenance cycles and too high costs for too long maintenance cycles. Incorporating the average maintenance times of the embedded computer into the calculation of the maintenance cycle of the embedded computer can reduce the inconvenience caused by too large or too small maintenance times with the universality of statistical data.

[0144] Refer to Figure 6 As shown, the human-machine collaborative decision-making module deploys a lightweight CNN model at the device end to realize local real-time analysis of the vibration spectrum, and obtains machine evaluations including:

[0145] Deploy a CNN model on the embedded computing to obtain the vibration signal of the embedded computer;

[0146] Use the CNN model to analyze the vibration signal of the embedded computer in real time and convert it into an image, denoted as the real-time signal;

[0147] Use artificial intelligence to analyze and store the vibration images of at least one embedded computer running normally, complete machine learning, and obtain the normal signal;

[0148] Artificial intelligence analyzes and compares the real-time signal with the normal signal to obtain the similarity between the real-time signal and the normal signal, and the similarity is expressed as a percentage;

[0149] The difference between 1 and the similarity is denoted as the machine evaluation;

[0150] The output of the CNN model is a feature vector processed by a convolutional layer, a pooling layer, and a fully connected layer.

[0151] It can be explained that by leveraging the powerful computing power of artificial intelligence to compare the similarity between the real-time signal and the normal signal of the vibration image, the advantages of artificial intelligence are fully utilized. An embedded computer operating normally has a definite vibration frequency. Representing the machine evaluation by the similarity between the real-time signal and the normal signal can improve the accuracy of the machine evaluation.

[0152] Referring to Figure 7 as shown, based on the high-precision digital model, the human evaluation includes:

[0153] Obtain the chip information, hardware information, and software operation status of the high-precision digital model;

[0154] Invite at least one expert to analyze the chip information, hardware information, and software operation status to obtain the evaluation of the embedded computer;

[0155] Analyze the authority and academic achievements of at least one expert, and use the analytic hierarchy process to obtain the analysis weight of each expert;

[0156] Calculate the sum of the products of the evaluations of at least one expert and their analysis weights to obtain the human evaluation.

[0157] It can be explained that by summarizing the evaluations of multiple experts and assigning different weights to each expert based on their authority and academic achievements, the situation of a single expert's misjudgment is avoided, greatly improving the accuracy of the human evaluation. The advantage of the human evaluation is to judge the probability of failure based on the rich experience of the experts.

[0158] Referring to Figure 8 as shown, determine whether the machine evaluation is greater than the preset value. If so, calculate the human evaluation;

[0159] The initial weight of the machine evaluation is equal to the initial weight of the human evaluation;

[0160] Multiply the sum of the machine evaluation plus 1 and the initial weight of the machine evaluation to obtain the weight of the machine evaluation;

[0161] Record the difference between 1 and the weight of the machine evaluation as the weight of the human evaluation;

[0162] The sum of the product of the weight of the machine evaluation and the machine evaluation and the product of the weight of the human evaluation and the human evaluation is the human-machine collaborative evaluation;

[0163] Obtain the quantity and priority of the components for the maintenance of the embedded computer;

[0164] Based on the quantity and priority of the components for the maintenance of the embedded computer, divide them into several maintenance levels;

[0165] Divide the human-machine collaborative evaluation into several interval segments, and each interval segment corresponds to a maintenance level from small to large;

[0166] Determine the interval to which the human-machine collaborative evaluation belongs and execute the corresponding maintenance level;

[0167] The steps for obtaining the preset value are as follows:

[0168] Count the number of times a failure occurs in at least one embedded computer evaluated by the machine;

[0169] Select the smallest machine evaluation with a failure count greater than 1 as the preset value.

[0170] It can be explained that human-machine collaboration combines the accurate numbers obtained from the powerful computing power of machine evaluation with the rich expert experience of human evaluation, greatly improving the accuracy of predicting the failure probability of embedded mobile devices. By dividing the maintenance level based on human-machine collaborative evaluation, it avoids the waste of human resources caused by using too high a maintenance level for too low a human-machine collaboration.

[0171] Furthermore, this solution also proposes a storage medium on which a computer-readable program is stored. When the computer-readable program is called, it executes the above-mentioned digital-twin-based predictive maintenance control system for embedded computers.

[0172] It can be understood that the storage medium can be a magnetic medium, such as a floppy disk, a hard disk, or a magnetic tape; an optical medium, such as a DVD; or a semiconductor medium, such as a solid-state drive (SSD).

[0173] In summary, the advantages of the present invention are as follows: The present invention proposes a digital-twin-based predictive maintenance control system for embedded computers, which uses digital twins to build a high-precision digital model of the embedded computer and constructs a one-to-one restored model in the virtual world, helping to achieve cost-free simulation operation and improve the prediction accuracy. The fault anomaly detection module can timely detect the faults of the embedded computer and avoid the situation where the embedded computer fails during maintenance and cannot be maintained. The predictive equipment maintenance module uses the high-precision digital model to determine the predicted life of the embedded computer, thereby dynamically adjusting the maintenance cycle, solving the problems of increased failure probability for embedded computers with too short a maintenance cycle and too high a cost for too long a maintenance cycle. The human-machine collaborative decision-making module combines the accurate prediction of the computer and the empirical prediction of humans, greatly improving the accuracy of fault probability prediction.

[0174] The basic principles, main features and advantages of the present invention have been shown and described above. Those skilled in the art should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification is only the principle of the present invention. Without departing from the spirit and scope of the present invention, various changes and improvements will occur to the present invention, and all these changes and improvements fall within the scope of the present invention claimed. The scope of protection required by the present invention is defined by the appended claims and their equivalents.

Claims

1. An embedded computer predictive maintenance control system based on digital twins, characterized in that, Including: A data fusion modeling module, which sets sensors to monitor the temperature and vibration data of the embedded computer in real time, uploads the data to the cloud, and uses digital twins to build a high-precision digital model; A fault and anomaly detection module, which, based on the high-precision digital model, evaluates the change of the chip's junction temperature, sets the warning critical point of the junction temperature. If the chip's junction temperature exceeds the warning critical point, it issues a warning. Based on the high-precision digital model, it uses wavelet packet decomposition to extract the energy entropy value of the vibration signal, and judges whether the energy entropy value exceeds the preset value. If so, it judges that the embedded computer has an anomaly; A predictive equipment maintenance module, which dynamically adjusts the maintenance cycle based on the prediction of the remaining life of the equipment using reinforcement learning; A human-machine collaborative decision-making module, which deploys a lightweight CNN model at the equipment end to realize local real-time analysis of the vibration spectrum, obtains a machine evaluation, obtains a human evaluation based on the high-precision digital model, combines the machine evaluation and the human evaluation, obtains a human-machine collaborative evaluation, and conducts maintenance control; The step of using wavelet packet decomposition to extract the energy entropy value of the vibration signal and judging whether the energy entropy value exceeds the preset value. If so, judging that the embedded computer has an anomaly includes: Obtaining the vibration signal of the high-precision digital model; Using wavelet packet decomposition on the vibration signal to obtain at least one divided frequency band; Judging whether the decomposition layer number of the wavelet packet decomposed vibration signal belongs to the standard decomposition layer number interval. If so, calculate the energy entropy value of the vibration signal using the entropy value calculation formula. If not, obtain the vibration signal of the high-precision digital model again and use wavelet packet decomposition on the vibration signal; Judging whether the energy entropy value exceeds the preset value. If so, judging that the embedded computer has an anomaly; The entropy value calculation formula is as follows: , where H is the entropy energy, is the ratio of the number of the i-th frequency band to the total number of all frequency bands, and n is the number of frequency bands after wavelet packet decomposition of the vibration signal; The wavelet packet decomposition is to let the signal pass through a series of filters with different center frequencies but the same bandwidth to obtain signals in at least one frequency band; The steps for obtaining the preset value and the standard decomposition layer number interval are as follows: Obtaining at least one vibration frequency when the embedded computer is running normally in at least one time period; Using wavelet packet decomposition on at least one vibration frequency to obtain the decomposition layer number and energy entropy value of at least one vibration frequency; Drawing a bar chart of the decomposition layer number and energy entropy value of the vibration frequency; Based on the bar chart, find the decomposition layer number interval corresponding to the concentrated distribution of the energy entropy value, denoted as the standard decomposition layer number interval, and the average entropy value corresponding to the concentrated distribution of the energy entropy value is denoted as the preset value.

2. The predictive maintenance control system of the embedded computer based on digital twin according to claim 1, characterized in that The step of setting sensors to monitor the temperature and vibration data of the embedded computer in real time, uploading the data to the cloud, and using digital twins to build a high-precision digital model includes: Obtaining the appearance size of the embedded computer and building a three-dimensional solid model of the embedded computer; Obtaining the hardware component model, usage time, and software composition of the embedded computer, and building a preliminary high-precision digital model of the embedded computer based on the three-dimensional solid model of the embedded computer; Obtaining the temperature, vibration, and electrical parameter data during the operation of the embedded computer, denoted as real parameters; Simulate the normal operation of the embedded computer using the preliminary high-precision digital model, record the temperature, vibration, and electrical parameter data of the high-precision digital model, and denote it as simulation parameters; Obtain the normal fluctuation range of each parameter during the normal operation of the embedded computer, calculate the length of the normal fluctuation range of each parameter, and denote half of the length of the normal fluctuation range of each parameter as the preset fluctuation value of each parameter; Compare the real parameters with the simulation parameters, calculate the difference between the real parameters and the simulation parameters, list each parameter as a target parameter one by one, and determine whether the difference of the target parameter is greater than the preset fluctuation value of the target parameter. If so, denote the target parameter as an influencing parameter; Adjust the hardware model, usage time, and software composition of the high-precision digital model, and record the changes in each parameter of the high-precision digital model; Quantify the adjusted hardware model and usage time of the high-precision digital model into numbers according to the release year of the model and the length of the usage time, and denote it as the independent variable parameter. Quantify the degree of change of each parameter into numbers, and denote it as the dependent variable parameter; Based on the changes in each parameter of the high-precision digital model, calculate the Pearson correlation coefficient between the independent variable parameter and the dependent variable parameter to obtain the correlation matrix between the independent variable parameter and the dependent variable parameter; Compare the difference between the influencing parameter and the real parameter, and determine the adjustment degree of the independent variable parameter corresponding to the influencing parameter based on the product of the Pearson coefficient of the correlation matrix of the influencing parameter and the difference; Summarize the adjustment degrees of the independent variable parameters corresponding to at least one influencing parameter to obtain the adjustment details of the hardware and software; Based on the obtained adjustment details of the hardware and software, adjust the preliminary high-precision digital model to obtain the high-precision digital model; The Pearson correlation coefficient is a statistic used to measure the linear correlation degree between two continuous variables, and its value range is from -1 to 1.

3. The predictive maintenance control system of an embedded computer based on digital twin according to claim 2, characterized in that, Regarding the evaluation of the change in the junction temperature of the evaluation chip, setting the junction temperature warning critical point includes: Obtain the temperature and air flow of the environment where the embedded computer is located; Use the data display in the high-precision digital model to obtain the average temperature at the chip node in the high-precision digital model, and denote it as the junction temperature of the chip; Obtain the usage time and load of the commonly used software used by the user for the embedded computer; Obtain the junction temperature when the user uses all the commonly used software simultaneously, and denote it as the maximum junction temperature; Evaluate the aging speed and performance of the chip when the maximum junction temperature is exceeded, and draw a change image of the junction temperature and the aging speed and performance of the chip; Find a point with the smallest derivative on the image, and denote the corresponding junction temperature as the warning critical point.

4. The predictive maintenance control system for an embedded computer based on digital twin according to claim 3, wherein Regarding the prediction based on the remaining life of the device, adopting reinforcement learning to dynamically adjust the maintenance cycle includes: Obtain the average daily usage time and average daily usage intensity of the user using the embedded computer; Simulate the daily operation of the high-precision digital model in the virtual world and accelerate the time flow rate in the virtual time; If the data simulation driver cannot use the software normally, it is determined that the data simulation driver is damaged; Obtain the running time of the high-precision digital model from running to damage, and denote it as the remaining life; Obtain the daily usage situation of the user, update the average daily usage time and usage intensity of the user using the embedded computer daily, and obtain the updated remaining life of the embedded computer; Based on the remaining life of the embedded computer, calculate the initial maintenance cycle of the embedded computer using the cycle calculation formula; During the initial maintenance cycle, determine whether the updated remaining life is greater than the updated remaining life of the previous day. If so, add one day to the initial maintenance cycle. If not, subtract one day from the initial maintenance cycle; The period calculation formula is as follows: , In the formula, T is the initial maintenance cycle, A is the remaining life of the embedded computer, and B is the average maintenance times of the embedded computer.

5. The predictive maintenance control system of the embedded computer based on digital twin according to claim 4, characterized in that, The human-machine collaborative decision-making module deploys a lightweight CNN model at the device end to realize local real-time analysis of the vibration spectrum, and the machine evaluation obtained includes: Deploy the CNN model on the embedded computer to obtain the vibration signal of the embedded computer; Use the CNN model to analyze the vibration signal of the embedded computer in real time and convert it into an image, denoted as the real-time signal; Use artificial intelligence to analyze and store the vibration images of at least one embedded computer running normally, complete machine learning, and obtain the normal signal; Artificial intelligence analyzes and compares the real-time signal with the normal signal to obtain the similarity between the real-time signal and the normal signal, and the similarity is expressed as a percentage; The difference between 1 and the similarity is denoted as the machine evaluation; The output of the CNN model is the feature vector processed by the convolutional layer, pooling layer, and fully connected layer.

6. The predictive maintenance control system of an embedded computer based on digital twin according to claim 5, characterized in that, The human evaluation obtained based on the high-precision digital model includes: Obtain the chip information, hardware information, and software running status of the high-precision digital model; Invite at least one expert to analyze the chip information, hardware information, and software running status to obtain the evaluation of the embedded computer; Analyze the authority and academic achievements of at least one expert, and use the analytic hierarchy process to obtain the analysis weight of each expert; Calculate the product sum of the evaluation of at least one expert and the analysis weight to obtain the human evaluation.

7. The predictive maintenance control system of the embedded computer based on digital twin according to claim 6, characterized in that, Merge the machine evaluation and the human evaluation to obtain the human-machine collaborative evaluation, and perform maintenance control including: Judge whether the machine evaluation is greater than the preset value. If so, calculate the human evaluation; The initial weight of the machine evaluation is equal to the initial weight of the human evaluation; Multiply the sum of the machine evaluation plus 1 by the initial weight of the machine evaluation to obtain the weight of the machine evaluation; The difference between 1 and the weight of the machine evaluation is denoted as the weight of the human evaluation; The sum of the product of the weight of the machine evaluation and the machine evaluation and the product of the weight of the human evaluation and the human evaluation is the human-machine collaborative evaluation; Obtain the quantity and priority of the components for the maintenance of the embedded computer; Based on the quantity and priority of the maintenance components of the embedded computer, divide them into several maintenance levels; Divide the human-machine collaborative evaluation into several intervals, and each interval corresponds to the maintenance level from small to large; Judge the interval to which the human-machine collaborative evaluation belongs and execute the corresponding maintenance level; The steps to obtain the preset value are as follows: Count the number of failures of at least one embedded computer with machine evaluation; Select the smallest machine evaluation with the number of failures greater than 1 as the preset value.

Images