Industrial equipment predictive maintenance method and system based on edge computing

Abstract

The invention provides an industrial equipment predictive maintenance method and system based on edge computing, and the method comprises the steps: carrying out the on-site collection, cleaning, fusion, storage and analysis of mass state monitoring data generated in the operation process of industrial equipment through an edge node disposed at an industrial site, according to the method, multi-dimensional features capable of reflecting equipment health conditions are extracted, comprehensive health indexes are constructed, health state evaluation is performed through an equipment fault early warning model, and an equipment maintenance strategy is dynamically generated based on a model output result to guide maintenance personnel to overhaul or replace the equipment in advance, so that the maintenance efficiency is improved. The fault occurrence probability and the influence range are furthest reduced. According to the method, the edge computing technology is adopted, the data transmission time delay and the bandwidth pressure can be remarkably reduced, the real-time performance of fault early warning and response is improved, and the data safety is enhanced.

Description

Technical Field

[0001] The present invention belongs to the technical field of industrial equipment maintenance, and in particular, relates to a method and system for predictive maintenance of industrial equipment based on edge computing. Background Art

[0002] Industrial equipment is an important material basis for modern industrial production and manufacturing. The normal operation of industrial equipment directly affects the production efficiency, product quality and economic benefits of enterprises. However, industrial equipment is usually composed of multiple complex subsystems, with many components, complex structures and harsh working conditions, which inevitably lead to various failures and abnormalities during operation, causing huge losses to the production and operation of enterprises. Predictive maintenance focuses on equipment status monitoring and fault warning. It uses sensors to collect key parameters such as vibration, temperature, pressure, and current of equipment in real time, and uses intelligent algorithms such as machine learning and big data analysis to build an indicator system that can reflect the health status of equipment, predict its degradation trend and remaining life, and perform maintenance or replacement of equipment at the best time to minimize failures and unplanned downtime. However, the implementation of predictive maintenance in actual industrial scenarios still faces many technical challenges.

[0003] In recent years, edge computing, as an emerging distributed computing model, has provided new ideas for solving the above problems. Applying edge computing to predictive maintenance of industrial equipment can fully tap the value of equipment status data, timely detect equipment anomalies, warn of potential failures, reduce the dependence of data transmission on the network, and ensure information security. However, at present, the industrial field still has the following problems on how to deeply integrate edge computing with predictive maintenance to form a standardized and systematic technical solution and implementation path:

[0004] (1) Most of them rely on cloud computing. The health monitoring and fault warning of industrial equipment rely on continuous data transmission to the cloud, which leads to high data transmission costs and cannot achieve real-time fault warning. When edge computing is not effectively applied, the real-time performance is poor and it is impossible to respond to changes in equipment status in a timely manner;

[0005] (2) Usually only a single sensor data, such as vibration or temperature, is used to monitor equipment health, which fails to fully utilize the advantages of multi-source heterogeneous data, resulting in inaccurate characterization of the degradation process of industrial equipment and a lack of a comprehensive equipment health assessment system;

[0006] (3) Fault warning mechanisms usually use simple thresholds or rules and lack adaptive adjustment capabilities. It is difficult to perform personalized management based on the actual usage and failure modes of different devices, resulting in low maintenance efficiency.

[0007] Therefore, we need to develop a predictive maintenance method and system for industrial equipment based on edge computing, which can optimize the health monitoring and maintenance of industrial equipment through edge computing, data fusion and adaptive early warning mechanism. Summary of the invention

[0008] The purpose of the present invention is to provide an industrial equipment predictive maintenance method and system based on edge computing to solve the problems mentioned in the above background technology that the existing industrial equipment predictive maintenance solutions lack real-time monitoring, are difficult to comprehensively evaluate equipment health, and the early warning mechanism is not intelligent enough.

[0009] To achieve the above object, the present invention provides an industrial equipment predictive maintenance method based on edge computing, and the method is specifically as follows:

[0010] Through edge computing nodes, time domain features, frequency domain features, and time-frequency domain features that reflect the health status of industrial equipment are extracted from the preprocessed status monitoring data, and feature selection and feature dimension reduction are performed to obtain multi-dimensional features;

[0011] The vibration intensity index, temperature anomaly index, current fluctuation index and sound anomaly index of the industrial equipment are calculated based on the multidimensional features, and the index aggregation is realized by using the OWA operator based on the order relationship to construct the comprehensive health index of the industrial equipment, and the weight of each sub-item in the comprehensive health index is dynamically adjusted by an adaptive optimization algorithm based on a sliding time window;

[0012] The comprehensive health index is evaluated through the equipment failure early warning model to obtain the health status level and failure risk assessment results as early warning information and output, and the failure probability of the industrial equipment within a limited time in the future is obtained through probability calculation;

[0013] For the multiple devices in the industrial equipment whose failure probability exceeds the preset warning probability, determining the maintenance priority and maintenance time window of the multiple devices according to the early warning information, the magnitude of the failure probability and the importance of the devices;

[0014] Considering the dual impact of current vibration level and accumulated damage on the health of the industrial equipment, the vibration intensity index is defined as:

[0015]

[0016] in, is the vibration intensity index, Indicates the number of samples in the current sampling period k. The root mean square value of the vibration velocity collected by the vibration sensor, is the vibration velocity RMS threshold, is the total number of vibration sensors, L is the time window length, and are weight coefficients reflecting the importance of instantaneous vibration and cumulative vibration, respectively, and ; It represents the root mean square value of the vibration velocity collected by the i-th vibration sensor in the j-th sampling period. The time window L is introduced to smooth the short-term fluctuations of the vibration intensity and obtain a more robust evaluation result.

[0017] Based on the above scheme, based on the multi-dimensional features, the vibration signal feature vector, temperature signal feature vector, current signal feature vector and sound signal feature vector of the industrial equipment are extracted, and based on these extracted feature vectors, the vibration intensity index, the temperature anomaly index, the current fluctuation index and the sound anomaly index are calculated respectively.

[0018] Based on the above scheme, the temperature anomaly index is calculated based on the temperature signal feature vector, and the formula is:

[0019]

[0020] in, Indicates the number of samples in the current sampling period k. The average temperature of the measuring points, is the normal temperature value, For the allowable temperature, is the maximum allowable temperature rise over two sampling periods, is the arithmetic mean of all temperature measurement points in the kth cycle, is the reference variance when the temperature distribution is uniform, is the total number of temperature sensors; , , is the weight coefficient; the temperature anomaly index The calculation comprehensively considers three abnormal factors: the temperature absolute value exceeds the standard, the temperature change rate exceeds the standard, and the temperature spatial distribution is uneven.

[0021] Based on the above scheme, the current fluctuation index is calculated based on the current signal characteristic vector as a quantitative index reflecting the degree of current imbalance. The calculation formula is as follows:

[0022]

[0023] in, The first The current value of the sampling point, For the The average value of the output current in the current cycle, for The total average value of the output current of each circuit, is the rated current, , It is the weight coefficient of the two indicators. The first item reflects the severity of the instantaneous current exceeding the standard, and the second item reflects the relative balance of multiple output currents. The worse the balance, the higher the risk of equipment failure.

[0024] Based on the above scheme, the sound abnormality index is calculated based on the sound signal feature vector, wherein the frequency band energy distribution dispersion is in the form of the ratio of the standard deviation to the mean to ensure that the index value falls within the interval [0, 1], and the calculation formula is as follows:

[0025]

[0026] in, It represents the sound pressure level of the ith sound measurement point in the current kth sampling period. For the current cycle The energy percentage of the frequency band, is the historical average of the energy percentage of the frequency band, is the total number of frequency bands, that is: Indicates the reference value / threshold of the sound pressure level, is the weight coefficient, and .

[0027] Based on the above scheme, the OWA operator based on the order relationship is used to realize the index aggregation as the comprehensive health index, and the calculation formula is as follows:

[0028]

[0029] in, For the The first of the vibration intensity index, the temperature anomaly index, the current fluctuation index and the sound anomaly index of a sampling period Large value, is the corresponding OWA weight, and The comprehensive health index , indicating that the industrial equipment is in The overall degradation degree of the sampling period, and The larger it is, the worse the device health is and the higher the risk of failure.

[0030] Based on the above solution, the adaptive optimization algorithm based on sliding time window is specifically as follows:

[0031] Define a sliding time window , the window length is W, which is used to achieve adaptive optimization of the comprehensive health index;

[0032] In each time window, the known health status labels of the industrial equipment, such as normal, slight wear, and severe wear, are used to construct Mapping model to state categories: ,in represents the real health status of the industrial equipment at the end of the current window k, is the classifier function, are the classifier parameters.

[0033] Based on the above scheme, the probability calculation is specifically as follows:

[0034] Assume that the degradation process of the industrial equipment from new to failure is governed by Distribution, its probability density function is: ,in The operating time of the industrial equipment; , is the shape parameter; , is the scale parameter;

[0035] In the future The calculation formula for the failure probability of the industrial equipment within the time period is as follows:

[0036]

[0037] in, , for the future The estimated increase in the comprehensive health indicators mentioned in the time period, Through the historical health indicator series Obtained by trend extrapolation estimation.

[0038] On the other hand, the present invention provides an industrial equipment predictive maintenance system based on edge computing, which is used to execute an industrial equipment predictive maintenance method based on edge computing described in the aforementioned scheme, and is composed of three parts: a field equipment layer, an edge computing layer, and a cloud service layer, wherein:

[0039] The field equipment layer includes various types of industrial equipment, and vibration, temperature, current, and sound sensors are arranged on the industrial equipment to collect status monitoring data of the industrial equipment in real time;

[0040] The edge computing layer is composed of several edge computing nodes deployed on site or near the industrial equipment, each of which is equipped with a data acquisition module, a data preprocessing module, a feature extraction module, a health assessment module and a fault warning module;

[0041] The cloud service layer is responsible for receiving device health status data and warning information from the edge computing layer for global device status monitoring and statistical analysis, and continuously optimizing and updating the health assessment and warning algorithms of the edge computing nodes in combination with massive historical data and mechanism models.

[0042] Compared with the prior art, the present invention has the following advantages and effects:

[0043] (1) Real-time monitoring of equipment status and fault warning are achieved through edge nodes. Edge computing reduces data transmission costs, improves the real-time nature of fault warnings, and achieves cloud-edge collaboration, taking into account both local real-time processing and global optimization.

[0044] (2) Extracting equipment status features from multiple dimensions, a fusion method for multi-source heterogeneous data including vibration, temperature, current, and sound is proposed to accurately characterize the equipment degradation process and construct a comprehensive health index;

[0045] (3) An adaptive multi-level fault warning mechanism is designed, and risk assessment is performed in combination with a probability model. Differentiated maintenance strategies are formulated for different fault risk levels, thereby optimizing maintenance resource allocation. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following is a brief introduction to the drawings required for the specific embodiments or the description of the prior art. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn according to the actual scale.

[0047] Figure 1 It is a flow chart of a method for predictive maintenance of industrial equipment based on edge computing provided by an embodiment of the present invention;

[0048] Figure 2 This is a structural diagram of an industrial equipment predictive maintenance system based on edge computing provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0049] In order to more clearly explain the purpose, technical solutions and advantages of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. The example implementation methods can be implemented in various forms and should not be understood as being limited to the examples described herein. On the contrary, these implementation methods are provided to make the present invention more comprehensive and complete, and to fully convey the concepts of the example implementation methods to those skilled in the art.

[0050] In addition, the described features, structures or characteristics may be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present invention. However, those skilled in the art will appreciate that the technical solution of the present invention can be practiced without one or more of the specific details, or other methods, components, devices, steps, etc. may be adopted. In other cases, known methods, devices, implementations or operations are not shown or described in detail to avoid blurring various aspects of the present invention.

[0051] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities may be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0052] The flowcharts shown in the accompanying drawings are only exemplary and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps can be decomposed, and some operations / steps can be combined or partially combined, so the actual execution order may change according to actual conditions.

[0053] The present invention is described in detail below in conjunction with specific embodiments:

[0054] Example 1

[0055] As attached Figure 1 As shown, Embodiment 1 of the present invention provides an industrial equipment predictive maintenance method based on edge computing, and the specific steps of the method are as follows:

[0056] Step S1: setting a reasonable data sampling frequency, collecting the state monitoring data of industrial equipment in real time through a variety of physical quantity sensors, including vibration, temperature, current, and noise; performing preprocessing operations such as denoising, interpolation, and normalization on the state monitoring data to improve data quality;

[0057] Step S2: extracting time domain features, frequency domain features, and time-frequency domain features reflecting the health status of the industrial equipment from the preprocessed status monitoring data, and performing feature selection and feature dimensionality reduction to obtain multidimensional features;

[0058] Step S3: constructing a comprehensive health index of the industrial equipment based on the multi-dimensional features to characterize the degree of degradation of the equipment, and dynamically adjusting the weight of the comprehensive health index through an adaptive optimization algorithm;

[0059] Step S4: Establishing an equipment failure early warning model, evaluating the comprehensive health index, timely discovering the abnormality and failure type of the industrial equipment, and estimating the failure risk level;

[0060] Step S5: Based on the fault type, fault risk level, and importance of the industrial equipment, formulate equipment maintenance priority and maintenance time window.

[0061] Preferably, the multiple physical quantity sensors in step S1 are installed at key parts of the industrial equipment, including vibration sensors, temperature sensors, current sensors and sound sensors, etc.; wherein the sampling frequency of the vibration sensor is 2000Hz, the sampling frequency of the temperature sensor is 1Hz, the sampling frequency of the current sensor is 100Hz, and the sampling frequency of the sound sensor is 8000Hz;

[0062] Specifically, the multiple physical quantity sensors collect various monitoring data during the operation of the industrial equipment in real time, and transmit the data to the nearest edge computing node in real time.

[0063] Preferably, the preprocessing operation in step S1 is to perform data preprocessing on the status monitoring data after the edge computing node receives the status monitoring data; the preprocessing process includes data cleaning, denoising and standardization, specifically:

[0064] Data cleaning is to detect and remove abnormal points, outliers and duplicate values ​​in the data; denoising uses the wavelet transform method, selects the db4 wavelet basis function, performs a 4-level wavelet decomposition on the data, and removes high-frequency noise; standardization uses the Z-score method to convert the data into a standard normal distribution with a mean of 0 and a standard deviation of 1, which can be calculated using the following formula: ,in, represents the standardized data, Represents the original data, represents the mean of the original data, Represents the standard deviation of the original data.

[0065] Preferably, the extraction of the time domain features, frequency domain features, and time-frequency domain features in step S2 is to extract various feature parameters from three dimensions of time domain, frequency domain, and time-frequency domain through edge computing nodes:

[0066] Specifically, the time domain features include the mean , Standard Deviation , RMS , Peak , peak-to-peak , skewness Kurtosis , can be calculated by the following formulas, where Indicates the number of signal sampling points, Indicates The amplitude of the sampling points:

[0067] ;

[0068] ;

[0069] ;

[0070] ;

[0071] ;

[0072] ;

[0073] ;

[0074] Specifically, the frequency spectrum is obtained by performing a fast Fourier transform (FFT) on the time domain signal, and the characteristic parameters of the frequency spectrum are extracted to obtain the frequency domain features, including the main frequency , spectrum peak Mean and spectral variance , can be calculated by the following formulas, where: Indicates frequency The amplitude at:

[0075] ;

[0076] ;

[0077] ;

[0078] ;

[0079] Specifically, the time-frequency domain features are extracted by short-time Fourier transform (STFT) or wavelet packet transform (WPT) to obtain the distribution of energy in different frequency bands; assuming that the time-frequency spectrum is divided into frequency bands, define the frequency bands Energy for: ,in, The time spectrum is time The amplitude at the frequency, and Respectively represent The lower and upper frequencies of the frequency band, Indicates the total duration of the signal; defines the total energy for: , then the frequency band can be calculated Energy percentage : 00% .

[0080] Preferably, the step S3 of constructing the health index of the industrial equipment is specifically:

[0081] Step S301: Based on the multi-dimensional features, obtain the vibration signal feature vector of the industrial equipment , temperature signal feature vector , current signal characteristic vector And the sound signal feature vector , respectively expressed as:

[0082]

[0083]

[0084]

[0085]

[0086] Specifically, the above-mentioned feature vectors are transmitted to the health assessment module in real time from each edge computing node at a set frequency (such as once per minute) through wired methods such as industrial Ethernet and field bus, or wireless methods such as 5G and NB-IoT, as input for constructing comprehensive health indicators.

[0087] Step S302: Based on the vibration signal feature vector Calculate the vibration intensity index , considering the dual impact of current vibration levels and cumulative damage on the health of the industrial equipment, Defined as:

[0088]

[0089] in, Indicates the number of samples in the current sampling period k. The root mean square value of the vibration velocity collected by the vibration sensor, is the vibration velocity RMS threshold, is the total number of vibration sensors, L is the time window length, and are weight coefficients reflecting the importance of instantaneous vibration and cumulative vibration, respectively, and , represents the root mean square value of the vibration velocity collected by the ith vibration sensor in the jth sampling period; the time window L is introduced to smooth the short-term fluctuations of the vibration intensity and obtain a more robust evaluation result; the weight coefficient and The optimal value can be determined using expert experience or data-driven methods based on the equipment type and degradation law.

[0090] Step S303: Based on the temperature signal feature vector Calculating temperature anomaly index , the formula is:

[0091]

[0092] The temperature anomaly index The calculation of takes into account three abnormal factors: the temperature absolute value exceeding the standard, the temperature change rate exceeding the standard, and the uneven temperature spatial distribution; among them, Indicates the number of samples in the current sampling period k. The average temperature of the measuring points, is the normal temperature value, For the allowable temperature, is the maximum allowable temperature rise over two sampling periods, is the arithmetic mean of all temperature measurement points in the kth cycle, is the reference variance when the temperature distribution is uniform, is the total number of temperature sensors; weight coefficient , , Reflects the relative importance of the three abnormal factors, satisfying The specific value can be optimized and obtained through grey correlation analysis, information entropy principle and other methods according to the influencing mechanism of temperature anomaly and historical operation data of equipment.

[0093] Step S304: Based on the current signal characteristic vector Calculate current fluctuation index , as a quantitative indicator reflecting the degree of current imbalance, is calculated as follows:

[0094]

[0095] in, The first The current value of the sampling point, For the The average value of the output current in the current cycle, for The total average value of the output current of each circuit, is the rated current, , It is the weight coefficient of the two indicators. The first item reflects the severity of the instantaneous current exceeding the standard, and the second item reflects the relative balance of multiple output currents. The worse the balance, the higher the risk of equipment failure.

[0096] Step S305: Based on the sound signal feature vector Calculate the sound anomaly index , where the frequency band energy distribution dispersion is in the form of the ratio of the standard deviation to the mean to ensure that the index value falls within the [0, 1] interval. The calculation formula is as follows:

[0097]

[0098] in, It represents the sound pressure level of the ith sound measurement point in the current kth sampling period. For the current cycle The energy percentage of the frequency band, is the historical average of the energy percentage of the frequency band, is the total number of frequency bands, and the meanings of the other symbols are consistent with the previous steps, namely: Indicates the reference value / threshold of the sound pressure level, is the weight coefficient, and , temperature anomaly index in step and current fluctuation index The calculation method is consistent.

[0099] Step S306: The vibration intensity index , Temperature abnormality index , Current fluctuation index , Abnormal sound index Perform weighted fusion to form the comprehensive health index ;

[0100] Specifically, in order to overcome the limitations of the arithmetic mean and geometric mean methods, this embodiment uses the OWA (Ordered Weighted Averaging) operator based on the order relationship to achieve health indicator aggregation. The calculation formula is as follows:

[0101]

[0102] in, For the Four sub-indicators for each sampling period , , The Large value, is the corresponding OWA weight, and ;OWA weight The selection of is usually based on fuzzy language description operators, for example: "most" corresponds to ; "at least half" corresponds to ; Compared with simple linear weighting, the OWA operator can effectively avoid the compensation effect between sub-indicators and highlight the sensitivity to severe anomalies;

[0103] It should be noted that the comprehensive health index , indicating that the industrial equipment is in The overall degradation degree of the sampling period, and The larger it is, the worse the device health is and the higher the risk of failure.

[0104] Step S307: define a sliding time window , the window length is W, which is used to achieve adaptive optimization of the comprehensive health index;

[0105] Specifically, in each time window, the known health status labels of the industrial equipment, such as normal, slight wear, severe wear, etc., are used to construct Mapping model to state categories: ,in represents the real health status of the industrial equipment at the end of the current window k, is the classifier function, is the classifier parameter;

[0106] The classifier model can be selected from Logistic regression, decision tree, support vector machine, and the classification loss of a single sample is further defined as:

[0107]

[0108] in is the health status of the industrial equipment predicted by the classifier model; specifically, within a time window of length W, the cumulative classification loss is:

[0109]

[0110] in, , is the weight vector of the four sub-indicators; , is the weight vector of the OWA aggregation operator; is the parameter vector of the health status classifier, that is, the parameter vector of the classifier model; the goal of the adaptive optimization is to find the optimal parameters , so that the cumulative classification loss is minimized, the definition is as follows:

[0111]

[0112] The constraints are:

[0113]

[0114]

[0115] For example, the above adaptive optimization problem can be solved by gradient descent method, and the optimal parameters obtained are It is used for health assessment in the next time window to realize adaptive online update of the weight of the comprehensive health indicator.

[0116] Preferably, the evaluation of the comprehensive health index in step S4 is based on the comprehensive health index , judging the health status of the industrial equipment through the equipment failure early warning model, obtaining the health status level and the failure risk assessment result as early warning information and outputting them;

[0117] Specifically, a classification standard for the health status of an equipment is defined, and the equipment status is divided into five levels: normal, attention, warning, critical, and fault. The corresponding health indicator thresholds are , then the state determination rule for the industrial equipment is:

[0118]

[0119] in, Indicated in The health status of the industrial equipment in each sampling period, the threshold The value of can be determined based on factors such as equipment type, historical operation data, industry standards, etc., and usually requires a lot of simulation tests and expert reviews; for example, in this embodiment, the threshold is set to , after evaluating the current health status of the industrial equipment, the equipment fault warning module outputs a fault risk assessment result corresponding to the health status level;

[0120] Furthermore, considering that simple threshold comparison may cause sudden changes in warnings and false alarms, this embodiment introduces probability calculation based on the equipment degradation process into the equipment failure warning model to obtain a smoother and more robust warning effect; based on reliability theory, it is assumed that the degradation process of the industrial equipment from new to failure obeys Distribution, its probability density function is:

[0121]

[0122] in, The operating time of the industrial equipment; , is the shape parameter; , is a scale parameter; in practice, comprehensive health indicators are often used Alternative time As a characterizing variable of the degradation process;

[0123] Specifically, based on the comprehensive health index , through the probability calculation, we can get The failure probability of the industrial equipment within a certain period of time is calculated as follows:

[0124]

[0125] in, , for the future The estimated increase in the comprehensive health indicators mentioned in the time period, Through the historical health indicator series Trend extrapolation estimation can be obtained using simple moving average method, exponential smoothing method, grey prediction method and ARIMA time series model;

[0126] Furthermore, when the failure probability Exceeding the preset warning probability When the industrial equipment A failure is very likely to occur within a certain period of time, which will trigger a high-level warning and recommend shutdown for maintenance; the warning probability The reasonable value of needs to balance the false alarm rate and missed alarm rate of the early warning, and can be simulated and optimized with the help of tools such as precision-recall curve and ROC curve.

[0127] In more implementation cases, in order to better characterize the future degradation trend of the equipment, the remaining service life of the industrial equipment at the current health level is output of estimates; The reliability function of the distribution shows that from the current moment At the time of failure The probability distribution of the remaining life is:

[0128]

[0129] in, represents the value of the comprehensive health index at the current moment, ΔHI represents the expected increase in the comprehensive health index, represents the scale parameter, Represents the shape parameter.

[0130] make , solve That is, the remaining service life is obtained The median estimate of:

[0131]

[0132] in Indicates the comprehensive health index The average degradation rate can be obtained by fitting the historical trend; similarly, other quantiles can be estimated The remaining useful life of , thereby obtaining the confidence interval of RUL; in this embodiment, the remaining life of the industrial equipment is predicted, which provides a direct reference for formulating a maintenance plan.

[0133] Preferably, the formulation of equipment maintenance priority and maintenance time window in step S5 is based on the early warning information to generate a specific equipment maintenance plan, including maintenance priority sorting and maintenance time window optimization of the industrial equipment;

[0134] Specifically, for the failure probability in the industrial equipment Exceeding the preset warning probability multiple devices, according to the health status level, failure probability The size and importance of the equipment determine the priority of maintenance and define the first Maintenance priority index of each device for:

[0135]

[0136] in, Indicates The health status level of the device, with a value of 1-5, corresponding to the five levels of Normal, Attention, Warning, Critical, and Failure; Indicates Devices in the warning time scale The probability of failure within Indicates The economic loss caused by a failure of a piece of equipment can be measured by indicators such as average maintenance cost and downtime loss. Indicates The importance of each device in the production line is comprehensively evaluated based on its topological position in the system, connectivity, mission criticality and other factors; are the weight coefficients of the four influencing factors, which are set according to the specific application scenario and expert experience and meet the requirements .

[0137] It should be noted that the maintenance priority index Quantified the The greater the value, the more urgent the maintenance of the equipment needs to be arranged sooner. By sorting the values ​​from large to small, you can get a maintenance priority queue.

[0138] It should be noted that premature maintenance of equipment will waste the effective service life of the equipment, while late maintenance will lead to the risk of sudden equipment failure. The optimal maintenance time should be determined based on both service life and failure risk.

[0139] Specifically, the maintenance time window of each single device in the maintenance priority queue is optimized, and the maintenance time window is solved by a target programming method, and its mathematical model is:

[0140]

[0141] The constraints are: ;

[0142] Among them, the objective function It consists of two parts: Part I From the current moment Maintenance time The cumulative use benefit of the equipment during this period can be simply understood as the probability that the equipment can work normally within this time window; Part II Indicated in Failure losses avoided by timely maintenance, including is the average failure cost, for Equipment reliability at all times; The physical meaning is the difference between the service life benefit and the maintenance benefit, and the maximum value should be taken; the constraints include two aspects: one is the maintenance time The probability of failure The tolerable risk threshold must not be exceeded , It can be 10%~20%. The second is the actual maintenance time window. Should fall within the user-specified interval to reflect external factors such as business continuity requirements and resource constraints.

[0143] In this embodiment, by collecting equipment status data in real time and processing it in combination with edge computing, accurate monitoring of the equipment health status and fault warning are achieved. Specifically, by installing a variety of physical quantity sensors (such as vibration, temperature, current and sound sensors) at key parts of the equipment, various types of monitoring data are collected in real time, and the collected data are pre-processed by denoising, interpolation and normalization through edge computing nodes to ensure data quality; multi-dimensional features in the time domain, frequency domain and time-frequency domain are extracted from these pre-processed data, and a more accurate feature set is obtained through feature selection and dimensionality reduction; based on the extracted multi-dimensional features, a comprehensive health index is constructed to characterize the degree of degradation of the equipment, and the weight of each feature is dynamically adjusted through an adaptive optimization algorithm; by establishing an equipment fault warning model, the comprehensive health index is used to evaluate the health status of the equipment, detect anomalies in real time and predict potential fault types and risk levels; combined with the importance of the equipment and the risk assessment results, differentiated maintenance strategies are formulated according to different fault risk levels and equipment operating conditions, thereby optimizing the allocation of maintenance resources and reducing maintenance costs; this embodiment fully considers the interaction of various indicators in the equipment degradation process, can accurately judge the health status of the equipment, provide a scientific basis for equipment maintenance, and significantly improve the effect of predictive maintenance of industrial equipment.

[0144] Example 2

[0145] As attached Figure 2 As shown, Embodiment 2 of the present invention provides an industrial equipment predictive maintenance system 200 based on edge computing, which is mainly composed of three parts: a field equipment layer 201, an edge computing layer 202, and a cloud service layer 203, as follows:

[0146] The field equipment layer 201 includes various types of industrial equipment, such as machine tools, pumps, fans, and compressors. Vibration, temperature, current, and sound sensors are arranged on the industrial equipment to collect status monitoring data of the industrial equipment in real time.

[0147] The edge computing layer 202 is composed of a number of edge computing nodes deployed on site or near the industrial equipment, each of which is equipped with a data acquisition module, a data preprocessing module, a feature extraction module, a health assessment module and a fault warning module;

[0148] Preferably, the data acquisition module is responsible for receiving the multi-source heterogeneous status monitoring data; the data preprocessing module is used to clean, denoise and standardize the status monitoring data; the feature extraction module is responsible for extracting the time domain, frequency domain and time-frequency domain characteristics of the equipment status; the health assessment module is used to construct the comprehensive health index of the industrial equipment; the fault warning module is used to determine the fault level and issue a warning signal; each edge computing node interacts with the field device layer 201 and the cloud service layer 203 through wired networks such as industrial Ethernet and fieldbus or wireless networks such as 5G and NB-IoT.

[0149] The cloud service layer 203 is deployed in a remote data center or cloud platform, and includes subsystems such as equipment asset management, big data analysis, machine learning training, and expert knowledge base; the cloud service layer 203 receives equipment health status data and warning information from the edge computing layer 202, which is used for global equipment status monitoring and statistical analysis on the one hand, and on the other hand, it combines massive historical data and mechanism models to continuously optimize and update the health assessment and warning algorithms of the edge computing nodes. The optimized algorithm model is then sent to the edge computing nodes to guide local data processing and decision-making. At the same time, the cloud service layer 203 can also provide on-site maintenance personnel with remote diagnosis, plan recommendation, spare parts management, expert consultation and other services for the industrial equipment.

[0150] This embodiment realizes the full process closed loop of predictive maintenance of industrial equipment under the system framework of end-edge-cloud collaboration; the device layer is responsible for the real-time perception of massive data, the edge layer is responsible for local processing and real-time decision-making of data, and the cloud layer is responsible for the accumulation and precipitation of knowledge and the iterative optimization of algorithms. The three layers work together and cooperate closely, which not only solves the efficiency problem of massive data processing, but also takes into account the advantages of cloud data aggregation. At the same time, the topological structure of the end-edge cloud is also very robust and flexible, and the number and computing power ratio of nodes at each layer can be flexibly adjusted according to application scenarios and business needs. Furthermore, the system of this embodiment also fully considers the needs of data security and privacy protection in the industrial Internet environment. The original data is encrypted and desensitized at the edge layer before being uploaded to the cloud, which can avoid the leakage of sensitive data and meet the needs of cloud data analysis. The edge node and the cloud platform use encrypted channels for data transmission, which can effectively resist malicious attacks on the network side.

[0151] In summary, this embodiment builds a complete set of industrial equipment predictive maintenance system based on edge computing technology. The system conforms to the technical trend of "edge intelligence" of the industrial Internet. While greatly improving the efficiency and refinement of equipment maintenance, it also strikes a good balance between the timeliness of data processing and the globality of data aggregation, providing an important enabling tool for the digital transformation and upgrading of industrial enterprises. Figure 2The end-edge-cloud collaborative architecture shown also provides a universal reference template for applications in other smart manufacturing scenarios.

[0152] It should be noted that the attached Figure 2 The system architecture shown is one implementation method, not the only one. In actual applications, the system architecture can be appropriately tailored and adjusted according to factors such as the type of industrial equipment, the degree of enterprise informatization, and network conditions. For example, for equipment with simple structure and low risk, the edge layer can be eliminated, and the equipment can directly upload data to the cloud for analysis; for occasions with extremely high data security requirements, the cloud service layer can also be eliminated, and data closed-loop processing can only be performed at the local edge layer. The openness and modular design of the system give users the flexibility of secondary development.

[0153] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. This application is intended to cover any modification, use or adaptation of the present invention, which follows the general principles of the present invention and includes common knowledge or customary techniques in the art that are not disclosed by the present invention. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present invention are indicated by the claims. It should be understood that the present invention is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. A predictive maintenance method for industrial equipment based on edge computing, characterized in that: include: Through edge computing nodes, time domain features, frequency domain features, and time-frequency domain features that reflect the health status of industrial equipment are extracted from the preprocessed status monitoring data, and feature selection and feature dimension reduction are performed to obtain multi-dimensional features; The vibration intensity index, temperature anomaly index, current fluctuation index and sound anomaly index of the industrial equipment are calculated based on the multidimensional features, and the index aggregation is realized by using the OWA operator based on the order relationship to construct the comprehensive health index of the industrial equipment, and the weight of each sub-item in the comprehensive health index is dynamically adjusted by an adaptive optimization algorithm based on a sliding time window; The comprehensive health index is evaluated through the equipment failure early warning model to obtain the health status level and failure risk assessment results as early warning information and output, and the failure probability of the industrial equipment within a limited time in the future is obtained through probability calculation; For the multiple devices in the industrial equipment whose failure probability exceeds the preset warning probability, determining the maintenance priority and maintenance time window of the multiple devices according to the early warning information, the magnitude of the failure probability and the importance of the devices; Considering the dual impact of current vibration level and accumulated damage on the health of the industrial equipment, the vibration intensity index is defined as: ,in, is the vibration intensity index, Indicates the number of samples in the current sampling period k. The root mean square value of the vibration velocity collected by the vibration sensor, is the vibration velocity RMS threshold, is the total number of vibration sensors, L is the time window length, and are weight coefficients reflecting the importance of instantaneous vibration and cumulative vibration, respectively, and ; Indicates the sampling cycle number, It represents the root mean square value of the vibration velocity collected by the i-th vibration sensor in the j-th sampling period. The time window L is introduced to smooth the short-term fluctuations of the vibration intensity and obtain a more robust evaluation result.

2. The method for predictive maintenance of industrial equipment based on edge computing according to claim 1, characterized in that: Based on the multidimensional features, the vibration signal feature vector, temperature signal feature vector, current signal feature vector and sound signal feature vector of the industrial equipment are extracted, and based on the extracted feature vectors, the vibration intensity index, the temperature anomaly index, the current fluctuation index and the sound anomaly index are respectively calculated.

3. The method for predictive maintenance of industrial equipment based on edge computing according to claim 2, characterized in that: The temperature anomaly index is calculated based on the temperature signal feature vector, and the formula is: ,in, Indicates the number of samples in the current sampling period k. The average temperature of the measuring points, is the normal temperature value, For the allowable temperature, is the maximum allowable temperature rise over two sampling periods, is the arithmetic mean of all temperature measurement points in the kth cycle, is the reference variance when the temperature distribution is uniform, is the total number of temperature sensors; , , is the weight coefficient; the temperature anomaly index The calculation comprehensively considers three abnormal factors: the temperature absolute value exceeds the standard, the temperature change rate exceeds the standard, and the temperature spatial distribution is uneven.

4. The method for predictive maintenance of industrial equipment based on edge computing according to claim 2, characterized in that: The current fluctuation index is calculated based on the current signal characteristic vector as a quantitative index reflecting the degree of current imbalance. The calculation formula is as follows: ,in, The first The current value of the sampling point, For the The average value of the output current in the current cycle, for The total average value of the output current of each circuit, is the rated current, , It is the weight coefficient of the two indicators. The first item reflects the severity of the instantaneous current exceeding the standard, and the second item reflects the relative balance of multiple output currents. The worse the balance, the higher the risk of equipment failure.

5. The method for predictive maintenance of industrial equipment based on edge computing according to claim 2, characterized in that: The sound abnormality index is calculated based on the sound signal feature vector, wherein the frequency band energy distribution dispersion is in the form of the ratio of the standard deviation to the mean, so as to ensure that the index value falls within the interval [0, 1], and the calculation formula is as follows: ,in, It represents the sound pressure level of the ith sound measurement point in the current kth sampling period. For the current cycle The energy percentage of the frequency band, is the historical average of the energy percentage of the frequency band, is the total number of frequency bands, that is: Indicates the reference value / threshold of the sound pressure level, is the weight coefficient, and .

6. The method for predictive maintenance of industrial equipment based on edge computing according to claim 1, characterized in that: The OWA operator based on the order relationship is used to realize the index aggregation as the comprehensive health index, and the calculation formula is as follows: ,in, For the The first of the vibration intensity index, the temperature anomaly index, the current fluctuation index and the sound anomaly index of a sampling period Large value, is the corresponding OWA weight, and The comprehensive health index , indicating that the industrial equipment is in The overall degradation degree of the sampling period, and The larger it is, the worse the device health is and the higher the risk of failure.

7. The method for predictive maintenance of industrial equipment based on edge computing according to claim 1, characterized in that: The adaptive optimization algorithm based on sliding time window is specifically: Define a sliding time window , the window length is W, which is used to achieve adaptive optimization of the comprehensive health index; In each time window, using the known health status labels of the industrial equipment, construct Mapping model to state categories: ,in represents the real health status of the industrial equipment at the end of the current window k, is the classifier function, are the classifier parameters.

8. The method for predictive maintenance of industrial equipment based on edge computing according to claim 1, characterized in that: The probability calculation is specifically as follows: Assume that the degradation process of the industrial equipment from new to failure is governed by Distribution, its probability density function is: ,in The operating time of the industrial equipment; , is the shape parameter; , is the scale parameter; In the future The calculation formula for the failure probability of the industrial equipment within the time period is as follows: ,in, , for the future The estimated increase in the comprehensive health indicator within the time period, Through the historical health indicator series Obtained by trend extrapolation estimation.

9. The method for predictive maintenance of industrial equipment based on edge computing according to claim 8, characterized in that: When the failure probability Exceeding the preset warning probability When the industrial equipment A failure is very likely to occur within the time, which will trigger a high-level warning and recommend shutdown for maintenance.

10. An industrial equipment predictive maintenance system based on edge computing, used to execute an industrial equipment predictive maintenance method based on edge computing according to any one of claims 1 to 9, characterized in that: It consists of three parts: field device layer, edge computing layer and cloud service layer; The field equipment layer includes various types of industrial equipment, and vibration, temperature, current, and sound sensors are arranged on the industrial equipment to collect status monitoring data of the industrial equipment in real time; The edge computing layer is composed of several edge computing nodes deployed on site or near the industrial equipment, each of which is equipped with a data acquisition module, a data preprocessing module, a feature extraction module, a health assessment module and a fault warning module; The cloud service layer is responsible for receiving device health status data and warning information from the edge computing layer for global device status monitoring and statistical analysis, and continuously optimizing and updating the health assessment and warning algorithms of the edge computing nodes in combination with massive historical data and mechanism models.

Images