Telemetry System for Predictive Maintenance of Equipment Based on Industrial Automation

Abstract

The present invention belongs to the technical field of equipment maintenance, and discloses a telemetry system for predictive maintenance of equipment based on industrial automation. The method includes: collecting equipment operation data of industrial automation equipment; preprocessing the equipment operation data to obtain preprocessed data; modeling the preprocessed data to obtain an equipment comprehensive prediction model; predicting the future operation state of the equipment based on the equipment comprehensive prediction model to obtain an equipment health assessment result; determining corresponding maintenance strategies and maintenance times based on the equipment health assessment result, and implementing maintenance on the equipment according to the corresponding maintenance strategies and maintenance times, which can not only reduce the operation cost of the enterprise, but also improve the equipment utilization rate and enhance the market competitiveness of the enterprise.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment maintenance, and more specifically, to a predictive maintenance telemetry system for equipment based on industrial automation. Background Art

[0002] A patent with the application publication number CN110705133A discloses a method and equipment for predictive maintenance, including: receiving first data sent by a data transmission device; classifying the first data, where the types of the first data include normal data, abnormal data, and fault data; establishing a prediction model using a variance algorithm and the first data, which can provide data support for equipment maintenance, realize practical equipment predictive maintenance, reduce maintenance costs, reduce downtime, and avoid production losses and material waste caused by equipment failures.

[0003] However, there are still many problems in the existing industrial automation equipment maintenance technology, which seriously restrict the improvement of production efficiency and equipment reliability. Traditional regular maintenance methods often cannot accurately reflect the actual condition of the equipment, resulting in waste of maintenance resources or insufficient maintenance; while simple fault detection systems often cannot detect potential problems in time, leading to sudden equipment failures, production line shutdowns, and huge economic losses; existing predictive maintenance systems usually only focus on a single type of data, such as only analyzing vibration or temperature, and it is difficult to comprehensively grasp the health status of the equipment; in terms of data processing, many systems lack effective noise processing capabilities, affecting the data quality and the reliability of the analysis results; in addition, the current equipment state modeling methods are often too simplified to accurately capture the non-linear dynamic characteristics in complex industrial environments; in terms of health assessment, most systems only consider short-term states and ignore long-term trends, resulting in lack of foresight in maintenance decisions; the formulation of maintenance strategies also often lacks flexibility and systematicness, and it is difficult to make optimal decisions according to the actual condition of the equipment and the specific needs of the enterprise; these problems are particularly prominent in actual production.

[0004] In view of this, the present invention proposes a predictive maintenance telemetry system for equipment based on industrial automation to solve the above problems. Summary of the Invention

[0005] To overcome the above-mentioned defects of the prior art and to achieve the above object, the present invention provides the following technical solution: A predictive maintenance telemetry system for equipment based on industrial automation, including: a data acquisition module for collecting equipment operation data of industrial automation equipment;

[0006] a data processing module for preprocessing the equipment operation data to obtain preprocessed data;

[0007] A modeling module, which is used to model the preprocessed data by using an improved Markov chain to obtain an equipment comprehensive prediction model;

[0008] A comprehensive evaluation module, which predicts the future operating state of the equipment based on the equipment comprehensive prediction model to obtain the equipment health assessment result;

[0009] A strategy maintenance module, which determines the corresponding maintenance strategy and maintenance time based on the equipment health assessment result, and implements maintenance on the equipment according to the corresponding maintenance strategy and maintenance time; each module is connected by wired and / or wireless means.

[0010] Furthermore, the equipment operation data includes vibration data, temperature data, current data, and power data.

[0011] Furthermore, the method for preprocessing the equipment operation data includes:

[0012] Clean the equipment operation data separately according to the data type. The method for data cleaning is to remove duplicate data and outliers; and process the noise data; obtain the preliminarily processed operation data; perform normalization or standardization processing on the preliminarily processed operation data to obtain the preprocessed data.

[0013] Furthermore, the method for processing the noise data includes:

[0014] Model the equipment operation data as a linear mixture of k source signals and noise respectively according to the data type, that is ; where is the time index, is the mixing matrix; is the observed signal;

[0015] Perform preliminary processing on the observed signal , and the preliminary processing is to subtract its mean value to make it have a zero mean, obtaining the preliminary observed signal ; Calculate the covariance matrix of each preliminary observed signal at each time ;

[0016] At each time , use the current estimated value of the covariance matrix to perform eigenvalue decomposition to obtain the eigenvector matrix and its diagonal matrix ; Based on the eigenvector matrix and its diagonal matrix construct the transformation matrix ; where is The transpose; at each time , multiply the preliminary observation signal on the left by the conversion matrix to obtain the middle-section signal ; invert the middle-section signal to obtain the standard original signal ; perform post-processing on the standard original signal to obtain the preliminary processed operation data.

[0017] Furthermore, the calculation method of the covariance matrix includes:

[0018] Define an adaptive forgetting factor (); initialize a covariance matrix as a diagonal matrix or an identity matrix; and a gain vector as a zero vector; for the preliminary observation signal , recursively calculate the gain vector and update the covariance matrix; the formula for calculating the gain vector is:

[0019] ; where, represents the estimated value of the covariance matrix at time , represents the gain vector calculated at time , which is a column vector; is the transpose of , is a positive constant;

[0020] The formula for updating the covariance matrix is:

[0021] ; where, is the regularization term coefficient, is the identity matrix; is the estimated value of the covariance matrix at time ;

[0022] Adaptive forgetting factor ; where, is the kurtosis of , is the kurtosis constant of a normal distribution; is the skewness of , and are the weight coefficients of the corresponding terms, is the exponential parameter; is the approximate entropy of ; is a positive adjustment coefficient.

[0023] Furthermore, the middle-section signal ​​​​The ways to perform inverse transformation include:

[0024] Transform the middle segment signal on an overcomplete dictionary for transformation representation, that is ; where is the sparse coefficient vector;

[0025] Add the transformation mixing matrix on the basis of the transformation representation, that is ; Define the optimization objective function FH = ; where is the square of the L2 norm therein, is 's L1 norm, is 's mixed norm, calculate the L2 norm for each row of , and then calculate the L1 norm for all rows' L2 norms; is the time-varying regularization coefficient; is the fixed regularization coefficient;

[0026] ; where is a basic L1 regularization coefficient, and are two positive exponential adjustment coefficients; is 's dimension, is 's L2 norm;

[0027] By minimizing the function value of the optimization objective function and normalizing each column of , the specific value of the transformation mixing matrix is obtained. At this time, use QR decomposition or maximum likelihood estimation method to decompose to obtain the inverse matrix W; Use the inverse matrix W to perform a linear transformation on the middle segment signal to obtain the standard original signal .

[0028] Furthermore, the ways to model the preprocessed data include:

[0029] Preset a high-dimensional phase space; Divide the high-dimensional phase space into grids to obtain a hierarchical grid model; Map the preprocessed data into the hierarchical grid model to obtain a trajectory sequence, and each time step corresponds to a state point in the hierarchical grid model;

[0030] Traverse each state point in the trajectory sequence, and determine the number of the grid it belongs to according to the coordinates of the state point; maintain a dictionary structure to store the number of visits to each grid. For the grid to which each state point belongs, increment the corresponding value in the dictionary by 1. After traversing the entire trajectory sequence, the values in the dictionary are the number of visits to each grid; based on the number of visits, calculate its visit probability, that is, the number of visits divided by the length of the trajectory sequence; regard each grid as a discrete state of the Markov chain, and for each grid, calculate the average residence time of the trajectory sequence in this grid, and regard the average residence time as the duration distribution of this discrete state.

[0031] Define the state transition count matrix , where the element records the number of times transferred from the -th grid to the -th grid. For the -th grid, traverse the trajectory sequence, accumulate the value, and calculate the state transition probability of the -th grid; where is the smoothing parameter; is the sum of all transfer times starting from the -th grid; represents the size of the state space;

[0032] Based on the average residence time, define the self-loop probability ; where is the average residence time of the trajectory sequence in the -th grid; based on the self-loop probability and the state transition probability construct the state transition matrix of the Markov chain. For each row of the state transition matrix, normalize the values of the state transition probabilities so that the row sum is 1; the Markov chain with this state transition matrix is the equipment comprehensive prediction model.

[0033] Furthermore, the method of dividing the high-dimensional phase space into grids includes:

[0034] Define the number of levels and the scale; that is, specify a scale parameter for each level; the scale parameter is the embedding dimension or the time delay;

[0035] Starting from the 0-th level, use a uniform grid to construct the initial grids. For the -th level ( ), based on the grid structure of the -th level, construct new grids by merging adjacent grids; construct layer by layer upward until the defined number of levels is reached; during this process, number the grids simultaneously.

[0036] The merging method includes:

[0037] At the level, calculate the entropy and the probability of being accessed for each grid, perform a weighted sum of the entropy and the probability of being accessed to obtain the statistical feature of the corresponding grid; calculate the similarity score between any two grids according to the statistical feature; construct a similarity matrix S from the similarity scores between all pairs of grids; based on the similarity matrix S, use the hierarchical clustering algorithm to cluster the grids to generate a hierarchical clustering tree; for each grid at the level, calculate its Silhouette coefficient; on the hierarchical clustering tree, starting from the root node, move down layer by layer to cut the height. For each cutting height, calculate the average Silhouette coefficient SC of all grids to obtain a mapping curve between the cutting height and SC; identify the inflection point in the mapping curve, and the cutting height corresponding to the inflection point is the optimal cutting height; according to the optimal cutting height, merge the grids belonging to the same class on the hierarchical clustering tree, and the merged grid inherits the statistical features of the grids before merging.

[0038] Furthermore, the method for obtaining the health assessment result of the device includes:

[0039] According to the current operation data of the device, determine its current state in the hierarchical grid model, and use this grid as the initial state of the device comprehensive prediction model;

[0040] Use the state transition matrix to perform n3-step prediction, where n3 is an integer greater than 1; obtain the short-term prediction distribution after n3 time steps starting from the initial state;

[0041] Let n3 approach infinity, and the Markov chain converges to a stable state distribution, denoted as the steady-state distribution; pre-define a set composed of one or several health states , and the states in it are the states of normal operation of the device; the element in the short-term prediction distribution represents the probability of being in the th state after n3 steps, and calculate the short-term health degree based on this probability ;

[0042] The element in the steady-state distribution represents the probability of being in the th state in the long term, then calculate the long-term health degree ; where, is the average sojourn time of the th state, and are two positive function adjustment parameters; the health assessment results include short-term health and long-term health.

[0043] Furthermore, the determination methods of the maintenance strategy and maintenance time include:

[0044] Define a set of maintenance strategies, which contains all optional maintenance strategies; define a time window T, which is a discrete time series composed of possible maintenance times; form a decision space D1 by taking the Cartesian product of the set of maintenance strategies and the time window T; define a decision function ;

[0045] ; where, and are strategy weight coefficients; is a positive constant, is also a positive constant; is the maintenance time, is the base exponential constant; is the weight factor related to the maintenance strategy , different correspond to different values;

[0046] Regard the decision space D1 as a discrete space, and each decision in the decision space D1 is a mass point; calculate the function value of each mass point on the decision function, denoted as the mass of the mass point.

[0047] Define a gravitational field , the elements in the gravitational field are the gravitational force and between any two mass points ; where is the gravitational constant, are two mass points and the Euclidean distance between, is the mass of the mass point , is the mass of the mass point , is a characteristic positive real-valued function;

[0048] ; where, is the base positive constant; and are two positive weight parameters; and are the maintenance times corresponding to the mass points and respectively, is to measure the mass point and the similarity function of the corresponding decision

[0049] In the gravitational field calculate the resultant force of all other particles on each particle, and select the particle with the maximum resultant force and its corresponding decision is the optimal maintenance strategy and maintenance time

[0050] The technical effects and advantages of the equipment predictive maintenance telemetry system based on industrial automation of the present invention are as follows:

[0051] The present invention realizes comprehensive and real-time monitoring of the operating state of industrial automation equipment, greatly improves the accuracy and timeliness of fault prediction. By intelligently analyzing short-term and long-term health indicators, the system can accurately identify potential problems and effectively prevent sudden equipment failures. This not only significantly reduces unplanned downtime, but also optimizes the maintenance plan and avoids problems of over-maintenance and under-maintenance; its dynamically adjusted maintenance strategy ensures the optimal allocation of maintenance resources, prolongs the service life of the equipment, and reduces the overall maintenance cost; the adaptive ability of the system makes it applicable to various complex industrial environments and can continuously optimize the prediction model according to the actual condition of the equipment; it greatly improves the stability and reliability of the production line, significantly improves production efficiency and product quality; in addition, the prediction function of the system provides a reliable basis for production planning and resource allocation, which helps to optimize the overall operation strategy; in the long run, it can not only reduce the operation cost, but also improve the equipment utilization rate and enhance the market competitiveness; at the same time, it helps to reduce resource consumption by reducing unnecessary repairs and replacements Brief Description of the Drawings

[0052] Figure 1 is a schematic diagram of the equipment predictive maintenance telemetry system based on industrial automation of the present invention

[0053] Figure 2 is a schematic diagram of the equipment predictive maintenance telemetry method based on industrial automation of the present invention Detailed Description of the Embodiments

[0054] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention

[0055] Embodiment 1

[0056] Please refer to Figure 1As shown in the figure, the telemetry system for predictive maintenance of equipment based on industrial automation in this embodiment includes: a data acquisition module for acquiring the equipment operation data of industrial automation equipment;

[0057] a data processing module for preprocessing the equipment operation data to obtain preprocessed data;

[0058] a modeling module for modeling the preprocessed data using an improved Markov chain to obtain an equipment comprehensive prediction model;

[0059] a comprehensive evaluation module for predicting the future operation state of the equipment based on the equipment comprehensive prediction model to obtain the equipment health degree evaluation result;

[0060] a strategy maintenance module for determining the corresponding maintenance strategy and maintenance time based on the equipment health degree evaluation result, and implementing maintenance on the equipment according to the corresponding maintenance strategy and maintenance time; each module is connected by wired and / or wireless means to realize data transmission between modules.

[0061] The equipment operation data includes vibration data, temperature data, current data, and power data; the data is collected in time series. Specifically, acceleration sensors are installed on the key components of the equipment (such as bearings, gearboxes, etc.); the acceleration sensors convert mechanical vibration into analog voltage or current signals, and a data acquisition card or a fieldbus module is used to sample and digitize the analog voltage or current signals to obtain vibration data; temperature data is obtained by installing temperature sensors (such as thermocouples, thermal resistors, etc.); a clamp-on current sensor or a leakage current sensor is used to measure the operating current of the equipment to obtain current data, and a power meter or a power sensor is used to measure the active power, reactive power, etc. of the equipment; the power meter or the power sensor outputs analog voltage or current signals, and the analog signals are sampled and digitized, and the sampling frequency is appropriately set according to the power fluctuation situation to obtain power data.

[0062] The collected analog and digital data needs to be collected, converted, and stored in real time by the data acquisition module to construct the equipment operation data, laying a foundation for subsequent data processing and analysis.

[0063] The methods for preprocessing the equipment operation data include:

[0064] Clean the device operation data separately according to the data types (vibration data, temperature data, current data, and power data). The method of data cleaning is as follows: remove duplicate data and outliers, such as data points that significantly exceed the reasonable range, check and repair data inconsistencies, such as inconsistent units and incorrect formats; and process noise data; obtain the preliminarily processed operation data; perform normalization or standardization on the preliminarily processed operation data to make variables with different dimensions within the same numerical range, and obtain the preprocessed data.

[0065] The methods for processing noise data include:

[0066] Model the device operation data as a linear mixture of k source signals and noise respectively according to the data types, that is ; where is the time index, is the mixing matrix; each row of the mixing matrix represents an observation signal channel; each column corresponds to a source signal, and the elements therein represent the contribution weights of the source signals to the observation signals; is the observation signal, that is, the data signal composed of different data types.

[0067] Preliminarily process the observation signal by subtracting its mean value to make it have a zero mean, and obtain the preliminary observation signal ; remove the DC component of the signal to make subsequent processing more convenient; calculate the covariance matrix of each time preliminary observation signal . Specifically, define an adaptive forgetting factor ( ), which controls the weights of new and old data; initialize a covariance matrix as a diagonal matrix or an identity matrix; and a gain vector as a zero vector; for the preliminary observation signal , recursively calculate the gain vector and update the covariance matrix; the formula for calculating the gain vector is:

[0068] ; where represents the estimated value of the covariance matrix at time , represents the gain vector calculated at time , which is a column vector; used to update the estimated value of the covariance matrix according to ; is the transpose of , is a small positive constant used to improve numerical stability and avoid division by zero.

[0069] The formula for updating the covariance matrix is as follows:

[0070] ; where is the regularization term coefficient, is the identity matrix, which is used to ensure that the estimated covariance matrix has an appropriate condition number and improve the robustness of the estimation; is the estimated value of the covariance matrix at time ;

[0071] Adaptive forgetting factor ; where is the kurtosis of is the kurtosis constant of a normal distribution, approximately equal to 3; is the skewness of and are the weight coefficients of the corresponding terms, controlling the influence degree, is the exponential parameter, controlling the sensitivity of the change range of the adaptive forgetting factor; is the approximate entropy of; reflecting the uncertainty and randomness of the data; is a positive adjustment coefficient, used to control the influence degree of the skewness.

[0072] At each time , using the estimated value of the current covariance matrix to perform eigenvalue decomposition, obtaining the eigenvector matrix and its diagonal matrix ; Based on the eigenvector matrix and its diagonal matrix to construct the transformation matrix ; where is the transpose of ; At each time , multiplying the preliminary observation signal on the left by the transformation matrix , obtaining the middle section signal ; Inverting the middle section signal to obtain the standard original signal.

[0073] Specifically, representing the middle section signal on an overcomplete dictionary , that is ; where is the sparse coefficient vector, most of whose elements are 0; The overcomplete dictionary is a pre-designed basis matrix (such as wavelet basis, Fourier basis, etc.), or a dictionary obtained by data-driven learning.

[0074] Add a transformation mixing matrix based on the transformation representation , that is ; Define the optimization objective function FH = ; where is the square of the L2 norm therein, is 's L1 norm, is 's mixed norm. Calculate the L2 norm for each row (corresponding to each source signal) of , and then calculate the L1 norm of the L2 norms of all rows; is the time-varying regularization coefficient, which is adaptively adjusted to obtain the optimal sparsity; is the fixed regularization coefficient, which is used to enhance the robustness of the estimation.

[0075] ; where is a basic L1 regularization coefficient, and are two positive exponential adjustment coefficients, controls the attenuation rate, controls relative to 's range of variation; is 's dimension, is 's L2 norm;

[0076] By minimizing the function value of the optimization objective function and normalizing each column of , the specific value of the transformation mixing matrix is obtained. At this time, use QR decomposition or maximum likelihood estimation method to decompose to obtain the inverse matrix W; Use the inverse matrix W to perform a linear transformation on the middle section signal to obtain the standard original signal .

[0077] Perform post-processing on the standard original signal to obtain the data after processing the noise data, and then obtain the preliminary processed operation data; The ways of performing post-processing include sorting and scaling; Specifically, use the known prior knowledge of the source signal to sort each component of , for example, sorting based on time-domain features (such as statistical moments, peaks, etc.), sorting based on frequency-domain features (such as dominant frequency, bandwidth, etc.) or sorting based on time-frequency features (such as wavelet packets, etc.); According to the known statistical characteristics of the source signal, scale each component of , for example, scaling based on variance, scaling based on peak value or scaling based on energy.

[0078] The ways to model the preprocessed data include:

[0079] Preset a high-dimensional phase space; partition the high-dimensional phase space into grids to obtain a hierarchical grid model. Specifically, define the number of levels and the scale; that is, specify a scale parameter for each level to control the grid granularity at that level; the scale parameter is the embedding dimension or the time delay.

[0080] Starting from the finest granularity level (level 0), use a uniform grid to construct the initial grids. For the level ( ), based on the grid structure of the level, construct new grids (coarse-grained) by merging adjacent grids; build layer by layer upwards until the defined number of levels is reached; during this process, number the grids simultaneously.

[0081] The merging methods include:

[0082] At the level, calculate the entropy and the probability of being visited for each grid, perform a weighted sum of the entropy and the probability of being visited to obtain the statistical characteristics of the corresponding grid; according to the statistical characteristics, calculate the similarity score (Euclidean distance, cosine similarity, or KL divergence) between any two grids; construct a similarity matrix S from the similarity scores between all pairs of grids; based on the similarity matrix S, use a hierarchical clustering algorithm (such as AGNES or DIANA) to cluster the grids, generating a hierarchical clustering tree (dendrogram).

[0083] For each grid at the level, calculate its Silhouette coefficient (the Silhouette coefficient is an index used to evaluate the clustering effect. It combines the compactness and separation of the clustering. The value of this coefficient ranges from -1 to 1, and the larger the value, the better the clustering effect); on the hierarchical clustering tree, starting from the root node, move down layer by layer to cut the height. For each cut height, calculate the average Silhouette coefficient SC of all grids to obtain a mapping curve between the cut height and SC; identify the inflection point in the mapping curve, and the cut height corresponding to the inflection point is the optimal cut height; according to the optimal cut height, merge the grids belonging to the same class on the hierarchical clustering tree, and the merged grids inherit the statistical characteristics of the grids before merging; for example, the probability of being visited of the coarse-grained grid can be the weighted average of the probabilities of the fine-grained grids.

[0084] Map the preprocessed data into a hierarchical lattice model to obtain a trajectory sequence, where each time step corresponds to a state point (phase space vector) in the hierarchical lattice model; traverse each state point in the trajectory sequence, and determine the number of the lattice it belongs to according to the coordinates of the state point (an efficient hash-based lookup method can be used to quickly locate the lattice); maintain a dictionary (hash table) structure to store the number of times each lattice is visited. For each lattice that a state point belongs to, increment the value corresponding to the dictionary by 1. After traversing the entire trajectory sequence, the value in the dictionary is the number of times each lattice is visited; based on the number of visits, calculate its visit probability, that is, the number of visits divided by the length of the trajectory sequence (the number of state points). If a new trajectory sequence arrives, there is no need to re-statistics all lattices. Only need to traverse the new trajectory, update the visit times of the corresponding lattices, and then recalculate the visit probability of each lattice, which is applicable to streaming data.

[0085] Regard each lattice as a discrete state of a Markov chain. For each lattice (discrete state), calculate the average sojourn time of the trajectory sequence in this lattice, and the average sojourn time is regarded as the duration distribution of this discrete state.

[0086] Define the state transition count matrix , where the element records the number of times transferred from the -th lattice to the -th lattice. For the -th lattice, traverse the trajectory sequence, accumulate the value, and calculate the state transition probability of the -th lattice; where, is the smoothing parameter, which is equivalent to the maximum likelihood estimate when is equal to 0; is the sum of all transfer times starting from the -th lattice; represents the size of the state space, that is, the total number of discrete states.

[0087] Due to the existence of the sojourn time, each state has a certain probability of self-looping. Therefore, based on the average sojourn time, define the self-loop probability ; where, is the average sojourn time of the trajectory sequence in the -th lattice.

[0088] Based on the self-loop probability and the state transition probability Construct the state transition matrix of the Markov chain. For each row of the state transition matrix, normalize the values of the state transition probabilities so that the row sum is 1; this ensures the randomness of the state transition matrix. The Markov chain with this state transition matrix is the equipment comprehensive prediction model. The state transition matrix describes the evolution law of the system state in the phase space and is the basis for subsequent state prediction and health assessment.

[0089] The ways to obtain the health assessment results of the equipment include:

[0090] According to the current operation data of the equipment, determine its current state (grid) in the hierarchical lattice model, and use this grid as the initial state of the equipment comprehensive prediction model.

[0091] Use the state transition matrix to perform n3-step prediction, where n3 is an integer greater than 1; obtain the short-term prediction distribution (represented by a vector or matrix) after n3 time steps starting from the initial state. This distribution reflects the uncertainty in the short term.

[0092] Let n3 approach infinity. The Markov chain will converge to a stable state distribution, denoted as the steady-state distribution (represented by a vector or matrix). The steady-state distribution is the unique steady-state distribution of the Markov chain and can be obtained through matrix calculation. The steady-state distribution reflects the average state distribution after long-term operation.

[0093] Pre-define a set composed of one or several health states , The states in it are the states of normal operation of the equipment. The elements in the short-term prediction distribution represent the probability of being in the th state after n3 steps. Based on this probability, calculate the short-term health degree ;

[0094] The elements in the steady-state distribution represent the probability of being in the th state in the long term. Then calculate the long-term health degree ; where, is the average sojourn time of the th state, and are two positive function adjustment parameters used to control the shape of the function. For those states that belong to the health state set H but have a relatively long average sojourn time, appropriately reduce their contribution to the long-term health degree. Because, if staying in a certain health state for a long time, it may also mean the existence of potential abnormalities. It should be noted that, and The same state can be indexed simultaneously; here, to distinguish the elements in the short-term prediction distribution and the steady-state distribution; the health assessment results include short-term health and long-term health.

[0095] The determination methods of the maintenance strategy and the maintenance time include:

[0096] Define a set of maintenance strategies, which contains all optional maintenance strategies, such as replacing parts, cleaning, adjusting, etc.; define a time window T, which is a discrete time series composed of possible maintenance times; form a decision space D1 by taking the Cartesian product of the set of maintenance strategies and the time window T; define a decision function ;

[0097] ; where and are the strategy weight coefficients, used to balance short-term health and long-term health; is a positive constant, used to control the attenuation rate, is also a positive constant, used to adjust the influence degree of long-term health; is the maintenance time, is the base exponential constant, such as ; is the weight factor related to the maintenance strategy , different correspond to different values; for example, it can be set that, if it is replacing parts then ; if it is cleaning then , if it is adjusting then ; in this way, the corresponding to the replacing parts strategy is the largest, indicating that its influence weight on long-term health is the highest.

[0098] Regard the decision space D1 as a discrete space, and each decision (maintenance strategy and maintenance time) in the decision space D1 is a particle; calculate the function value of each particle on the decision function, denoted as the mass of the particle.

[0099] Define a gravitational field , and the elements in the gravitational field are the gravitational force and between any two particles ; where is the gravitational constant, is the Euclidean distance between the two particles and , is the particle The quality (the function value on the decision function), is the mass of the particle , is a characteristic positive real-valued function that dynamically adjusts the influence degree of distance on gravity according to the characteristics of different particle pairs.

[0100] ; where is a basic normal constant, such as ; and are two positive weight parameters used to control the influence degrees of time difference and policy difference; and are the maintenance times corresponding to the particles and respectively, is a similarity function for measuring the decisions corresponding to the particles and (a weighted function that synthesizes the type of decision, the mass of the particle corresponding to the decision, and the cost of implementing the decision).

[0101] In the gravitational field , calculate the resultant force of all other particles on each particle, and select the particle with the largest resultant force. Its corresponding decision is the optimal maintenance strategy and maintenance time; according to the obtained optimal maintenance strategy and maintenance time, perform corresponding maintenance on the equipment.

[0102] Comprehensively consider the short-term and long-term health conditions of the equipment, and search for the optimal maintenance strategy and maintenance time in the decision space, so as to achieve efficient predictive maintenance of the equipment.

[0103] In this embodiment, it realizes comprehensive and real-time monitoring of the operation status of industrial automation equipment, greatly improves the accuracy and timeliness of fault prediction. By intelligently analyzing short-term and long-term health indicators, the system can accurately identify potential problems and effectively prevent sudden equipment failures. This not only significantly reduces unplanned downtime, but also optimizes the maintenance plan and avoids the problems of over-maintenance and under-maintenance; its dynamically adjusted maintenance strategy ensures the optimal allocation of maintenance resources, prolongs the service life of the equipment, and reduces the overall maintenance cost; the adaptive ability of the system makes it applicable to various complex industrial environments and can continuously optimize the prediction model according to the actual conditions of the equipment; it greatly improves the stability and reliability of the production line, significantly improves production efficiency and product quality; in addition, the prediction function of the system provides a reliable basis for production planning and resource allocation, helping to optimize the overall operation strategy; in the long run, it can not only reduce the operation cost, but also improve the equipment utilization rate and enhance the market competitiveness; at the same time, by reducing unnecessary repairs and replacements, it also helps to reduce resource consumption.

[0104] Example 2

[0105] Please refer to Figure 2 As shown, for the parts not described in detail in this embodiment, refer to the description content of Embodiment 1. A device predictive maintenance telemetry method based on industrial automation is provided, including:

[0106] S1. Collect the device operation data of industrial automation devices;

[0107] S2. Preprocess the device operation data to obtain preprocessed data;

[0108] S3. Use an improved Markov chain to model the preprocessed data to obtain a device comprehensive prediction model;

[0109] S4. Based on the device comprehensive prediction model, predict the future operation state of the device to obtain the device health assessment result;

[0110] S5. Determine the corresponding maintenance strategy and maintenance time based on the device health assessment result, and implement maintenance on the industrial automation device according to the corresponding maintenance strategy and maintenance time.

[0111] Example 3

[0112] This embodiment publicly provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the operation mode of the above-provided device predictive maintenance telemetry method based on industrial automation.

[0113] Since the electronic device introduced in this embodiment is the electronic device used to implement the device predictive maintenance telemetry method based on industrial automation in the embodiments of the present application, based on the device predictive maintenance telemetry method based on industrial automation introduced in the embodiments of the present application, those skilled in the art can understand the specific implementation manners and various variations of the electronic device in this embodiment. Therefore, the specific implementation of how this electronic device implements the method in the embodiments of the present application will not be described in detail here. As long as those skilled in the art implement the electronic device used to implement the device predictive maintenance telemetry method based on industrial automation in the embodiments of the present application, it falls within the scope of protection of the present application.

[0114] The above formulas are all dimensionless and take their numerical values for calculation. The formulas are obtained by software simulation of collecting a large amount of data to approximate the real situation. The preset parameters and threshold selection in the formulas are set by those skilled in the art according to the actual situation.

[0115] The above are only the preferred embodiments of the present invention. The protection scope of the present invention is not limited to the above embodiments. All technical solutions falling within the concept of the present invention belong to the protection scope of the present invention. It should be noted that for those of ordinary skill in the art, several improvements and refinements made without departing from the principle of the present invention should also be regarded as within the protection scope of the present invention.

Claims

1. Equipment predictive maintenance telemetry system based on industrial automation, characterized by: include: Data acquisition module, used to collect equipment operation data of industrial automation equipment; A data processing module is used to pre-process the equipment operation data to obtain pre-processed data; A modeling module is used to model the preprocessed data using an improved Markov chain to obtain a comprehensive prediction model for the equipment; The method of modeling the preprocessed data includes: Preset a high-dimensional phase space; divide the high-dimensional phase space into grids to obtain a hierarchical lattice model; map the preprocessed data into the hierarchical lattice model to obtain a trajectory sequence, where each time step corresponds to a state point in the hierarchical lattice model; Maintain a dictionary structure to store the number of visits to each grid, and regard each grid as a discrete state of the Markov chain. For each grid, count the average residence time of the trajectory sequence in the grid, and regard the average residence time as the duration distribution of the discrete state; Define the state transition count matrix , where the elements Recorded from The grid is transferred to The number of grids, for the grids, traverse the trajectory sequence, and accumulate , and calculate the value of The state transition probability of each grid is defined based on the average residence time. The state transition matrix of the Markov chain is constructed based on the self-cycle probability and the state transition probability. The Markov chain with the state transition matrix is ​​the comprehensive prediction model of the equipment. The comprehensive evaluation module predicts the future operation status of the equipment based on the equipment comprehensive prediction model and obtains the health evaluation result of the equipment; The strategy maintenance module determines the corresponding maintenance strategy and maintenance time based on the health assessment results of the equipment, and performs maintenance on the equipment according to the corresponding maintenance strategy and maintenance time; each module is connected by wired and / or wireless means; The maintenance strategy and maintenance time are determined in the following ways: Define a maintenance strategy set, which includes all optional maintenance strategies; define a time window T, which is a discrete time series consisting of possible maintenance times; form a decision space D1 by taking the Cartesian product of the maintenance strategy set and the time window T; define a decision function; The decision space D1 is regarded as a discrete space, and each decision in the decision space D1 is a mass point; the function value of each mass point on the decision function is calculated and recorded as the mass of the mass point; Defining the Gravity Field , gravity field The elements in are any two particles and The gravitational force between In the gravity field For each particle, calculate the resultant force of all other particles on it, and select the particle with the largest resultant force. , and its corresponding decision is the optimal maintenance strategy and maintenance time.

2. The equipment predictive maintenance telemetry system based on industrial automation according to claim 1 is characterized in that: The equipment operation data includes vibration data, temperature data, current data and power data.

3. The equipment predictive maintenance telemetry system based on industrial automation according to claim 2 is characterized in that: The method of preprocessing the equipment operation data includes: The equipment operation data is cleaned according to the data type. The data cleaning method is as follows: remove duplicate data and outliers; process noise data; obtain preliminary processed operation data; normalize or standardize the preliminary processed operation data to obtain preprocessed data.

4. The equipment predictive maintenance telemetry system based on industrial automation according to claim 3 is characterized in that: The method of processing noise data includes: The equipment operation data is modeled as k source signals according to the data type. and noise A linear mixture of ;in, is the time index, is the mixing matrix; To observe the signal; The observed signal Perform preliminary processing, which is to subtract its mean to make it have zero mean and obtain the preliminary observation signal ; Calculate each time Initial observation signal The covariance matrix of At every time , using the current estimate of the covariance matrix Perform eigenvalue decomposition to obtain the eigenvector matrix and its diagonal matrix ; Based on the eigenvector matrix and its diagonal matrix Constructing the transformation matrix ;in, for The transpose of , the initial observation signal Left multiply by the transformation matrix , get the middle signal ; The middle signal Perform reverse transformation to obtain the standard original signal ; For the standard original signal Post-processing is performed to obtain preliminary processing operation data.

5. The equipment predictive maintenance telemetry system based on industrial automation according to claim 4 is characterized in that: The covariance matrix is ​​calculated by: Defining the adaptive forgetting factor ( ); initialize a covariance matrix as a diagonal matrix or a unit matrix; and a gain vector as a 0 vector; for the initial observation signal , recursively calculate the gain vector and update the covariance matrix; the formula for calculating the gain vector is: ;in, Indicates at time The estimated value of the covariance matrix when , Indicates at time The calculated gain vector is a column vector; for The transpose of is a positive constant; The formula for updating the covariance matrix is: ;in, is the regularization coefficient, is the unit matrix; For in time The estimated value of the covariance matrix when ; Adaptive forgetting factor ;in, for The kurtosis of is the kurtosis constant of a normal distribution; for The skewness of and is the weight coefficient of the corresponding item, is the index parameter; for The approximate entropy of is a positive adjustment coefficient.

6. The equipment predictive maintenance telemetry system based on industrial automation according to claim 5, characterized in that: The mid-segment signal Ways to reverse the transformation include: The middle signal In an overcomplete dictionary The above transformation is expressed as ;in, is a sparse coefficient vector; Add a transformation mixing matrix based on the transformation representation ,Right now ; Define the optimization objective function FH= ;in, is the square of the L2 norm within it, for The L1 norm of for The mixed norm of Calculate the L2 norm for each row of , and then calculate the L1 norm of the L2 norm of all rows; is the time-varying regularization coefficient; is a fixed regularization coefficient; ;in, is a basic L1 regularization coefficient, and are two positive exponential adjustment coefficients; for The dimension of for The L2 norm of ; By minimizing the function value of the objective function, Normalize each column of to obtain the transformed mixing matrix The specific value of , in this case, QR decomposition or maximum likelihood estimation method is used to decompose , get the inverse matrix W; use the inverse matrix W to correct the mid-segment signal Perform linear transformation to obtain the standard original signal .

7. The equipment predictive maintenance telemetry system based on industrial automation according to claim 6, characterized in that: Traverse each state point in the trajectory sequence, and determine the number of the grid it belongs to according to the coordinates of the state point; for each grid to which the state point belongs, add 1 to the value of the corresponding dictionary. After traversing the entire trajectory sequence, the value in the dictionary is the number of visits to each grid; based on the number of visits, calculate its visit probability, that is, the number of visits divided by the length of the trajectory sequence; State transition probability ;in, is the smoothing parameter; It is from The sum of all transfer times starting from the grid; Indicates the size of the state space; Self-loop probability ;in, For the trajectory sequence The average residence time of a grid.

8. The equipment predictive maintenance telemetry system based on industrial automation according to claim 7, characterized in that: The method of gridding the high-dimensional phase space includes: Define the number of levels and scales; that is, specify a scale parameter for each level; the scale parameter is the embedding dimension or delay; Starting from level 0, use a uniform grid to construct the initial grid. level( ), based on A hierarchical grid structure, where new grids are constructed by merging adjacent grids; this is done layer by layer until a defined number of levels is reached; the grids are numbered during this process; The merging methods include: In the The entropy and the probability of being visited of each grid are counted, and the entropy and the probability of being visited are weighted and summed to obtain the statistical characteristics of the corresponding grid; based on the statistical characteristics, the similarity score between any two grids is calculated; the similarity scores between all grids are constructed into a similarity matrix S; based on the similarity matrix S, the grids are clustered using a hierarchical clustering algorithm to generate a hierarchical clustering tree; for each The Silhouette coefficient of each grid is calculated; on the hierarchical clustering tree, starting from the root node, the cutting height is moved downward layer by layer, and for each cutting height, the average Silhouette coefficient SC of all grids is calculated to obtain a mapping curve between the cutting height and SC; the inflection point in the mapping curve is identified, and the cutting height corresponding to the inflection point is the optimal cutting height; according to the optimal cutting height, the grids belonging to the same category on the hierarchical clustering tree are merged, and the merged grids inherit the statistical characteristics of the grids before the merger.

9. The equipment predictive maintenance telemetry system based on industrial automation according to claim 8, characterized in that: The method for obtaining the health evaluation result of the device includes: According to the current operation data of the equipment, its current state is determined in the hierarchical grid model, and the grid is used as the initial state of the equipment comprehensive prediction model; Using the state transfer matrix, perform n3-step prediction, where n3 is an integer greater than 1; starting from the initial state, after n3 time steps, the short-term prediction distribution is achieved; As n3 tends to infinity, the Markov chain converges to a stable state distribution, which is recorded as the steady-state distribution; Predefine a set of one or more health states , The state in is the state of normal operation of the equipment; the elements in the short-term prediction distribution Indicates that after n3 steps, it is in the The probability of a state, based on which the short-term health is calculated ; Elements in a Steady-State Distribution Indicates that the long-term The probability of a state is calculated to obtain the long-term health ;in, For the The average dwell time of a state, and are two positive function adjustment parameters; the health assessment results include short-term health and long-term health.

10. The equipment predictive maintenance telemetry system based on industrial automation according to claim 9, characterized in that: The decision function ;in, and is the strategy weight coefficient; is a normal number, is also a normal number; For maintenance time, is the basic exponential constant; Yes and maintenance strategy Related weight factors, different Corresponding to different value; Two particles and The attraction between ;in is the gravitational constant, There are two particles and The Euclidean distance between For mass point Quality, For mass point Quality, is a characteristic positive real-valued function; ;in, is the underlying positive constant; and are two positive weight parameters; and Particle and The corresponding maintenance time, For measuring point and The corresponding decision similarity function.

Images