A coating process abnormality prediction system and method

Abstract

This invention relates to the field of coating process monitoring and analysis technology, specifically to a coating process anomaly prediction system and method. The method includes: acquiring the core processes of coating production; establishing a data flow model between the core processes based on a multiple linear regression model; optimizing the regression coefficients of the model using Gaussian process regression combined with particle swarm optimization to determine the optimal regression coefficients; performing anomaly propagation path inference based on a Bayesian network path probability inference model; determining the edge weights of directed edges between nodes based on the calculation results of the improved Bayesian formula, and determining the influence coefficients between processes; processing the influence coefficients between processes, determining the predicted abnormal process for each process when an anomaly occurs, and sending notification information. This invention achieves accurate anomaly prediction and efficient notification, enhancing the accuracy and targeting of anomaly prediction, and is more suitable for the anomaly prevention and control needs of complex coating production processes.

Description

Technical Field

[0001] This invention relates to the field of coating process monitoring and analysis technology, specifically to a coating process anomaly prediction system and method. Background Technology

[0002] Paint production involves multiple core processes, including raw material proportioning, mixing, grinding, and color matching. These processes are closely linked through material transfer, and any abnormality in parameters at any stage can trigger a chain reaction, resulting in substandard product performance.

[0003] In traditional production, anomaly detection often relies on manual inspection or single parameter monitoring, making it difficult to capture hidden correlations between processes in a timely manner. Due to the lack of systematic modeling of parameter correlations and accurate anomaly propagation reasoning, there are often delays in fault location and low troubleshooting efficiency, which seriously affect production continuity and product quality stability, and urgently needs to be addressed. Summary of the Invention

[0004] The purpose of this invention is to address the problems existing in the background art by proposing a coating process anomaly prediction system and method.

[0005] The technical solution of the present invention: a method for predicting anomalies in a coating process, comprising the following steps: The core processes of coating are identified, and a data flow model between these core processes is established based on a multiple linear regression model. The regression coefficients of the model were optimized by combining Gaussian process regression with particle swarm optimization algorithm to determine the optimal regression coefficients; Anomaly propagation path inference is performed based on a Bayesian network-based path determination probability inference model. The edge weights of directed edges between nodes are determined based on the calculation results of the improved Bayesian formula, and the influence coefficients between processes are determined. The influence coefficients between processes are processed to determine the predicted abnormal process for each process when an anomaly occurs and to send notification information.

[0006] Preferably, the expression for the data flow model is as follows: ; In the formula, Let p be the key parameter of process k; where k is the process number and k is a positive integer; p is the key parameter number. The regression constant; These are the relevant parameters that affect the key parameter p in process k; represents the regression coefficients corresponding to the relevant parameters; i is the relevant parameter number, i is a positive integer; n is the total number of relevant parameters in process k; This is the data flow error term.

[0007] Preferably, the regression coefficients of the model are optimized, and the specific steps include: Initialize the particle swarm, which involves treating all regression coefficients of the multiple linear regression model as a single particle and randomly generating a number of particles to obtain the particle swarm. The position of each particle is substituted into the transfer function to calculate the predicted value; then the predicted value is compared with the actual measured value, and the mean square error is used as the fitness value. Based on the mean squared error, the historical best position of the particle and the global best position of the entire population are selected. The velocity and position of the particle are adjusted according to the update formula of the particle swarm optimization algorithm. The position of the particle is updated iteratively to determine the optimal solution of the particle and obtain the optimal regression coefficient of the model.

[0008] Preferably, the parameter weight vector is characterized by the optimal regression coefficients. .

[0009] Preferably, the method for inferring anomaly propagation paths includes the following steps: Acquire knowledge of coating production processes and historical production data to determine the dependencies between each node; Based on knowledge of coating production processes, determine the influence relationships between nodes and establish directed edges between nodes; Using historical production data, we calculate the conditional probabilities between each node and construct a conditional probability table. When an anomaly is detected in a certain process, it is marked as an abnormal process, and the remaining processes are marked as pending processes. The abnormal nodes and pending processes are used as evidence input for Bayesian network inference.

[0010] The preferred, improved Bayesian formula is as follows: ; In the formula, B represents the target process abnormality; Wa and Wj represent the pending processes abnormality; a and j are the pending process numbers, and a and j are both positive integers; Za and Zj represent the parameter weight vectors of the pending processes. This represents the probability that the process Wa to be determined will be abnormal when the target process is abnormal and the parameter weight vector is Z. This represents the conditional probability of the target process being abnormal when the pending process Wa is abnormal and the parameter weight vector is Z. It is obtained through historical data statistics. The prior probability of an anomaly occurring in the pending process Wj is calculated based on the historical anomaly frequency; the denominator is the sum of the probabilities of all upstream pending processes that may cause an anomaly in the target process, with parameter weight vector Z, and N represents the number of upstream processes that may affect the target process.

[0011] Preferably, the method for processing the influence coefficient between processes includes: The influence coefficients of different processes are sorted in descending order to obtain the influence coefficient ranking. Based on the influence coefficient ranking, the influence coefficient sub-ranking of the upstream process of each process is determined. The first element of the influence coefficient sub-ranking is marked as the process to be predicted as abnormal and the inspection is notified.

[0012] Preferably, the method for determining that a process has an anomaly includes using a data flow model to calculate the calculated values ​​of key parameters of the downstream process, calculating the difference between these values ​​and the actual measured values ​​to obtain the measurement difference, and comparing the measurement difference with a preset difference threshold; if the measurement difference is greater than the difference threshold, the process is determined to have an anomaly; otherwise, the process is determined to be normal. The actual measured values ​​of key parameters are compared with the preset standard range. If the actual measured value does not fall within the standard range, the process is judged to be abnormal; otherwise, the process is judged to be normal.

[0013] This invention also discloses a coating process anomaly prediction system, which applies the above-mentioned coating process anomaly prediction method, specifically including the following steps: The first model building module is used to obtain the core processes of coating and establish a data flow model between the core processes based on a multiple linear regression model. The model optimization module is used to optimize the regression coefficients of the model by combining Gaussian process regression with particle swarm optimization algorithm to determine the optimal regression coefficients; The second model building module is used to determine the path probability inference model based on Bayesian networks and perform anomaly propagation path inference. The calculation module is used to determine the edge weights of directed edges between nodes and the influence coefficients between processes based on the calculation results of the improved Bayes formula. The predictive analysis module is used to process the influence coefficients between processes, determine the predicted abnormal process when an anomaly occurs in each process, and send notification information.

[0014] Compared with the prior art, the above-mentioned technical solution of the present invention has the following beneficial technical effects: This invention integrates multiple methods, including multiple linear regression, Gaussian process regression combined with particle swarm optimization, Bayesian networks, and improved Bayesian formulas, to construct a complete system from core process transfer function modeling to anomaly prediction and notification. This achieves accurate anomaly prediction and efficient notification, enhancing the accuracy and relevance of anomaly prediction, and is more suitable for the anomaly prevention and control needs of complex processes in paint production. Attached Figure Description

[0015] Figure 1 This is a block diagram of a method according to Embodiment 1 of the present invention. Detailed Implementation

[0016] Example 1, as Figure 1As shown, the present invention proposes a method for predicting anomalies in a coating process, comprising: The core processes of coating are identified, and a data flow model between these core processes is established based on a multiple linear regression model. The regression coefficients of the model were optimized by combining Gaussian process regression with particle swarm optimization algorithm to determine the optimal regression coefficients; The expression for the data flow model is as follows: ; In the formula, Let p be the key parameter of process k; where k is the process number and k is a positive integer; p is the key parameter number. The regression constant; These are the relevant parameters that affect the key parameter p in process k; represents the regression coefficients corresponding to the relevant parameters; i is the relevant parameter number, i is a positive integer; n is the total number of relevant parameters in process k; For data stream error terms; For example, the core processes in paint production are ingredient mixing, grinding, dispersion, paint blending, filtration, and packaging; key parameters of the grinding process, including the median particle size distribution after the dispersion process, are obtained through a data flow model. and slurry viscosity ; Among them, the median diameter affects the particle size distribution after the dispersion process. Including grinding bead particle size Stirring speed and grinding time ; Affects slurry viscosity Including slurry solids content and grinding temperature ; The median diameter of the particle size distribution after the dispersion process and slurry viscosity The computational expressions obtained through the dataflow model are as follows: ; .

[0017] The regression coefficients of the model are optimized using Gaussian process regression combined with particle swarm optimization. The specific steps include: Initialize the particle swarm, which involves treating all regression coefficients of the multiple linear regression model as a single particle and randomly generating a number of particles to obtain the particle swarm. The position of each particle is substituted into the transfer function to calculate the predicted value; then the predicted value is compared with the actual measured value, and the mean squared error is used as the fitness value; it should be noted that the smaller the fitness value, the more accurate the model prediction, and the better the corresponding particle position; Based on the mean squared error, the particle's own historical best position and the global best position of the entire population are selected. The particle's velocity and position are adjusted according to the update formula of the particle swarm optimization algorithm. The particle's position is updated iteratively to determine the optimal solution of the particle and obtain the optimal regression coefficient of the model. Among them, the particle's own historical best position is the position that the particle found in the previous iteration process that minimizes the fitness value; the global best position of the entire population is the position that the entire particle swarm found in the previous iteration process that minimizes the fitness value. The parameter weight vector is represented by the optimal regression coefficient. .

[0018] Based on a Bayesian network-based path determination probabilistic inference model, anomaly propagation path inference is performed, including the following steps: Acquire knowledge of coating production processes and historical production data to determine the dependencies between each node; Based on the knowledge of coating production process, determine the influence relationship between nodes and establish directed edges between nodes; for example, the knowledge of coating production process records that the abnormal output particle size of the grinding process will affect the dispersion effect of the dispersion process, so establish a directed edge between the grinding process and the dispersion process, with the direction from the grinding process to the dispersion process. Using historical production data, we calculate the conditional probabilities between each node and construct a conditional probability table. When an anomaly is detected in a certain process, it is marked as an abnormal process, and the remaining processes are marked as pending processes. The abnormal nodes and pending processes are used as evidence input for Bayesian network inference. The edge weights of directed edges between nodes are determined based on the calculation results of the improved Bayesian formula, and the influence coefficients between processes are determined. The improved Bayes formula is as follows: ; In the formula, B represents the target process abnormality; Wa and Wj represent the pending processes abnormality; a and j are the pending process numbers, and a and j are both positive integers; Za and Zj represent the parameter weight vectors of the pending processes. This represents the probability that the process Wa to be determined will be abnormal when the target process is abnormal and the parameter weight vector is Z. This represents the conditional probability of the target process being abnormal when the pending process Wa is abnormal and the parameter weight vector is Z. It is obtained through historical data statistics. The prior probability of an anomaly occurring in the pending process Wj is calculated based on the historical anomaly frequency; the denominator is the sum of the probabilities of all upstream pending processes that may cause an anomaly in the target process, with parameter weight vector Z, where N represents the number of upstream processes that may affect the target process. The influence coefficients between processes are processed to determine the predicted abnormal process for each process when an anomaly occurs and to send notification information.

[0019] Methods for handling the influence coefficients between processes include: The influence coefficients of different processes are sorted in descending order to obtain the influence coefficient ranking. Based on the influence coefficient ranking, the influence coefficient sub-ranking of the upstream process of each process is determined. The first element of the influence coefficient sub-ranking is marked as the process to be predicted as abnormal and the inspection is notified. The method for determining that a process has an anomaly includes using a data flow model to calculate the calculated values ​​of key parameters for downstream processes, calculating the difference between these calculated values ​​and the actual measured values, obtaining the measurement difference, and comparing the measurement difference with a preset difference threshold; if the measurement difference is greater than the difference threshold, the process is determined to have an anomaly; otherwise, the process is determined to be normal. The actual measured values ​​of key parameters are compared with the preset standard range. If the actual measured value does not fall within the standard range, the process is judged to be abnormal; otherwise, the process is judged to be normal. The standard range is calculated based on historical parameter data. By integrating multiple methods such as multiple linear regression, Gaussian process regression combined with particle swarm optimization, Bayesian networks, and improved Bayesian formulas, a complete system from core process transfer function modeling to anomaly prediction and notification was constructed. This system enables accurate anomaly prediction and efficient notification, enhances the accuracy and pertinence of anomaly prediction, and is more suitable for the anomaly prevention and control needs of complex processes in paint production.

[0020] Example 2: The coating process anomaly prediction system proposed in this invention is applied to the coating process anomaly prediction method proposed in Example 1, specifically including: The first model building module is used to obtain the core processes of coating and establish a data flow model between the core processes based on a multiple linear regression model. The model optimization module is used to optimize the regression coefficients of the model by combining Gaussian process regression with particle swarm optimization algorithm to determine the optimal regression coefficients; The second model building module is used to determine the path probability inference model based on Bayesian networks and perform anomaly propagation path inference. The calculation module is used to determine the edge weights of directed edges between nodes and the influence coefficients between processes based on the calculation results of the improved Bayes formula. The predictive analysis module is used to process the influence coefficients between processes, determine the predicted abnormal process when an anomaly occurs in each process, and send notification information.

[0021] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A method for predicting anomalies in a coating process, characterized in that, Includes the following steps: The core processes of coating are identified, and a data flow model between these core processes is established based on a multiple linear regression model. The regression coefficients of the model were optimized by combining Gaussian process regression with particle swarm optimization algorithm to determine the optimal regression coefficients; Anomaly propagation path inference is performed based on a Bayesian network-based path determination probability inference model. The edge weights of directed edges between nodes are determined based on the calculation results of the improved Bayesian formula, and the influence coefficients between processes are determined. The influence coefficients between processes are processed to determine the predicted abnormal process for each process when an anomaly occurs and to send notification information.

2. The method for predicting anomalies in a coating process according to claim 1, characterized in that, The expression for the data flow model is as follows: ； In the formula, Let p be the key parameter of process k; where k is the process number and k is a positive integer; p is the key parameter number. The regression constant; These are the relevant parameters that affect the key parameter p in process k; represents the regression coefficients corresponding to the relevant parameters; i is the relevant parameter number, i is a positive integer; n is the total number of relevant parameters in process k; This is the data flow error term.

3. The method for predicting anomalies in a coating process according to claim 2, characterized in that, The specific steps for optimizing the regression coefficients of the model include: Initialize the particle swarm, which involves treating all regression coefficients of the multiple linear regression model as a single particle and randomly generating a number of particles to obtain the particle swarm. The position of each particle is substituted into the transfer function to calculate the predicted value; then the predicted value is compared with the actual measured value, and the mean square error is used as the fitness value. Based on the mean squared error, the historical best position of the particle and the global best position of the entire population are selected. The velocity and position of the particle are adjusted according to the update formula of the particle swarm optimization algorithm. The position of the particle is updated iteratively to determine the optimal solution of the particle and obtain the optimal regression coefficient of the model.

4. The method for predicting anomalies in a coating process according to claim 3, characterized in that, The parameter weight vector is represented by the optimal regression coefficient. .

5. The method for predicting anomalies in a coating process according to claim 4, characterized in that, The method for inferring abnormal propagation paths includes the following steps: Acquire knowledge of coating production processes and historical production data to determine the dependencies between each node; Based on knowledge of coating production processes, determine the influence relationships between nodes and establish directed edges between nodes; Using historical production data, we calculate the conditional probabilities between each node and construct a conditional probability table. When an anomaly is detected in a certain process, it is marked as an abnormal process, and the remaining processes are marked as pending processes. The abnormal nodes and pending processes are used as evidence input for Bayesian network inference.

6. The method for predicting anomalies in a coating process according to claim 5, characterized in that, The improved Bayes formula is as follows: ； In the formula, B represents the target process abnormality; Wa and Wj represent the pending processes abnormality; a and j are the pending process numbers, and a and j are both positive integers; Za and Zj represent the parameter weight vectors of the pending processes. This represents the probability that the process Wa to be determined will be abnormal when the target process is abnormal and the parameter weight vector is Z. This represents the conditional probability of the target process being abnormal when the pending process Wa is abnormal and the parameter weight vector is Z. It is obtained through historical data statistics. The prior probability of an anomaly occurring in the pending process Wj is calculated based on the historical anomaly frequency; the denominator is the sum of the probabilities of all upstream pending processes that may cause an anomaly in the target process, with parameter weight vector Z, and N represents the number of upstream processes that may affect the target process.

7. The method for predicting anomalies in a coating process according to claim 5, characterized in that, Methods for handling the influence coefficients between processes include: The influence coefficients of different processes are sorted in descending order to obtain the influence coefficient ranking. Based on the influence coefficient ranking, the influence coefficient sub-ranking of the upstream process of each process is determined. The first element of the influence coefficient sub-ranking is marked as the process to be predicted as abnormal and the inspection is notified.

8. The method for predicting anomalies in a coating process according to claim 7, characterized in that, Methods for determining process anomalies include using a data flow model to calculate the calculated values ​​of key parameters in downstream processes, calculating the difference between these values ​​and the actual measured values, obtaining the measurement difference, and comparing the measurement difference with a preset difference threshold. If the measurement difference is greater than the difference threshold, the process is determined to be abnormal; Otherwise, the process is considered normal; The actual measured values ​​of key parameters are compared with the preset standard range. If the actual measured value does not fall within the standard range, the process is judged to be abnormal; otherwise, the process is judged to be normal.

9. A coating process anomaly prediction system, applied to the coating process anomaly prediction method according to any one of claims 1 to 8, characterized in that, Specifically, it includes: The first model building module is used to obtain the core processes of coating and establish a data flow model between the core processes based on a multiple linear regression model. The model optimization module is used to optimize the regression coefficients of the model by combining Gaussian process regression with particle swarm optimization algorithm to determine the optimal regression coefficients; The second model building module is used to determine the path probability inference model based on Bayesian networks and perform anomaly propagation path inference. The calculation module is used to determine the edge weights of directed edges between nodes and the influence coefficients between processes based on the calculation results of the improved Bayes formula. The predictive analysis module is used to process the influence coefficients between processes, determine the predicted abnormal process when an anomaly occurs in each process, and send notification information.

Images