A real-time early warning method for coating equipment failures

Abstract

The present invention relates to the technical field of data processing. More specifically, the present invention relates to a method for real-time early warning of coating equipment failures. The method includes collecting pressure data and vibration data, obtaining the instability of each pressure data according to the surrounding data of each pressure data, obtaining the characteristic value of each pressure data according to the instability of each pressure data, obtaining the preference degree of each clustering result according to the correlation between the characteristic value of each pressure data and the abnormality degree of each pressure data in each clustering result, thereby obtaining the optimal number of midpoints, and then obtaining the optimal clustering result. According to the optimal clustering result, the final abnormality degree of each pressure data is obtained, and then the failure of the coating equipment is judged. The present invention improves the accuracy of anomaly detection.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing. More specifically, the present invention relates to a method for real-time early warning of coating equipment failures. Background Art

[0002] In modern industrial production, improving production efficiency is one of the important goals pursued by enterprises. As a key equipment on the production line, the operating condition of the coating equipment directly affects the efficiency of the entire production line. Therefore, a real-time early warning method is needed to detect and handle equipment failures in a timely manner to reduce downtime and production losses. During the operation of the coating machine, the pressure between the coating roller and the support roller directly affects the uniform distribution of the coating and the coating quality. Therefore, by performing real-time anomaly detection on the pressure data between the coating roller and the support roller, the failures of the coating equipment can be effectively detected.

[0003] The currently published patent application document with the publication number CN117473278A discloses a method for improving FMEA based on the fuzzy analytic hierarchy process, fuzzy VIKOR method, and PAM clustering. Improvements are made to the deficiencies in three aspects of the traditional FMEA: the evaluation of the failure mode level, the determination of the risk factor weight, and the ranking of the importance of the failure modes. First, triangular fuzzy numbers are used to make the evaluation of the failure mode more in line with the actual situation. Secondly, the comprehensive weight is used to improve the weight of the risk factors. Then, the fuzzy VIKOR method is used to improve FMEA, and the comprehensive evaluation value combining the group benefit value and the individual regret value is used as the basis for ranking. Finally, the PAM clustering algorithm is used to perform clustering analysis on the failure modes according to the RPN value and the comprehensive evaluation value of the failure modes.

[0004] The PAM (Partitioning Around Medoid) algorithm is a method for data clustering. By clustering the pressure data, the abnormal data in the data sequence can be effectively identified, and the real-time early warning of coating equipment failures can be better completed. However, the number of medoids in the PAM algorithm needs to be set manually. Due to the diverse changes in the pressure data between the coating roller and the support roller during the operation of the coating machine, if the number of medoids is too small, the clustering result may be too general to accurately reflect the internal structure of the pressure data; if the set number of medoids is too large, the clustering result may be too fragmented, increasing the difficulty of subsequent anomaly detection. Summary of the Invention

[0005] In order to solve the problem that the number of medoids set manually in the PAM algorithm is inaccurate due to the diverse changes in the pressure data, the present invention proposes a method for real-time early warning of coating equipment failures, which includes the following steps:

[0006] Collect pressure data and vibration data; take the mean of the absolute values of the differences between the fitted values and the actual values of all the surrounding data of each pressure data as the instability of each pressure data; take the product of the instability of each pressure data and the Pearson correlation coefficient of the instability sequence of each pressure data and the instability sequence of the vibration data corresponding to each pressure data as the eigenvalue of each pressure data;

[0007] Preset the initial number of midpoints as K, and the iteration step of the number of midpoints as 1. Cluster the pressure data according to the initial number of midpoints to obtain the first clustering result; according to the first clustering result and the eigenvalue of each pressure data, obtain the preference degree of the first clustering result; add one to K as the number of midpoints for the second time, and obtain the second clustering result, and according to the method for obtaining the preference degree of the first clustering result, obtain the preference degree of the second clustering result; according to the preference degrees of the first clustering result to the second clustering result, judge whether the initial number of midpoints continues to iterate; if the iteration stops, the second clustering result is the best clustering result; if the iteration continues, stop until the best clustering result is obtained;

[0008] According to the best clustering result, obtain the final abnormality degree of each pressure data, and then judge the fault of the coating equipment.

[0009] The innovation of the present invention is that according to the change characteristics of the pressure data and its surrounding data, the eigenvalue of each pressure data is obtained, which can reflect the abnormal situation of each pressure data. Then, the pressure data is clustered each time by iterating the number of midpoints of the PAM algorithm, and according to the eigenvalue of each pressure data, the preference degree of each clustering result is obtained to get the best clustering result. Finally, according to the best clustering result, the final abnormality degree of each pressure data is obtained, and then the fault of the coating equipment is judged, avoiding the problems of instability and uncertainty of the clustering result caused by the artificial setting of the number of midpoints in the traditional algorithm often relying on experience and subjective judgment, thus improving the quality and accuracy of clustering and further improving the efficiency of anomaly detection.

[0010] Preferably, the acquisition of all the surrounding data of each pressure data includes:

[0011] Preset the number of neighborhood data as M, and take the i-th pressure data, the M pressure data before the i-th pressure data, and the M pressure data after it as the surrounding data of the i-th pressure data.

[0012] It is convenient to obtain the eigenvalue of each pressure data according to all the surrounding data of each pressure data.

[0013] Preferably, the acquisition of the fitted values of all the surrounding data of each pressure data includes:

[0014] Use the least squares method to fit the data around the \(i\)-th pressure data to obtain the fitted values of each piece of data around the \(i\)-th pressure data.

[0015] Preferably, the acquisition of the instability sequences of the respective pressure data and the instability sequences of the vibration data corresponding to the respective pressure data includes:

[0016] Successively record the instability of each piece of data around the \(i\)-th pressure data as the instability sequence of the \(i\)-th pressure data;

[0017] Preset the number \(M\) of neighborhood data. Take the \(i\)-th vibration data, the \(M\) vibration data before the \(i\)-th vibration data, and the \(M\) vibration data after it as the data around the \(i\)-th vibration data; successively record the instability of each piece of data around the vibration data corresponding to the \(i\)-th pressure data as the instability sequence of the vibration data corresponding to the \(i\)-th pressure data.

[0018] Preferably, the obtaining of the preference degree of the first clustering result according to the first clustering result and the eigenvalue of each pressure data includes:

[0019] In the first clustering result, use the LOF algorithm to obtain the local outlier factor of each pressure data in its affiliated clustering cluster, denoted as the abnormality degree of each pressure data; according to the order of the sampling time of the pressure data, sort the abnormality degrees of each pressure data to obtain an abnormality degree sequence, and sort the eigenvalues of each pressure data to obtain an eigenvalue sequence; take the Pearson correlation coefficient between the abnormality degree sequence and the eigenvalue sequence as the preference degree of the first clustering result.

[0020] It is convenient to subsequently obtain the best clustering result according to the preference degree of each clustering result.

[0021] Preferably, the determination of whether to continue iterating the number of initial midpoints according to the preference degrees of the first clustering result and the second clustering result includes:

[0022] Use the least squares method to fit the preference degrees of the first and second clustering results to obtain a fitted curve. If the derivative value of the fitted curve is negative, the iteration of the number of initial midpoints stops at this time. If the derivative value of the fitted curve is positive, the iteration of the number of initial midpoints continues at this time.

[0023] Stopping the iteration of the number of midpoints according to the decreasing trend of the preference degree of each clustering result can obtain an accurate clustering result.

[0024] Preferably, the obtaining of the final abnormality degree of each pressure data according to the best clustering result, and then judging the fault of the coating equipment includes:

[0025] In the optimal clustering result, the LOF algorithm is used to obtain the local outlier factor of each pressure data in its affiliated clustering cluster, which is denoted as the final anomaly degree of each pressure data. If the final anomaly degree of any pressure data is greater than the anomaly threshold T, the pressure data is abnormal data. At this time, the coating equipment fails, and the remote monitoring and alarm of the coating equipment are immediately triggered to notify relevant personnel for maintenance in a timely manner.

[0026] The quality and accuracy of clustering are improved, and thus the accuracy of anomaly detection is improved.

[0027] Preferably, the acquisition of pressure data and vibration data includes:

[0028] The pressure sensor is arranged between the coating roller and the support roller of the coater, and the vibration sensor is arranged on the coating roller. The preset sampling time is 2 times per second. During the operation of the coater, pressure data and vibration data are acquired.

[0029] The present invention has the following beneficial effects: The purpose of the present invention is to obtain the characteristic value of the pressure data according to the change characteristics of the pressure data and its local data, which can reflect the abnormal situation of the pressure data. Then, the pressure data is clustered each time by iterating the number of midpoints of the PAM algorithm, and according to the correlation between the characteristic value of each pressure data and the anomaly degree of each pressure data in each clustering result, the preferred degree of each clustering result is obtained to obtain the optimal clustering result. Finally, according to the optimal clustering result, the final anomaly degree of each pressure data is obtained, and then the failure of the coating equipment is judged, avoiding the problems of instability and uncertainty of the clustering result caused by the artificial setting of the number of midpoints in the traditional algorithm often relying on experience and subjective judgment, thereby improving the quality and accuracy of clustering, and further improving the anomaly detection efficiency. Description of the Drawings

[0030] By reading the following detailed description with reference to the drawings, the above and other objects, features and advantages of the exemplary embodiments of the present invention will become easily understood. In the drawings, several embodiments of the present invention are shown in an exemplary rather than restrictive manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0031] Figure 1 is the step flowchart of a method for real-time early warning of coating equipment failure in an embodiment of the present invention. Detailed Embodiments

[0032] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are 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 skilled in the art without creative efforts belong to the scope of protection of the present invention.

[0033] The following will describe in detail the specific embodiments of the present invention in conjunction with the accompanying drawings.

[0034] Please refer to Figure 1 , which shows a step flowchart of a method for real-time warning of coating equipment failures provided by an embodiment of the present invention. The method includes the following steps:

[0035] S001. Collect pressure data and vibration data.

[0036] In the embodiment of the present invention, a pressure sensor is arranged between the coating roller and the support roller of the coater, and a vibration sensor is arranged on the coating roller. The preset sampling time is 2 times per second. During the operation of the coater, pressure data and vibration data are collected. It should be noted that the pressure data and the vibration data are time-aligned.

[0037] S002. According to the surrounding data of each pressure data, obtain the instability of each pressure data, and according to the instability of each pressure data, obtain the characteristic value of each pressure data.

[0038] It should be noted that the number of midpoints in the existing PAM algorithm needs to be set manually. However, the pressure data between the coating roller and the support roller during the operation of the coater is diverse. It is difficult to determine the optimal number of midpoints by the manual setting method. If the number of midpoints is too small, the clustering result may be too general to accurately reflect the internal structure of the pressure data; if the number of midpoints is too large, the clustering result may be too fragmented, increasing the difficulty of subsequent anomaly detection. Therefore, the present invention needs to iteratively and adaptively determine the number of midpoints to obtain the optimal clustering result of the pressure data between the coating roller and the support roller during the operation of the coater, thereby improving the anomaly detection result and realizing the real-time warning of coating equipment failures.

[0039] It should be further noted that when the coater malfunctions (for example, the bearing lubrication of the coating roller is insufficient or other failures), it will cause unstable fluctuations in the pressure data between the coating roller and the support roller. Therefore, the instability of each pressure data can be obtained first. It is represented by the mean fitting error of all the surrounding data of the pressure data through curve fitting of the surrounding data of the pressure data. Since curve fitting is a process of finding a curve to make the pressure data as close to the curve as possible, the smoother the change of the pressure data, the more conducive it is to the progress of curve fitting. Therefore, the smaller the mean fitting error of all the surrounding data of the pressure data, the more stable the pressure data, and vice versa, the more unstable the data;

[0040] Also, since the pressure data collected by the pressure sensor will be affected by the surrounding environmental noise, it will also cause unstable fluctuations in the pressure data. However, the instability of the pressure data caused by some failures of the coater is because the coating roller vibrates, resulting in fluctuations in the pressure data, that is, unstable fluctuations in the vibration data will cause unstable fluctuations in the pressure data. Therefore, the instability of each pressure data can be corrected by using the correlation between the pressure data and its corresponding vibration data as the credibility. The greater the correlation, the greater the credibility of the instability of each pressure data. Therefore, according to the instability of each pressure data and the correlation between the instability of each pressure data and its corresponding vibration data, the characteristic value of each pressure data is obtained.

[0041] [[ID= 6]]In the embodiment of the present invention, the i-th pressure data, the M pressure data before the i-th pressure data, and the M pressure data after it are used as the surrounding data of the i-th pressure data; the i-th vibration data, the M vibration data before the i-th vibration data, and the M vibration data after it are used as the surrounding data of the i-th vibration data; the preset number of neighborhood data M = 7. In other embodiments, the implementer can preset the value of M according to the specific implementation method.

[0042] Use the least squares method to fit the surrounding data of the i-th pressure data to obtain the fitting value of each surrounding data of the i-th pressure data;

[0043] Obtain the instability of each pressure data:

[0044] ;

[0045] In the formula, represents the instability of the i-th pressure data; represents the number of surrounding data of the i-th pressure data; represents the value of the j-th surrounding data of the i-th pressure data; The fitted value of the j-th surrounding data representing the i-th pressure data; || represents the absolute value symbol; the greater the difference between the fitted value and the actual value of each surrounding data of the i-th pressure data, the more unstable the i-th pressure data is.

[0046] The instability of each surrounding data of the i-th pressure data is sequentially recorded as the instability sequence of the i-th pressure data, and the instability of each surrounding data of the vibration data corresponding to the i-th pressure data is sequentially recorded as the instability sequence of the vibration data corresponding to the i-th pressure data; it should be noted that the vibration data corresponding to the sampling moment of the pressure data is the vibration data corresponding to the pressure data.

[0047] Obtain the characteristic value of each pressure data:

[0048] ;

[0049] In the formula, represents the characteristic value of the i-th pressure data; represents the instability of the i-th pressure data; norm() represents the normalization function; represents the Pearson correlation coefficient between the instability sequence of the i-th pressure data and the instability sequence of the vibration data corresponding to the i-th pressure data. The larger its value, the more credible the instability of the i-th pressure data is. The larger the value of , the more abnormal the i-th pressure data is.

[0050] S003. According to the correlation between the characteristic value of each pressure data and the abnormality degree of each pressure data in each clustering result, obtain the preference degree of each clustering result, and then obtain the optimal number of midpoints, and then obtain the optimal clustering result.

[0051] It should be noted that it is known that the number of midpoints in the existing PAM algorithm needs to be set manually. However, the pressure data between the coating roller and the support roller during the operation of the coater is diverse, and it is difficult to determine the optimal number of midpoints by the manual setting method. Therefore, in the present invention, the number of midpoints is iterated, and each clustering result is obtained. According to any clustering result, the abnormality degree of each pressure data in any clustering result is obtained. Also, since the characteristic value of each pressure data reflects the abnormality of each pressure data, the more relevant the abnormality degree of each pressure data in this clustering result is to the characteristic value of each pressure data, the more accurate this clustering result is. Therefore, according to the correlation between the abnormality degree of each pressure data and the characteristic value of each pressure data in each clustering result, obtain the preference degree of each clustering result. Finally, when the preference degree of the clustering result in the iteration process starts to decline, stop the iteration to obtain the optimal number of midpoints.

[0052] In an embodiment of the present invention, the preset initial number of midpoints is K. According to the initial number of midpoints, the PAM algorithm is used to cluster the pressure data to obtain the first clustering result. In the first clustering result, the LOF algorithm is used to obtain the local outlier factor of each pressure data in its affiliated clustering cluster, which is denoted as the abnormality degree of each pressure data. According to the order of the sampling time of the pressure data, the abnormality degrees of each pressure data are sorted to obtain an abnormality degree sequence, and the eigenvalue sequences of each pressure data are sorted to obtain an eigenvalue sequence. The Pearson correlation coefficient between the abnormality degree sequence and the eigenvalue sequence is used as the preference degree of the first clustering result.

[0053] The preset iteration step of the number of midpoints is 1. Add 1 to K as the second number of midpoints. According to the second number of midpoints, obtain the second clustering result and the preference degree of the second clustering result. The least squares method is used to fit the preference degrees of the first to second clustering results to obtain the first fitting curve. If the derivative value of the first fitting curve is negative, the iteration of the initial number of midpoints stops, and the second clustering result is the best clustering result.

[0054] If the derivative value of the first fitting curve is positive, the iteration of the initial number of midpoints continues. Add 1 to the second number of midpoints as the third number of midpoints, obtain the third clustering result and the preference degree of the third clustering result. The least squares method is used to fit the preference degrees of the first to third clustering results to obtain the second fitting curve, and the derivative value of the second fitting curve is judged, and so on, until the derivative value of the v-th fitting curve is negative, then stop. The (v + 1)-th clustering result is the best clustering result.

[0055] It should be noted that the method for obtaining the preference degree of each clustering result is the same as that of the first clustering result.

[0056] S004. According to the best clustering result, obtain the final abnormality degree of each pressure data, and then judge the fault of the coating equipment.

[0057] It should be noted that after obtaining the best clustering result, then according to the best clustering result, obtain the final abnormality degree of each pressure data to identify abnormal data.

[0058] In an embodiment of the present invention, in the best clustering result, the LOF algorithm is used to obtain the local outlier factor of each pressure data in its affiliated clustering cluster, which is denoted as the final abnormality degree of each pressure data. If the final abnormality degree of any pressure data is greater than the abnormality threshold T, the pressure data is abnormal data. At this time, the coating equipment fails, and the remote monitoring and alarm of the coating equipment are immediately triggered to notify relevant personnel for maintenance in time.

[0059] In an embodiment of the present invention, the preset anomaly threshold T = 0.8. In other embodiments, the implementer can preset the value of the anomaly threshold T according to the specific implementation.

[0060] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A real-time early warning method for coating equipment failures, characterized in that, Including: Collecting pressure data and vibration data; Taking the mean of the absolute values of the differences between the fitted values and the actual values of all the surrounding data of each pressure data as the instability of each pressure data; Taking the product of the instability of each pressure data and the Pearson correlation coefficient of the instability sequence of each pressure data and the instability sequence of the vibration data corresponding to each pressure data as the eigenvalue of each pressure data; The instability of each pressure data: ； In the formula, represents the instability of the i-th pressure data; represents the number of surrounding data of the i-th pressure data; represents the value of the j-th surrounding data of the i-th pressure data; represents the fitted value of the j-th surrounding data of the i-th pressure data; || represents the absolute value symbol; The eigenvalue of each pressure data: ； In the formula, represents the eigenvalue of the i-th pressure data; represents the instability of the i-th pressure data; norm() represents the normalization function; represents the Pearson correlation coefficient between the instability sequence of the i-th pressure data and the instability sequence of the vibration data corresponding to the i-th pressure data; Presetting the initial number of midpoints as K, presetting the iteration step of the number of midpoints as 1, clustering the pressure data according to the initial number of midpoints to obtain the first clustering result; According to the first clustering result and the eigenvalue of each pressure data, obtaining the preference degree of the first clustering result; Adding one to K as the second number of midpoints, and obtaining the second clustering result, and according to the method of obtaining the preference degree of the first clustering result, obtaining the preference degree of the second clustering result; Judging whether the initial number of midpoints continues to iterate according to the preference degrees of the first clustering result to the second clustering result; If the iteration stops, the second clustering result is the best clustering result; If the iteration continues, stop until the best clustering result is obtained; The judging whether the initial number of midpoints continues to iterate according to the first clustering result and the preference degree of the second clustering result includes: Using the least squares method to fit the preference degrees of the first and second clustering results to obtain a fitted curve. If the derivative value of the fitted curve is negative, the iteration of the initial number of midpoints stops at this time. If the derivative value of the fitted curve is positive, the iteration of the initial number of midpoints continues at this time; According to the best clustering result, obtaining the final abnormality degree of each pressure data, and further judging the fault of the coating equipment.

2. The real-time early warning method for coating equipment failures according to claim 1, characterized in that, The obtaining of all the surrounding data of each pressure data includes: Presetting the number of neighborhood data as M, and taking the i-th pressure data, the M pressure data before the i-th pressure data, and the M pressure data after the i-th pressure data as the surrounding data of the i-th pressure data.

3. A real-time warning method for coating equipment failures according to claim 1, characterized in that The obtaining of the fitted values of all the surrounding data of each pressure data includes: Using the least squares method to fit the surrounding data of the i-th pressure data to obtain the fitted value of each surrounding data of the i-th pressure data.

4. A real-time warning method for coating equipment failures according to claim 1, characterized in that The obtaining of the instability sequence of each pressure data and the instability sequence of the vibration data corresponding to each pressure data includes: Successively recording the instability of each surrounding data of the i-th pressure data as the instability sequence of the i-th pressure data; Presetting the number of neighborhood data as M, and taking the i-th vibration data, the M vibration data before the i-th vibration data, and the M vibration data after the i-th vibration data as the surrounding data of the i-th vibration data; Successively recording the instability of each surrounding data of the vibration data corresponding to the i-th pressure data as the instability sequence of the vibration data corresponding to the i-th pressure data.

5. A real-time warning method for coating equipment failures according to claim 1, characterized in that The obtaining of the preference degree of the first clustering result according to the first clustering result and the eigenvalue of each pressure data includes: In the first clustering result, the LOF algorithm is used to obtain the local outlier factor of each pressure data in its affiliated clustering cluster, denoted as the abnormality degree of each pressure data; according to the order of the sampling time of the pressure data, the abnormality degrees of each pressure data are sorted to obtain an abnormality degree sequence, and the eigenvalue sequences of each pressure data are sorted to obtain an eigenvalue sequence; the Pearson correlation coefficient between the abnormality degree sequence and the eigenvalue sequence is used as the preference degree of the first clustering result.

6. A real-time warning method for coating equipment failures according to claim 1, characterized in that, According to the optimal clustering result, the final abnormality degree of each pressure data is obtained, and then the faults of the coating equipment are judged, including: In the optimal clustering result, the LOF algorithm is used to obtain the local outlier factor of each pressure data in its affiliated clustering cluster, denoted as the final abnormality degree of each pressure data. If the final abnormality degree of any pressure data is greater than the abnormality threshold T, the pressure data is abnormal data. At this time, the coating equipment fails, and the remote monitoring and alarm of the coating equipment are immediately triggered to notify relevant personnel for maintenance in time.

7. A real-time warning method for coating equipment failures according to claim 1, characterized in that The acquisition of pressure data and vibration data includes: The pressure sensor is placed between the coating roller and the support roller of the coater, and the vibration sensor is placed on the coating roller. The preset sampling time is 2 times per second. During the operation of the coater, the pressure data and the vibration data are collected.

Images