A UV curing machine fault detection method based on data analysis

Abstract

The present application relates to the technical field of electric parameter measurement and industrial equipment fault diagnosis, and particularly relates to a UV curing machine fault detection method based on data analysis, which comprises the following steps: continuously collecting voltage data points of the UV curing machine, determining suspected noise factors by the difference between the voltage data points and the mean of surrounding data segments and the difference between adjacent voltage data points, and calculating noise probability in combination with the suspected noise factors and fluctuation possibility. In the recursive aggregation of the condensed hierarchical clustering algorithm, each two clusters are taken as a group, a score value is calculated according to the noise probability and distance of the data points in the group, a group of data points according to which single-link aggregation is determined is determined according to the score value, the condensed hierarchical clustering algorithm is optimized, and the optimized algorithm is used for abnormal detection on the voltage data points to determine whether the UV curing machine has voltage failure. The method improves the accuracy of UV curing machine fault detection.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of electrical parameter measurement and industrial equipment fault diagnosis. Specifically, it relates to a UV curing machine fault detection method based on data analysis. BACKGROUND

[0002] In modern manufacturing, UV curing machines are commonly used in processes such as surface coating and curing of photocurable resin. As an indispensable device in these production lines, the voltage stability of the UV curing machine directly determines the service life of electrical components and the curing effect. Voltage abnormalities such as transient fluctuations and persistent shifts are the main causes of equipment failure. Light causes incomplete curing and product scrap, while heavy causes lamp tube burnout and power module damage.

[0003] Traditional manual detection and regular inspection cannot capture subtle voltage abnormalities in real time, and passive maintenance is often performed after the equipment is shut down due to voltage failure, significantly increasing maintenance costs and downtime losses.

[0004] In the prior art, the condensed hierarchical clustering algorithm is a commonly used data anomaly detection algorithm. The condensed hierarchical clustering algorithm is used to identify abnormal patterns in voltage data to determine whether there is a voltage fault, so as to intervene in time to prevent electrical component damage. This clustering method is a recursive aggregation process, and a single-link aggregation method is used in the recursive aggregation process. Specifically, the distance between the two closest data points in two clusters is selected as the distance between the two clusters, and clustering merging is performed based on this distance to generate a hierarchical clustering result tree. That is, during the clustering operation, the distance between the closest data points in the two clusters is used as the key for single-link aggregation, which determines the final clustering result.

[0005] However, during the collection of voltage data points of the UV curing machine, electromagnetic interference from electrical equipment such as motors and switching power supplies around the UV curing machine can cause noise data in the voltage. These noise data can interfere with the normal readings of the sensor, causing distortion in the voltage data points. When using a single-link aggregation method to merge two clusters, if there is noise data in the two clusters, the noise data will reduce the accuracy of the distance calculation between the two clusters, affecting the accuracy of the clustering result, and ultimately leading to false positives in the voltage during the operation of the UV curing machine, and thus reducing the accuracy and effectiveness of voltage fault detection. SUMMARY

[0006] In order to solve the problem of inaccurate clustering result caused by noise data interference when detecting abnormal situation of voltage data point of the UV curing machine, and further causing misjudgment of voltage abnormal situation of the UV curing machine, and affecting the accuracy and effectiveness of the UV curing machine voltage fault detection, the application provides a UV curing machine fault detection method based on data analysis.

[0007] In one aspect, the application provides a UV curing machine fault detection method based on data analysis, comprising:

[0008] Continuously collect voltage data points of the UV curing machine during operation, and divide the surrounding data segment of each voltage data point;

[0009] Based on the difference between each voltage data point and the mean value of its surrounding data segment, and the difference between each voltage data point and the adjacent voltage data point on the left and right, determine the suspected noise factor of each voltage data point; according to the suspected noise factor of each voltage data point and the fluctuation possibility of each voltage data point, determine the noise probability of each voltage data point;

[0010] In the process of recursive aggregation of the multiple clusters obtained by using the agglomerative hierarchical clustering algorithm to cluster the voltage data points, each two clusters are taken as a cluster group, one voltage data point is selected from each of the two clusters of each group to form a group of data points, and the score value of each group of voltage data points of each cluster group is determined according to the noise probability and distance of each group of data points;

[0011] According to the score value of each group of voltage data points of each cluster group, determine a group of voltage data points on which each cluster group is based when single-link aggregation, so as to realize the optimization of the agglomerative hierarchical clustering algorithm; use the optimized agglomerative hierarchical clustering algorithm to detect the voltage data points of the UV curing machine during operation, and determine whether the UV curing machine has voltage fault according to the abnormal detection result.

[0012] This technical solution continuously collects abundant voltage data points and defines a local data range to facilitate the analysis of the characteristics of each data point within that local range. By comprehensively considering the suspected noise factor and fluctuation probability of each voltage data point, the noise probability is determined, enabling more accurate identification of noisy data and avoiding misclassification of normally fluctuating voltage data points as noise. It also improves the accuracy of judging true noise data, providing more reliable input data for subsequent clustering algorithm optimization. During the recursive aggregation process of the agglomerative hierarchical clustering algorithm, the noise probability of each data point in each cluster and the distance between each data point are calculated to determine the score. This provides a quantitative standard for determining the data points used in subsequent single-link aggregation, allowing for a comprehensive consideration of the noise probability and distance factors of data points during clustering. This avoids erroneous clustering caused by noisy data, improving the accuracy of clustering results and making the anomaly detection results of voltage data points more accurate. Based on accurate anomaly detection results, it is possible to more accurately determine whether a UV curing machine has a voltage fault, improving the accuracy and reliability of UV curing machine fault detection.

[0013] Furthermore, the suspected noise factor for each voltage data point was determined based on the following method:

[0014] Calculate the first The absolute value of the difference between the voltage data point and its left-side adjacent data point, and the... The absolute value of the difference between each voltage data point and its adjacent data point to the right is used to denot the larger absolute value of the difference. ;

[0015] No. Suspected noise factors at each voltage data point In the formula, For the first The absolute value of the difference between a voltage data point and the mean of its surrounding data segments. This is a hyperparameter.

[0016] This technical solution, through this comprehensive calculation method, can more fully and accurately reflect the likelihood that each voltage data point is noise data. It considers both the deviation of the voltage data point within a local range and the differences between the voltage data point and its adjacent voltage data points, making the identification of noise data more reliable.

[0017] Furthermore, the noise probability of each voltage data point is determined based on the following method: the normalized value of the ratio of the suspected noise factor of each voltage data point to the fluctuation probability of each voltage data point is used as the noise probability of each voltage data point.

[0018] This technical solution avoids misjudgments that might arise from judging noise data solely based on data distribution characteristics by comprehensively considering the suspected noise factor and fluctuation probability of each voltage data point. Specifically, it takes into account that normal voltage fluctuations at certain stages of operation may cause voltage data points to deviate significantly from surrounding voltage data points, but these voltage data points are not noise data. By introducing the fluctuation probability factor, the true noise data can be identified more accurately, reducing misjudgments of normal fluctuation data.

[0019] Furthermore, the fluctuation probability of each voltage data point is determined based on the following method: obtaining the time interval between the corresponding time of each voltage data point and the start time of the UV curing machine, as well as the current data point at the corresponding time of each voltage data point; selecting all historical voltage data points consistent with each voltage data point, and calculating the average of all historical current data points corresponding to all selected historical voltage data points as the reference current value of that voltage data point.

[0020] Calculate the probability of fluctuation for each voltage data point: In the formula, For the first The probability of fluctuation at each voltage data point For the first The time interval between each voltage data point and the start-up time of the UV curing machine. For the first The absolute value of the difference between the current data point at the corresponding time and the reference current value at that voltage data point. It is a natural exponential function.

[0021] This technical solution takes into account that voltage fluctuations may be more frequent during the initial startup of a UV curing machine, but the voltage will gradually stabilize over time, reducing the likelihood of fluctuations. By combining the factors of time interval and current to determine the probability of fluctuations, the voltage fluctuation characteristics of the UV curing machine at different operating stages can be more accurately quantified.

[0022] Furthermore, the score for each voltage data point in each cluster is calculated based on the following formula:

[0023] In the formula, For the first The first cluster The score of the group voltage data points and The first The first cluster The noise probability of the first and second voltage data points in a group of voltage data points. For the first The first cluster a distance of two voltage data points of each group of voltage data points, is a natural exponential function, wherein the distance of two voltage data points is determined based on an absolute value of a difference of the two voltage data points.

[0024] The technical solution determines the score value by comprehensively considering the noise probability of each group of voltage data points of each cluster and the distance between two voltage data points contained in each group of voltage data points. When selecting a group of voltage data points for single-linkage aggregation, a group of voltage data points with low noise probability and short distance is preferentially selected, which can effectively exclude the interference of noise data in the clustering process and improve the accuracy of the clustering result.

[0025] Further, the determination method of a group of voltage data points for single-linkage aggregation of each cluster is as follows: calculating the score value of all groups of voltage data points of each cluster, and selecting a group of voltage data points with the highest score value as a group of voltage data points for single-linkage aggregation of the cluster.

[0026] The technical solution selects a group of voltage data points with the highest score value as a group of voltage data points for single-linkage aggregation of the cluster. Compared with the traditional method of directly selecting the nearest data points in two clusters as distance measurement, this method can more accurately reflect the real distance relationship between two clusters because it selects a group of voltage data points with small noise influence and high data reliability.

[0027] Further, the abnormality detection of voltage data points of the UV curing machine during operation includes:

[0028] In the process of recursive aggregation in the agglomerative hierarchical clustering algorithm, a distance threshold is preset. If the inter-cluster distance of a cluster and another cluster when merging is less than the distance threshold, the cluster after merging of the cluster and the other cluster is taken as a normal cluster. If the inter-cluster distance of a cluster and another cluster when merging is greater than or equal to the distance threshold, the cluster and the other cluster are separated and taken as two suspected abnormal clusters.

[0029] A number threshold is preset. If the number of voltage data points of a suspected abnormal cluster is greater than the preset number threshold, the suspected abnormal cluster is removed, and the remaining suspected abnormal clusters are taken as abnormal clusters. The voltage data points contained in the abnormal clusters are taken as abnormal voltage data points, so as to complete the abnormality detection of the voltage data points.

[0030] Further, the method for determining whether the UV curing machine has voltage failure according to the abnormality detection result is: presetting a proportion threshold; if the proportion of the number of abnormal voltage data points to the total number of voltage data points exceeds the proportion threshold within M minutes before the current analysis moment, it is determined that the UV curing machine has voltage failure, and relevant personnel are notified to check; if the proportion of the number of abnormal voltage data points to the total number of voltage data points does not exceed the proportion threshold within M minutes before the current analysis moment, it is determined that the UV curing machine does not have voltage failure; M is a preset value.

[0031] Further, the method for dividing the surrounding data segment of each voltage data point is: taking each voltage data point as the center, selecting a preset number of voltage data points on both sides of the voltage data point, to form the surrounding data segment of the voltage data point.

[0032] Further, the reference current value of the voltage data point is determined based on the following method: for any one voltage data point, all historical voltage data points consistent with the voltage data point are selected, the mean value of all historical current data points corresponding to the selected all historical voltage data points is calculated as the reference current value of the voltage data point.

[0033] The present application has the following effects:

[0034] The present application can accurately identify noise data in voltage data points through a series of methods, can effectively exclude noise interference when using the condensed hierarchical clustering algorithm to detect the abnormality of voltage data points, optimizes the condensed hierarchical clustering algorithm, improves the accuracy and stability of the clustering result, and based on the accurate clustering result, can accurately judge the voltage abnormality of the UV curing machine during operation, and further makes the voltage failure detection of the UV curing machine more accurate. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 is a method flowchart of the present application;

[0036] Figure 2 is a method flowchart of step S2 of the present application;

[0037] Figure 3 is a method flowchart of step S3 of the present application;

[0038] Figure 4 is a schematic diagram of the hierarchical clustering result tree of the present application. DETAILED DESCRIPTION

[0039] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application.

[0040] Reference Figure 1The application provides a UV curing machine fault detection method based on data analysis, which comprises steps S1-S5.

[0041] S1: collecting voltage data points of the UV curing machine in the running process.

[0042] In order to accurately obtain the voltage data points of the UV curing machine in the running process, a voltage sensor is selected to perform the data collection task, and the collection duration is set to one hour, during which the voltage data points are collected at a frequency of five times per second. This frequency can more densely capture the subtle changes of the voltage of the UV curing machine in the running process, and provide sufficient and detailed data basis for subsequent analysis.

[0043] S2: evaluating the noise probability of each voltage data point.

[0044] In order to accurately evaluate the noise probability of each voltage data point, this step needs to analyze the value performance and value change performance of each voltage data point in its surrounding data segment to obtain the suspected noise factor of each voltage data point. However, in the running process of the UV curing machine, the voltage may be in a state of gentle fluctuation and rise, so the suspected noise factor determined by the relative size of a data point in its surrounding data points is easy to misjudge the normal fluctuation of the voltage as an abnormality. Therefore, the suspected noise factor and the fluctuation possibility of each voltage data point are comprehensively determined to accurately reflect the noise probability of each voltage data point.

[0045] As shown in the figure, it comprises the following steps: Figure 2

[0046] S21: determining the suspected noise factor of each voltage data point.

[0047] Since the noise data points usually have a large numerical difference from their surrounding data points, when analyzing the suspected noise factor of each voltage data point, the more outstanding the value performance of a voltage data point in its surrounding data segment is, the more likely it is a noise data, and the larger the suspected noise factor is. The larger the value change between a voltage data point and its adjacent voltage data points is, the more likely it is a noise data, and the more outstanding the value performance of the voltage data point in the surrounding data segment is, and the larger the corresponding suspected noise factor is.

[0048] Therefore, taking each voltage data point as the center, a predetermined number of voltage data points are selected on both sides of the voltage data point to form the surrounding data segment of the voltage data point, and the predetermined number is 20 (an empirical value). If the number of voltage data points on one side of the voltage data point is less than 20, it can be supplemented on the other side.

[0049] For the first voltage data point, the surrounding data segment of the first voltage data point is composed of the first voltage data point and the second voltage data point. ​the absolute value of the difference between the first voltage data point and its left adjacent voltage data point, and the absolute value of the difference between the first voltage data point and its right adjacent voltage data point, and the larger one is recorded as the absolute value of the difference between the first voltage data point and its adjacent voltage data point. This operation can capture the mutation of the voltage data point.

[0050] In particular, if a voltage data point only has a left adjacent voltage data point or only has a right adjacent voltage data point, then the absolute value of the difference between the voltage data point and its only one adjacent voltage data point.

[0051]

[0052]

[0053] In the formula, is the absolute value of the difference between the first voltage data point and the mean value of the surrounding data segment of the first voltage data point, is a hyperparameter, which exists to prevent the case where is 0.

[0054] If the difference between a voltage data point and its adjacent voltage data point is large, it means that the voltage data point deviates from the surrounding voltage data points in value, and the voltage data point is more likely to be noise data. In the normal trend of voltage data points, the voltage values between adjacent voltage data points usually do not jump greatly, and if there is a large jump, it is likely to be disturbed by noise.

[0055] The mean value of the surrounding data segment of a voltage data point represents the typical characteristics of the surrounding data segment, and the difference between a voltage data point and the mean value of its surrounding data segment reflects the deviation of the voltage data point from the overall level of its surrounding data segment. The larger the difference, the more the voltage data point deviates from the normal range of its surrounding data segment, and the more likely the voltage data point is noise data.

[0056] S22: Calculate the fluctuation possibility of each voltage data point.

[0057] ​​​​​​The suspected noise factor of each voltage data point is based on the numerical performance of each voltage data point in its surrounding data segment. This step further considers that the voltage data points of the UV curing machine also have certain regular fluctuations, which may be confused with noise data, so it is necessary to further combine the analysis of other dimensions based on the suspected noise factor of the voltage data point to more accurately reflect the noise probability of a voltage data point.

[0058] Specifically, this step considers that the voltage data points will have certain fluctuations for a period of time after the UV curing machine is started, and will tend to be stable after a period of time. At the same time, during the operation of the UV curing machine, the voltage data points and the current data points will usually have a certain correlation, that is, when the voltage increases, the current will also increase accordingly, and vice versa.

[0059] Therefore, this step will determine the fluctuation possibility of each voltage data point by analyzing the correlation between each voltage data point and the current data point and the time interval between the time corresponding to the voltage data point and the current start time of the UV curing machine.

[0060] If the time interval between a voltage data point and the current start time of the UV curing machine is smaller, it can be explained that the possibility of the corresponding time of the voltage data point being in the early stage of the current start of the UV curing machine is greater, and the possibility of the corresponding time of the voltage data point existing voltage regular fluctuations is greater, and the noise probability of the voltage data point is smaller. At the same time, if the correlation between the voltage data point and the current data point at the corresponding time is stronger, it further increases the credibility of the possibility that the corresponding time of the voltage data point exists voltage regular fluctuations. According to the above dimensions, and combined with the obtained suspected noise factor of each voltage data point, the noise probability of the voltage data point is determined.

[0061] In one embodiment, the method for obtaining the fluctuation possibility of each voltage data point is as follows:

[0062] Obtain the time interval between the corresponding time of each voltage data point and the current start time of the UV curing machine, and the current data point at the corresponding time of each voltage data point;

[0063] Select all historical voltage data points consistent with each voltage data point, calculate the mean value of all historical current data points corresponding to the selected historical voltage data points as the reference current value of the voltage data point;

[0064] For example, the voltage data point The current data point at the corresponding time ,

[0065] All historical voltage data points of X5 are , , , , the current data point at the corresponding time is , , , ;

[0066] and is all historical voltage data points consistent with , and and the mean value 120 of the current data point at the corresponding time of and is the reference current value of the voltage data point .

[0067] Calculate the fluctuation possibility of each voltage data point, for the first voltage data point, its fluctuation possibility is:

[0068]

[0069] In this formula, is the fluctuation possibility of the first voltage data point, is the time interval between the corresponding time of the first voltage data point and the start time of the current UV curing machine, represents the absolute value of the difference between the current data point at the corresponding time of the first voltage data point and the reference current value of the voltage data point, is the natural exponential function. Reflecting the running stage, reflecting the current deviation, the product of the two constitutes a comprehensive index, which represents the size of the state deviation from the steady state at the corresponding time of the first voltage data point. The greater the product, the farther the deviation from the steady state, and the smaller the fluctuation possibility, and vice versa. Therefore, by build a negative correlation with the fluctuation possibility to meet the above logic.

[0070] The time interval between the current analysis time and the start time of the current curing machine is smaller, the first voltage data point is more likely to be in the early stage of the current curing machine startup, so the first voltage data point corresponding time exists voltage regular fluctuation possibility is greater, then the first The lower the noise probability of each voltage data point, the greater the probability of noise, and vice versa. The smaller the value, the better. The stronger the correlation between the voltage data point and the current data at the same time, the more it indicates that the voltage data point has a stronger correlation with the current data at the same time. The greater the probability of regular voltage fluctuations at a given voltage data point, the higher the reliability. The lower the noise probability of each voltage data point, the greater the probability of noise, and vice versa.

[0071] S23: Determine the noise probability of each voltage data point by combining the suspected noise factor and the fluctuation probability.

[0072] In one embodiment, the noise probability of each voltage data point is determined based on the following:

[0073] The normalized value of the ratio of the suspected noise factor to the fluctuation probability of each voltage data point is used as the noise probability of each voltage data point.

[0074] For the The formula for calculating the noise probability for each voltage data point is:

[0075]

[0076] In this formula, For the first Noise probability of each voltage data point Indicates the first Suspected noise factors for each voltage data point For the first The probability of fluctuation at each voltage data point As a normalization function, it maps the calculation results to a standard interval, giving them a clear probabilistic meaning.

[0077] In this formula, The smaller the value, the better. The more stable the voltage state of the UV curing machine at the corresponding time for each voltage data point, the less likely data fluctuations will occur. The larger, the more it means The larger the value, the more likely it is due to noise rather than data fluctuations. The greater the noise probability of the first voltage data point, the higher the probability of the second voltage data point. The more individual voltage data points there are, the more likely they are to be noisy data. Conversely, The larger the value, the more likely it is to be the first. At the corresponding time of the first voltage data point, the voltage state of the UV curing machine is relatively unstable, and the possibility of data fluctuations is greater at this time. The lower the noise probability of each voltage data point.

[0078] When the device is in stable operation, the value of is very small (tending to 0), which, as the denominator in the formula, significantly amplifies the effect of . This fully complies with the engineering logic: in a stable state, even a small voltage jump should be given high attention, and its probability as noise should be higher.

[0079] When the device is starting or power adjusting. At this time, the value of is larger (tending to 1), and the amplification effect on is greatly weakened or even suppressed, which makes the value of calculated from the expected voltage fluctuation in this stage relatively low, thereby effectively avoiding misjudgment of normal fluctuations as noise.

[0080] In this way, the present application no longer statically and independently considers each data jump, but comprehensively judges the data points in the dynamic time and working condition background in which they are generated. This dynamic evaluation mechanism based on the working condition improves the accuracy and robustness of noise recognition, and lays a solid data foundation for subsequent accurate clustering analysis and reliable fault diagnosis.

[0081] S3: Optimizing the agglomerative hierarchical clustering algorithm based on the noise probability of each voltage data point.

[0082] Specifically, the basis for determining two clusters in single-link aggregation in the recursive aggregation process of the agglomerative hierarchical clustering algorithm is optimized.

[0083] As shown in Figure 3 , the method comprises the following steps:

[0084] S31: Calculate the score value of each group of voltage data points of each group of clusters.

[0085] Through step S2, the noise probability of each voltage data point is obtained, and each group of clusters contains voltage data points contained therein, thereby obtaining a plurality of groups of voltage data points.

[0086] If the noise probability of a group of voltage data points is smaller, it means that this group of voltage data points is more reliable, and the possibility of being used as the aggregation basis of the two clusters to which it belongs is greater, and the corresponding score value is also higher. The distance between a group of data points refers to the absolute value of the difference between the two data points (numerical value), and the smaller the distance, the smaller the difference between the two data points, and the greater the possibility of being used as the aggregation basis of the two clusters to which it belongs, and the corresponding score value will also be higher.

[0087] In one embodiment, the score value of each group of voltage data points of each group of clusters is calculated based on the following formula:

[0088]

[0089] In the formula, is the score value of the first group voltage data point of the first group cluster, the higher the score value, the more suitable the first group voltage data point as a group of voltage data points on which the first group cluster depends in single-link aggregation, and respectively are the noise probability of the first voltage data point and the noise probability of the second voltage data point of the first group voltage data point of the first group cluster, the noise probability reflects the reliability of the voltage data point, the higher the noise probability, the greater the possibility of the voltage data point being interfered by noise, and the lower the reliability of the voltage data point; on the contrary, the lower the noise probability, the more reliable the voltage data point. is the distance between the two voltage data points of the first group voltage data point of the first group cluster, is a natural exponential function.

[0090] In the formula, is the sum of the noise probabilities, the smaller the sum of the noise probabilities, the more reliable the two noise data points, the higher the score value of the two noise data points, and the greater the possibility of being the basis for aggregation, that is, and present a negative correlation, the smaller the distance, the closer the two noise data points, and the greater the possibility of being the basis for aggregation, and also present a negative correlation. Therefore, part of the negative correlation between and is constructed by the negative exponential function, and and present a negative correlation.

[0091] Through the formula, the score value of each group voltage data point of each group cluster is obtained, so that the possibility of different groups of voltage data points as the basis for aggregation can be intuitively compared by the size of the score value, and it is convenient to select the voltage data points on which each group cluster depends in single-link aggregation.

[0092] S32: determining a group of voltage data points on which each group cluster depends in single-link aggregation based on the score value.

[0093] The score value of all group voltage data points of each group of clusters is calculated, and the group voltage data point with the highest score value is selected as the group voltage data point on which the single-link aggregation of the group of clusters is based.

[0094] For example, A and B are a group of clusters, voltage data point A1 in A and voltage data point B1 in B are a group of voltage data points, the score value is 0.8, voltage data point A2 in A and voltage data point B2 in B are a group of voltage data points, the score value is 0.5, voltage data point A3 in A and voltage data point B3 in B are a group of voltage data points, the score value is 0.6, and A1 and B1 are selected as the group voltage data points on which the single-link aggregation of A and B is based.

[0095] In this way, the group voltage data points on which the single-link aggregation of any group of clusters is based are obtained.

[0096] S33: Calculate the inter-cluster distance based on the group voltage data points on which the single-link aggregation is based, and perform the clustering operation.

[0097] S331: Obtain the group voltage data points on which the single-link aggregation of each group of clusters is based.

[0098] For example, A1 and B1 are obtained as the group voltage data points on which the single-link aggregation of A and B is based.

[0099] S332: Calculate the inter-cluster distance.

[0100] The absolute value of the difference between the two voltage data points included in the group voltage data points on which the single-link aggregation of each group of clusters is based is taken as the distance between the two clusters included in the group of clusters.

[0101] For example, A and B are a group of clusters, voltage data point A1 in A and voltage data point B1 in B are a group of voltage data points, and A1 and B1 are the group voltage data points on which the single-link aggregation of A and B is based. The absolute value of the difference between A1 and B1 is taken as the distance between A and B.

[0102] In this way, the distance between the two clusters included in any group of clusters, i.e., the distance between any two clusters, can be obtained, and these distances are recorded.

[0103] S333: Recursively aggregate and generate a clustering result tree.

[0104] In the recursive aggregation process, for each cluster, the distances between the cluster and all other clusters are sorted, and the cluster with the closest distance is selected for merging to form a new cluster. After merging, the new cluster is re-computed with the distances to the remaining clusters according to steps S1 (continuously collecting voltage data points of the UV curing machine during operation) to S3. Then, the new cluster is merged with the cluster with the closest distance, and this process is repeated. Through this step-by-step merging, a complete hierarchical clustering result tree is finally generated, which shows the merging relationship between clusters and the hierarchical structure of the clustering process.

[0105] As shown in Figure 4

[0106] Initial state: There are 5 clusters, C1, C2, C3, C4, and C5.

[0107] First iteration: Take any two clusters as a group, obtain a set of voltage data points for single-linkage aggregation of each cluster in the group, and calculate the distance between clusters based on this set of voltage data points. The distance between C1 and C2 is the smallest, i.e., the association between them is the closest, so C1 and C2 are merged into a new cluster, denoted as C12.

[0108] Second iteration: The remaining clusters are C12, C3, C4, and C5. Again, obtain a set of voltage data points for single-linkage aggregation of each cluster in the group, and calculate the distance between clusters based on this set of voltage data points. At this time, the distance between C4 and C5 is the smallest, so C4 and C5 are merged into a new cluster, denoted as C45.

[0109] Third iteration: The remaining clusters are C12, C3, and C45. Continue to obtain a set of voltage data points for single-linkage aggregation of each cluster in the group, and calculate the distance between clusters based on this set of voltage data points. At this time, the distance between C3 and C45 is the smallest, so C3 and C45 are merged into a new cluster, denoted as C345.

[0110] Fourth iteration: The last remaining clusters are C12 and C345, and their distance is calculated. They are merged into a final cluster, denoted as C12345, which is the root node of the hierarchical clustering result tree.

[0111] Through this iterative merging process, a final hierarchical clustering result tree is formed, and the distance between each two merged clusters is recorded, showing the process of merging into different levels of clusters.

[0112] ​S4: Using the optimized agglomerative hierarchical clustering algorithm to detect anomalies in the voltage data points during the operation of the UV curing machine.

[0113] Specifically, the agglomerative hierarchical clustering algorithm is used to cluster the voltage data points during the operation of the UV curing machine, and a hierarchical clustering result tree is obtained. Each non-leaf node of the tree is a fused new cluster, which is fused from the clusters corresponding to its child nodes. Each cluster contains multiple voltage data points.

[0114] In the agglomerative hierarchical clustering algorithm, a preset distance threshold is used to filter and judge the clustering results, which is a general method for identifying abnormal clusters:

[0115] In the recursive aggregation process of the agglomerative hierarchical clustering algorithm, the preset distance threshold is 0.6 (empirical value). If the inter-cluster distance of a cluster and another cluster when merging is less than 0.6, the merging relationship of the two clusters will be retained, and the merged cluster will be analyzed as a whole cluster, and the whole cluster will be considered as a normal cluster. Because when the inter-cluster distance is less than the distance threshold, it means that the clusters are similar in features or attributes, which conforms to the clustering pattern of the overall data, so it can be retained as a normal cluster.

[0116] If the inter-cluster distance of a cluster and another cluster when merging is greater than or equal to 0.6, the merging relationship of the two clusters will be cut off, and the two clusters will be separated and treated as two suspected abnormal clusters. Because when the inter-cluster distance is greater than or equal to the threshold, it means that the two clusters have large differences in features or attributes from other clusters, which is not consistent with the clustering pattern of the overall data, so they are separated and further analyzed as suspected abnormal clusters.

[0117] This operation of identifying suspected abnormal clusters through distance threshold can quickly and effectively filter out data that may be abnormal from a large number of clustering results, which helps to further explore the causes of these anomalies.

[0118] S5: According to the anomaly detection result, evaluate whether the UV curing machine has a voltage fault.

[0119] During the operation of the UV curing machine, the collected voltage data points are clustered and analyzed to detect anomalies, and then determine whether there is a voltage fault.

[0120] The specific steps and judgment criteria are as follows:

[0121] First, determine the abnormal voltage data points: the preset number threshold is 5 (empirical value), after completing the cluster analysis of the voltage data points and the preliminary determination of the suspected abnormal clusters, further screening is performed on each suspected abnormal cluster. If the number of voltage data points of a suspected abnormal cluster is greater than 5, it is considered that the suspected abnormal cluster may be a false abnormality caused by data fluctuations or other non-critical factors. If the number of voltage data points of a suspected abnormal cluster is less than or equal to 5, and the remaining suspected abnormal clusters are formally determined as abnormal clusters.

[0122] Because the agglomerative hierarchical clustering algorithm is based on the distance between clusters for merging. When the number of data points in a cluster is small, it may be in a relatively isolated position in the clustering space, far from other clusters. This means that the data points in this cluster have a large difference in features or attributes from most other data points, and do not conform to the overall data distribution pattern. If it is random fluctuations, data points should be more evenly distributed in each cluster, rather than concentrated in a few isolated small clusters.

[0123] Next, evaluate whether the UV curing machine has a voltage failure.

[0124] The preset proportion threshold is 2% (empirical value); if the number of abnormal voltage data points accounts for more than 2% of the total number of voltage data points within M minutes before the current analysis time (when the abnormality of the current analysis voltage data point is analyzed), it is determined that the UV curing machine has a voltage failure, and relevant personnel are notified for inspection; if the number of abnormal voltage data points accounts for less than 2% of the total number of voltage data points within M minutes before the current analysis time, it is determined that the UV curing machine does not have a voltage failure, wherein M is a preset value, and the value is 5 (empirical value). This dynamic evaluation method based on a time window takes into account the dynamic changes in the running state.

[0125] By monitoring recent data, real-time fluctuations in the voltage of the UV curing machine can be captured in a timely manner. If the proportion of abnormal data points exceeds the threshold within a short period of time, it indicates that the UV curing machine may have a problem and needs to be intervened in a timely manner, and relevant personnel are notified for inspection, which can avoid further damage to the UV curing machine due to continued operation and ensure the safe and stable operation of the UV curing machine.

Claims

1. A UV curing machine failure detection method based on data analysis, characterized by, The method comprises the following steps: continuously collecting voltage data points of the UV curing machine during operation, and dividing a surrounding data segment of each voltage data point; Determine a suspected noise factor of each voltage data point based on a difference between the voltage data point and a mean of data segments around the voltage data point, and a difference between the voltage data point and adjacent voltage data points on left and right sides of the voltage data point, including: calculating an absolute value of a difference between the first voltage data point and an adjacent data point on a left side of the first voltage data point, and an absolute value of a difference between the first voltage data point and an adjacent data point on a right side of the first voltage data point, and recording a larger one of the absolute values as the suspected noise factor of the first voltage data point. ;​​ No. Suspected noise factors at each voltage data point In the formula, For the first The absolute value of the difference between a voltage data point and the mean of its surrounding data segments. For hyperparameters; the noise probability of each voltage data point is determined based on the suspected noise factor and the fluctuation probability of each voltage data point, including: using the normalized value of the ratio of the suspected noise factor to the fluctuation probability of each voltage data point as the noise probability of each voltage data point; obtaining the fluctuation possibility of each voltage data point: obtaining the time interval between the corresponding time of each voltage data point and the starting time of the UV curing machine, and the current data point at the corresponding time of each voltage data point; The fluctuation possibility of each voltage data point: , wherein, is the fluctuation possibility of the i th voltage data point, is the time interval between the corresponding moment of the i th voltage data point and the starting moment of the UV curing machine, is the absolute value of the difference between the current data point at the corresponding moment of the i th voltage data point and the reference current value of the voltage data point, is a natural exponential function; using the condensed hierarchical clustering algorithm to cluster the voltage data points to obtain a plurality of clusters, in the recursive aggregation process of the plurality of clusters, taking each two clusters as a cluster group, selecting one voltage data point from each of the two clusters of each cluster group to form a group of data points, and determining the score value of each group of voltage data points of each cluster group according to the noise probability and distance of each group of data points; determining a group of voltage data points that each cluster group relies on during single-link aggregation according to the score value of each group of voltage data points of each cluster group, so as to optimize the condensed hierarchical clustering algorithm; using the optimized condensed hierarchical clustering algorithm to perform anomaly detection on the voltage data points of the UV curing machine during operation, and determining whether the UV curing machine has a voltage fault according to the anomaly detection result.

2. The UV curing engine failure detection method of claim 1, wherein, The score value of each group of voltage data points of each cluster group is calculated based on the following formula: ； In the formula, For the first The first cluster The score of the group voltage data points and The first The first cluster The noise probability of the first and second voltage data points in a group of voltage data points. For the first The first cluster The distance between two voltage data points in a group of voltage data points. It is a natural exponential function, where the distance between two voltage data points is determined based on the absolute value of the difference between the two voltage data points.

3. The UV curing engine failure detection method of claim 1, wherein, The determination method of the group of voltage data points that each cluster group relies on during single-link aggregation is: calculate the score value of all groups of voltage data points of each cluster group, and select a group of voltage data points with the highest score value as the group of voltage data points that the cluster group relies on during single-link aggregation.

4. The UV curing engine failure detection method of claim 1, wherein, The anomaly detection on the voltage data points of the UV curing machine during operation comprises: in the process of recursive aggregation in the condensed hierarchical clustering algorithm, a distance threshold is preset, if the inter-cluster distance of a cluster and another cluster when they are merged is less than the distance threshold, the cluster after the cluster and the other cluster are merged is regarded as a normal cluster; if the inter-cluster distance of a cluster and another cluster when they are merged is greater than or equal to the distance threshold, the cluster and the other cluster are separated and treated as two suspected abnormal clusters; a quantity threshold is preset; if the number of voltage data points of a suspected abnormal cluster is greater than the preset quantity threshold, the suspected abnormal cluster is removed, and the remaining suspected abnormal clusters are regarded as abnormal clusters, and the voltage data points contained in the abnormal clusters are regarded as abnormal voltage data points, so as to complete the anomaly detection on the voltage data points.

5. The UV curing engine failure detection method of claim 1, wherein, The method for determining whether the UV curing machine has a voltage fault according to the anomaly detection result is: a proportion threshold is preset; if the proportion of the number of abnormal voltage data points to the total number of voltage data points exceeds the proportion threshold within M minutes before the current analysis time, it is determined that the UV curing machine has a voltage fault, and relevant personnel are notified to check; if the proportion of the number of abnormal voltage data points to the total number of voltage data points does not exceed the proportion threshold within M minutes before the current analysis time, it is determined that the UV curing machine does not have a voltage fault; M is a preset value.

6. The UV curing engine failure detection method of claim 1, wherein, The method for dividing the surrounding data segment of each voltage data point is: taking each voltage data point as the center, selecting a preset number of voltage data points on both sides of the voltage data point to form the surrounding data segment of the voltage data point.

7. The UV curing engine fault detection method of claim 1, wherein, The reference current value of the voltage data point is determined based on the following method: For any one voltage data point, all historical voltage data points consistent with the voltage data point are selected, and the mean value of all historical current data points corresponding to the selected historical voltage data points is calculated as the reference current value of the voltage data point.

Images