A packaging machine fault intelligent detection method and system

Abstract

The present invention relates to the field of data processing technology, and more specifically, to a packaging machine fault intelligent detection method and system, the method comprising: obtaining the average daily running speed of the packaging machine in a complete service life, obtaining a reference data sequence, determining the noise level of the reference data sequence by using the difference between the data change amount of each reference data and the data change amount of the expected running speed at the corresponding time, clustering the data change amount of each reference data, calculating the data complexity of the reference data sequence by using the obtained discrete degree of each clustering cluster and the overall clustering performance of all clustering clusters, and then determining the penalty parameter of the reference data sequence in combination with the noise level and data complexity of the reference data sequence, so as to use the penalty parameter to assist in training an anomaly detection model, and perform fault detection on the packaging machine. The present invention can improve the accuracy of abnormal data detection.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and more specifically, to a packaging machine fault intelligent detection method and system. Background Art

[0002] Since metal powder packaging machines are affected by many factors during high-speed operation, such as mechanical wear, electrical failure, dust pollution, etc., various faults are prone to occur. These faults will not only cause production line shutdowns and affect production progress, but may also pollute metal powders and reduce product quality. With the rapid development of artificial intelligence, big data, and machine learning technologies, intelligent detection methods have gradually become an effective way to solve these problems.

[0003] In the related technology, a method and system for detecting abnormal network data is disclosed in the patent application document with publication number CN118035648A. The method includes: adding a penalty term for the complexity of a decision subtree to a focal loss function, minimizing the focal loss function with the added penalty term, assigning weights to a plurality of pre-trained weak learners based on decision subtrees, linearly combining each weak learner based on the corresponding weight to obtain a strong learner, and obtaining network abnormal data based on the strong learner and the pre-processed network data to be detected.

[0004] However, when adding the penalty item of decision subtree complexity, the above scheme does not take into account the influence of noise data and the diversity and complexity of the intrinsic patterns of the data itself, so that the added penalty item may not accurately control the complexity of the anomaly detection model, resulting in overfitting and underfitting of the data when the anomaly detection model is used to detect anomalies on the data, which may reduce the accuracy of the determined anomaly detection results to a certain extent. Summary of the invention

[0005] In order to solve the problem of low accuracy in abnormal data detection, the present invention proposes a packaging machine fault intelligent detection method and system.

[0006] According to a first aspect of an embodiment of the present invention, there is provided a packaging machine fault intelligent detection method, comprising:

[0007] Obtain the average daily operating speed of the packaging machine during its complete service life to obtain a reference data sequence;

[0008] Determine the expected running speed of each reference data in the reference data sequence at the corresponding time, calculate the difference between the data change amount of each reference data and the data change amount of the expected running speed at the corresponding time, count the number of differences greater than a preset threshold, and use the ratio of the number to the number of all reference data as the noise level of the reference data sequence;

[0009] Cluster the data variation of each reference data to obtain multiple clusters, determine the discrete degree of each cluster, and calculate the data complexity of the reference data sequence. The data complexity is positively correlated with the discrete degree and negatively correlated with the overall clustering performance of all clusters. The overall clustering performance reflects the clustering effect of clustering the data variation of the reference data;

[0010] The penalty parameter of the reference data sequence is calculated. The penalty parameter is positively correlated with the noise level and the complexity of the data. The penalty parameter is used to assist in training the pre-built anomaly detection model so as to perform fault detection on the packaging machine based on the trained anomaly detection model.

[0011] The present invention determines the penalty parameter of the reference data sequence in combination with the noise level and data complexity of the reference data sequence, and can accurately control the complexity of the anomaly detection model based on the penalty parameter auxiliary training, thereby avoiding overfitting and underfitting of the anomaly detection model when performing anomaly detection on the reference data sequence, and further improving the accuracy of abnormal data detection in the reference data sequence; and when determining the penalty parameter of the reference data sequence, it is based on the data characteristics of the reference data sequence itself, which can reduce the difficulty of integrating and deploying the anomaly detection model into the packaging machine for fault detection.

[0012] In an exemplary embodiment of the present invention, the penalty parameter of the reference data sequence satisfies the following relationship:

[0013] ;

[0014] In the formula, is the penalty parameter of the reference data sequence; is the noise level of the reference data series; is the complexity of the reference data series; For natural numbers An exponential function with base .

[0015] In an exemplary embodiment of the present invention, a penalty parameter is used to assist in training a pre-built anomaly detection model, so as to perform fault detection on a packaging machine based on the trained anomaly detection model, including:

[0016] The pre-built anomaly detection model is trained using the pre-acquired training sample set. During the training process, the degree of penalty for misclassifying the training sample set is controlled by the penalty parameter. When the minimization loss function of the anomaly detection model tends to converge, the trained anomaly detection model is obtained.

[0017] The reference data sequence is input into the trained anomaly detection model for anomaly detection, so as to detect the fault of the packaging machine according to the results of anomaly detection.

[0018] The present invention utilizes the calculated penalty parameter to assist in training a pre-built anomaly detection model, which can accurately control the complexity of the anomaly detection model, thereby improving the accuracy of the anomaly detection result.

[0019] In an exemplary embodiment of the present invention, the data complexity of the reference data sequence satisfies the following relationship:

[0020] ;

[0021] In the formula, is the data complexity of the reference data series; For the The degree of dispersion of clusters; is the number of clusters; is the overall clustering performance of all clusters; is the summation symbol.

[0022] The present invention determines the data complexity of the reference data sequence by comprehensively considering the discreteness of each cluster and the overall clustering performance of all clusters, thereby ensuring the accuracy of the determined data complexity.

[0023] In an exemplary embodiment of the present invention, a method for obtaining the overall clustering performance of all clusters includes:

[0024] Obtain the distance between the cluster centers of each cluster, and use the cluster with the shortest distance to the cluster center of any cluster as the nearest neighbor cluster of any cluster, and then determine the nearest neighbor cluster of each cluster;

[0025] The average value of the distance between the cluster center of each cluster and the cluster center of the nearest neighbor cluster is calculated, and the reciprocal of the average distance is taken as the overall clustering performance of all clusters.

[0026] The present invention calculates the distance between the cluster centers of each cluster, can evaluate the similarity between the clusters, and then use the similarity to determine the overall clustering performance of all clusters, and can accurately evaluate the clustering effect of clustering the data change amount of each reference data.

[0027] In an exemplary embodiment of the present invention, the degree of discreteness of the clusters satisfies the following relationship:

[0028] ;

[0029] In the formula, For the The degree of dispersion of clusters; For the The first The amount of data change corresponding to each data point; For the The amount of data change corresponding to the cluster center of each cluster; For the The number of data points in a cluster; is the absolute value symbol.

[0030] In an exemplary embodiment of the present invention, a method for obtaining a data variation of reference data includes:

[0031] Calculate the mean of the difference between any reference data except the first and last positions in the reference data sequence and the left and right adjacent reference data to obtain the data change amount of the corresponding reference data, and then determine the data change amount of all reference data except the first position in the reference data sequence;

[0032] The difference between any reference data at the first and last positions in the reference data sequence and the reference data adjacent to it on one side is calculated to obtain the data change amount of the corresponding reference data, thereby determining the data change amount of the reference data at the first and last positions in the reference data sequence.

[0033] The present invention calculates the data change amount of each reference data, can capture the dynamic change trend and situation of the reference data, and then uses the difference between the dynamic change of the actual running speed and the expected running speed to determine the noise level of the reference data sequence, which can ensure the accuracy of the determined noise level.

[0034] In an exemplary embodiment of the present invention, determining the expected running speed corresponding to the time of each reference data in the reference data sequence includes:

[0035] Perform least squares fitting on all reference data in the reference data sequence to obtain the fitting value corresponding to each reference data. The fitting value corresponding to any reference data is used as the expected running speed at the corresponding time of any reference data, thereby determining the expected running speed at the corresponding time of each reference data in the reference data sequence.

[0036] According to a second aspect of an embodiment of the present invention, a packaging machine fault intelligent detection system is provided. The packaging machine fault intelligent detection system includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the first aspect of the embodiment of the present invention.

[0037] The present invention has the following effects:

[0038] The present invention can capture the difference between the dynamic change of the reference data and the corresponding expected running speed by calculating the difference between the data change amount of each reference data and the data change amount of the expected running speed at the corresponding time, thereby ensuring the accuracy of the determined noise level, and by clustering the data change amount of each reference data and comprehensively obtaining multiple features of the clustering clusters to determine the data complexity of the reference data sequence, the accuracy of the determined data complexity can be ensured, so that the penalty parameter of the reference data sequence can be accurately determined, so as to accurately control the complexity of the anomaly detection model, thereby avoiding underfitting and overfitting phenomena when performing anomaly detection on the reference data sequence, and further improving the accuracy of abnormal data detection, and when determining the penalty parameter of the reference data sequence, it is based on the data characteristics of the reference data sequence itself, which can reduce the difficulty of integrating and deploying the anomaly detection model into the packaging machine for fault detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] By reading the following detailed description with reference to the accompanying drawings, the above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0040] Figure 1 The present invention is a schematic diagram of the steps of an intelligent detection method for packaging machine faults according to an embodiment of the present invention. DETAILED DESCRIPTION

[0041] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0042] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0043] Reference Figure 1 , a packaging machine fault intelligent detection method, including steps S1 to S4, specifically as follows:

[0044] S1: Obtain the average daily operating speed of the packaging machine during its complete service life to obtain a reference data sequence.

[0045] Specifically, a packaging machine can be randomly selected from packaging machines of the same model in the same usage environment, and the operating speed data of the selected packaging machine during the complete usage cycle can be obtained. Then, the obtained operating speed data belonging to the same day can be averaged to obtain the corresponding daily average operating speed, thereby determining the daily average operating data within the complete usage cycle and obtaining a reference data sequence.

[0046] Among them, the reference data in the reference data sequence is the average daily operating data, and the complete use cycle is the working cycle of using the packaging machine to complete the packaging of one or more products. It should be noted that since the operating speed of the packaging machine also changes within a certain range over time, using the average daily operating speed as reference data can accurately evaluate the overall level of the operating speed of the packaging machine on any day within the complete use cycle, thereby ensuring the accuracy of the determined operating conditions of the packaging machine.

[0047] S2: Determine the expected running speed of each reference data in the reference data sequence at the corresponding time, calculate the difference between the data change amount of each reference data and the data change amount of the expected running speed at the corresponding time, count the number of differences greater than a preset threshold, and use the ratio of the number to the number of all reference data as the noise level of the reference data sequence.

[0048] It should be noted that, since the packaging machine needs to adapt to the packaging needs of different types of products, the operating speed of the packaging machine often has different expected operating speeds at different times, that is, the expected operating speed of the packaging machine should be data that changes dynamically over time. If the difference between the reference data and the expected operating speed at the corresponding time is simply compared, it may not be possible to fully capture the dynamic characteristics of the data changing over time. The data change amount used in the present invention can reflect the changing trend and speed of the reference data or the expected operating speed, thereby being able to more accurately reflect the dynamic behavior of the reference data or the expected operating speed, and further ensure the accuracy of the degree to which each reference data is affected by noise.

[0049] In an exemplary embodiment of the present invention, the reference data refers to the average daily operating data of the packaging machine on any day within a complete use cycle, which is used to determine the operating condition of the packaging machine; the expected operating speed can be an ideal value of the packaging machine's operating speed, which is used as control data to evaluate the extent to which the reference data is affected by noise.

[0050] The data variation may be data that can reflect the dynamic change of any reference data or any expected operating speed, and is used to determine the degree to which the reference data is affected by noise. For example, when the data variation of any reference data is large, it means that the operating speed of the packaging machine changes rapidly at the time corresponding to the reference data. If the data variation of the expected operating speed at the corresponding time is small, it means that the ideal value of the operating speed of the packaging machine changes slowly at the time, which means that the reference data corresponding to the time is greatly affected by noise. Therefore, the difference between the data variation of the reference data and the data variation of the corresponding expected operating speed can be used to evaluate the degree to which the reference data is affected by noise.

[0051] The noise level may be data that can reflect the amount of noise data in the reference data sequence, and is used as a data basis to adjust the sensitivity to noise data when performing anomaly detection on the reference data sequence. For example, when the noise level of the reference data sequence is large, it means that the amount of noise data in the reference data sequence is large, and when performing anomaly detection on the reference data sequence, it is necessary to reduce the sensitivity to noise data, so as to avoid overfitting of the reference data in the reference data sequence.

[0052] Optionally, when the noise level of the reference data sequence is small, it means that the amount of noise data in the reference data sequence is small. Then, when performing anomaly detection on the reference data sequence, it is necessary to increase the sensitivity to the noise data, so as to avoid underfitting of the reference data in the reference data sequence, and then accurately detect the abnormal data in the reference data sequence.

[0053] In an exemplary embodiment of the present invention, the expected running speed of each reference data corresponding to the time in the reference data sequence can be determined by the following steps:

[0054] Perform least squares fitting on all reference data in the reference data sequence to obtain the fitting value corresponding to each reference data. The fitting value corresponding to any reference data is used as the expected running speed at the corresponding time of any reference data, thereby determining the expected running speed at the corresponding time of each reference data in the reference data sequence.

[0055] Optionally, by fitting each reference data in the reference data sequence using the least squares method, the expected running speed of each reference data corresponding to the time can be accurately determined in a simple manner, thereby reducing the difficulty of determining the expected running speed. Among them, least squares fitting of the data set is a prior art and is not described in detail in this embodiment.

[0056] In an exemplary embodiment of the present invention, the determination of the data change amount of the reference data can be achieved by the following steps:

[0057] Calculate the mean of the difference between any reference data except the first and last positions in the reference data sequence and the adjacent reference data on the left and right to obtain the data change amount of the corresponding reference data, and then determine the data change amount of all reference data except the first position in the reference data sequence; calculate the difference between any reference data at the first and last positions in the reference data sequence and the reference data adjacent on one side to obtain the data change amount of the corresponding reference data, and then determine the data change amount of the reference data at the first and last positions in the reference data sequence.

[0058] Specifically, the data change amount of any reference data except the first and last positions in the reference data sequence is calculated to satisfy the following relationship:

[0059] ;

[0060] In the formula, For the The data change of each reference data; For the Reference data; For the Reference data; For the Reference data; is the absolute value symbol.

[0061] Optionally, when calculating the data change amount of the first reference data in the reference data sequence, the absolute value of the difference between the first reference data and the second reference data can be used as the data change amount of the first reference data; when calculating the data change amount of the last reference data in the reference data sequence, if the number of reference data in the reference data sequence is , then the data change of the last reference data is Reference data and The absolute value of the difference between the reference data amounts can be used to obtain the data change amount of each reference data in the reference data sequence.

[0062] It should be noted that the method for determining the data change amount of the expected running speed of each reference data in the reference data sequence is exactly the same as the method for determining the data change amount of each reference data in the reference data sequence, and is not elaborated herein in this embodiment.

[0063] Optionally, the difference between the data change amount of each reference data and the data change amount of the corresponding expected running speed may be calculated, and the difference satisfies the following relationship:

[0064] ;

[0065] In the formula, For the The difference between the data change amount of the reference data and the data change amount of the corresponding expected running speed; For the The data change of each reference data; For the The data change of the expected running speed corresponding to the reference data; is the absolute value symbol.

[0066] in, Can reflect the The larger the value, the greater the influence of noise data on the reference data, and the greater the possibility that the corresponding reference data is noise data. On the contrary, the smaller the value, the smaller the influence of noise data on the reference data, and the smaller the possibility that the corresponding reference data is noise data.

[0067] Optionally, the preset threshold can be set to , then count Medium to large The amount of data is recorded as , then the noise level of the reference data series can be calculated ,in, is the difference between the data change of the first reference data and the data change of the corresponding expected running speed, is the difference between the data change of the second reference data and the data change of the corresponding expected running speed, For the The difference between the data change of the reference data and the data change of the corresponding expected running speed, is the number of reference data in the reference data sequence.

[0068] S3: Cluster the data changes of each reference data to obtain multiple clusters, determine the discreteness of each cluster, and calculate the data complexity of the reference data sequence. The data complexity is positively correlated with the discreteness and negatively correlated with the overall clustering performance of all clusters. The overall clustering performance reflects the clustering effect of clustering the data changes of the reference data.

[0069] In an exemplary embodiment of the present invention, the degree of dispersion may be data that measures the density of data distribution in the clusters, and is used to determine the data complexity of the reference data sequence. For example, when the overall level of the degree of dispersion of all clusters is high, it means that the data distribution in all clusters is relatively discrete, and the distribution of the reference data in the reference data sequence is relatively chaotic, and the corresponding data complexity of the reference data sequence is relatively high.

[0070] Specifically, the degree of discreteness of each cluster satisfies the following relationship:

[0071] ;

[0072] In the formula, For the The degree of dispersion of clusters; For the The first The amount of data change corresponding to each data point; For the The amount of data change corresponding to the cluster center of each cluster; For the The number of data points in a cluster.

[0073] The data complexity may be data that can reflect the complexity of the data structure of the reference data sequence and the similarity characteristics between the reference data in the reference data sequence, and the overall clustering performance may be data that can reflect the clustering effect of clustering the data variation of each reference data, which is used to determine the similarity characteristics between the reference data in the reference data sequence, and further determine the data complexity of the reference data sequence. For example, when the overall clustering performance is larger, it means that the clustering effect of clustering the data variation of each reference data is better, and the corresponding similarity between the reference data in the reference data sequence is larger, which means that the data complexity of the reference data sequence is relatively lower.

[0074] Optionally, K-Means (k-means clustering algorithm) can be used to cluster the data changes of each reference data, or other clustering algorithms such as K center point clustering algorithm, i.e., K-Mediods clustering algorithm, etc. can be used for clustering. This embodiment does not specifically limit the selection of clustering algorithm.

[0075] In an exemplary embodiment of the present invention, the overall clustering performance of all clusters may be determined by the following steps:

[0076] Get the distance between the cluster centers of each cluster, take the cluster with the shortest distance to the cluster center of any cluster as the nearest neighbor cluster of any cluster, and then determine the nearest neighbor cluster of each cluster; calculate the average value of the distance between the cluster center of each cluster and the cluster center of the nearest neighbor cluster, and take the inverse of the average value of the distance as the overall clustering performance of all clusters.

[0077] Optionally, the Euclidean distance between the cluster centers of each cluster can be calculated, and then the cluster with the shortest Euclidean distance to the cluster center of any cluster is taken as the nearest neighbor cluster of the cluster, and then the nearest neighbor cluster of each cluster is determined, and the average value of the Euclidean distance between the cluster center of each cluster and the cluster center of the nearest neighbor cluster is calculated, and then the inverse of the average value is taken as the overall clustering performance of all clusters.

[0078] Optionally, the cosine distance between the cluster centers of each cluster may be calculated to determine the nearest neighbor cluster of each cluster, and then determine the overall clustering performance of all clusters. The present invention does not specifically limit the distance type used to determine the overall clustering performance of all clusters.

[0079] Furthermore, after determining the degree of discreteness of each cluster and the overall clustering performance of all clusters, the data complexity of the reference data sequence can be calculated. Specifically, the data complexity of the reference data sequence satisfies the following relationship:

[0080] ;

[0081] In the formula, is the data complexity of the reference data series; For the The degree of dispersion of clusters; is the number of clusters; is the overall clustering performance of all clusters; is the summation symbol.

[0082] In another embodiment, the data complexity of the reference data sequence may be determined in other ways, such as by a difference calculation formula instead of a proportional calculation formula.

[0083] S4: Calculate the penalty parameter of the reference data sequence. The penalty parameter is positively correlated with the noise level and the complexity of the data. Use the penalty parameter to assist in training the pre-built anomaly detection model to perform fault detection on the packaging machine based on the trained anomaly detection model.

[0084] In an exemplary embodiment of the present invention, the penalty parameter refers to a parameter that can assist in training a pre-built anomaly detection model, determined based on the noise level and complexity of the reference data sequence, and is used to prevent overfitting and underfitting of the anomaly detection model when performing anomaly detection on the reference data sequence. For example, when the noise level of the reference data sequence is large, a larger penalty parameter can be set to avoid the anomaly detection model from being too strict in determining the boundary of normal data, so that the anomaly detection model will not pay too much attention to the noise data, which will lead to the occurrence of overfitting. When the complexity of the reference data sequence is high, it means that the reference data sequence may contain multiple fault modes or data with diverse normal data modes. That is, setting a larger penalty parameter can improve the flexibility of the anomaly detection model, so that the data can be flexibly fitted and the robustness of the anomaly detection model can be improved.

[0085] Specifically, the penalty parameter of the reference data sequence satisfies the following relationship:

[0086] ;

[0087] In the formula, is the penalty parameter of the reference data sequence; is the noise level of the reference data series; is the data complexity of the reference data series; For natural numbers An exponential function with base .

[0088] In another embodiment, the penalty parameter of the reference data sequence may also be determined by other methods, such as by a gamma transform calculation formula instead of a summation calculation formula.

[0089] Optionally, use By controlling the value of the penalty parameter within the range of 0-1, the difficulty of using the penalty parameter to assist in training the anomaly detection model can be reduced.

[0090] Among them, the anomaly detection model refers to a model built based on anomaly detection algorithm for detecting abnormal data in a reference data sequence. For example, a one-class support vector machine (One-Class Support Vector Machine, One-Class SVM) can be used to build an anomaly detection model. Of course, other supervised learning algorithms for anomaly detection can also be used to build an anomaly detection model, such as a support vector machine, etc., and then a kernel function, such as a Gaussian kernel function, a linear kernel function or a polynomial kernel function is used to map the reference data sequence to a high-dimensional space, thereby performing anomaly detection on the reference data sequence. It should be noted that the kernel function used in the present invention is the RBF kernel function, that is, the Gaussian kernel function.

[0091] In an exemplary embodiment of the present invention, the following steps can be used to implement fault detection of a packaging machine using a trained anomaly detection model:

[0092] The pre-constructed anomaly detection model is trained using the pre-acquired training sample set. During the training process, the penalty degree for misclassifying the training sample set is controlled by the penalty parameter. When the minimized loss function of the anomaly detection model tends to converge, the trained anomaly detection model is obtained. The reference data sequence is input into the trained anomaly detection model for anomaly detection, so as to detect faults of the packaging machine according to the results of the anomaly detection.

[0093] Specifically, the average daily operating speed of the selected packaging machine in any complete cycle can be used as a training sample set, and then the training sample set can be divided into a training set and a test value according to a certain ratio, and the normal data in the training sample set can be marked as 1, and the abnormal data can be marked as 0, so that the training set can be used to train the anomaly detection model built based on One-Class SVM. During the training process, the calculated penalty parameter is used to control the penalty degree for misclassifying the normal data and the abnormal data in the training set, so as to avoid overfitting and underfitting of the anomaly detection model, thereby improving the accuracy of the model in detecting abnormal data and the flexibility and robustness of the model. Then, the test set is used to evaluate the anomaly detection model. When the minimization loss function of the anomaly detection model tends to converge, a trained anomaly detection model is obtained. This embodiment does not specifically limit the division ratio of the training set and the sample set.

[0094] Furthermore, the trained anomaly detection model can be deployed in the packaging machine, and then the reference data sequence can be input into the trained anomaly detection model for anomaly detection. When abnormal data in the reference data sequence is detected, an alarm device such as a timely alarm can be used to sound an alarm, thereby reminding the operator that the packaging machine has failed at the corresponding time, and taking corresponding measures to repair the packaging machine in time, thereby ensuring the reliable operation of the packaging machine. It should be noted that the specific steps of deploying the trained model to the corresponding equipment are existing technologies and are not described in detail in this embodiment.

[0095] The present invention also provides a packaging machine fault intelligent detection system, the system includes a memory and a processor, and a computer program is stored in the memory, the computer program integrates the function of an intelligent packaging machine fault detection method, when the computer program is executed, the accuracy of abnormal data detection can be improved through an intelligent packaging machine fault detection method.

[0096] In the description of this specification, "plurality" or "several" means at least two, such as two, three or more, etc., unless otherwise clearly and specifically defined.

[0097] Although this specification has shown and described a number of embodiments of the present invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art will conceive of many modifications, changes and alternatives without departing from the ideas and spirit of the present invention. It should be understood that in the practice of the present invention, various alternatives to the embodiments of the present invention described herein may be employed.

Claims

1. An intelligent detection method for packaging machine faults, It is characterized in that include: Obtain the average daily operating speed of the packaging machine during its complete service life to obtain a reference data sequence; Determine the expected running speed of each reference data in the reference data sequence at the corresponding time, calculate the difference between the data change amount of each reference data and the data change amount of the expected running speed at the corresponding time, count the number of differences greater than a preset threshold, and use the ratio of the number to the number of all reference data as the noise level of the reference data sequence; Cluster the data variation of each reference data to obtain multiple clusters, determine the discrete degree of each cluster, and calculate the data complexity of the reference data sequence. The data complexity is positively correlated with the discrete degree and negatively correlated with the overall clustering performance of all clusters. The overall clustering performance reflects the clustering effect of clustering the data variation of the reference data; The degree of discreteness of clusters satisfies the following relationship: ; In the formula, For the The degree of dispersion of clusters; For the The first The amount of data change corresponding to each data point; For the The amount of data change corresponding to the cluster center of each cluster; For the The number of data points in a cluster; The data complexity of the reference data sequence satisfies the following relationship: ; In the formula, is the data complexity of the reference data series; is the number of clusters; is the overall clustering performance of all clusters; The method for obtaining the overall clustering performance of all clusters includes: obtaining the distance between the cluster centers of each cluster, taking the cluster with the shortest distance to the cluster center of any cluster as the nearest neighbor cluster of any cluster, and then determining the nearest neighbor cluster of each cluster; calculating the average value of the distance between the cluster center of each cluster and the cluster center of the nearest neighbor cluster, and taking the reciprocal of the average value of the distance as the overall clustering performance of all clusters; Calculate the penalty parameter of the reference data sequence. The penalty parameter is positively correlated with the noise level and the complexity of the data. Use the penalty parameter to assist in training the pre-built anomaly detection model so as to perform fault detection on the packaging machine based on the trained anomaly detection model. The penalty parameter of the reference data sequence satisfies the following relationship: ; In the formula, is the penalty parameter of the reference data sequence; is the noise level of the reference data series; is the data complexity of the reference data sequence.

2. According to claim 1, a packaging machine fault intelligent detection method, It is characterized in that The method of using the penalty parameter to assist in training a pre-built anomaly detection model, so as to perform fault detection on the packaging machine based on the trained anomaly detection model, comprises: The pre-constructed anomaly detection model is trained using the pre-acquired training sample set. During the training process, the degree of penalty for misclassifying the training sample set is controlled by the penalty parameter. When the minimization loss function of the anomaly detection model tends to converge, a trained anomaly detection model is obtained. The reference data sequence is input into a trained anomaly detection model to perform anomaly detection, so as to perform fault detection on the packaging machine according to the result of the anomaly detection.

3. According to claim 1, a packaging machine fault intelligent detection method, It is characterized in that The method for obtaining the data variation of the reference data includes: Calculate the mean of the difference between any reference data except the first and last positions in the reference data sequence and the left and right adjacent reference data to obtain the data change amount of the corresponding reference data, and then determine the data change amount of all reference data except the first position in the reference data sequence; The difference between any reference data at the head and tail positions in the reference data sequence and the reference data adjacent to the reference data on one side is calculated to obtain the data change amount of the corresponding reference data, thereby determining the data change amount of the reference data at the head and tail positions in the reference data sequence.

4. According to claim 1, a packaging machine fault intelligent detection method, It is characterized in that The determining of the expected running speed corresponding to the time of each reference data in the reference data sequence includes: Perform least squares fitting on all reference data in the reference data sequence to obtain fitting values ​​corresponding to each reference data, and use the fitting value corresponding to any reference data as the expected running speed at the time corresponding to the any reference data, thereby determining the expected running speed at the time corresponding to each reference data in the reference data sequence.

5. An intelligent fault detection system for packaging machines, It is characterized in that The packaging machine fault intelligent detection system comprises a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method as claimed in any one of claims 1 to 4.

Images