Insulated cable production quality detection method and system

Abstract

The invention relates to the technical field of quality detection, in particular to an insulated cable production quality detection method and system, and the method comprises the steps: constructing a plurality of real-time parameter features according to the production parameter sequence of a plurality of production devices; inputting the plurality of real-time features into isolated trees in an isolated forest, and weighting the output abnormal value according to the normalized preset weight of each isolated tree to obtain a comprehensive abnormal value; and comparing the comprehensive abnormal value with an abnormal threshold value to obtain a detection result. According to the technical scheme, the accurate quality detection result of the insulated cable can be obtained according to the production parameters of the production devices in the production process.

Description

Technical Field

[0001] The present application relates to the technical field of quality inspection, and in particular to a method and system for inspecting the production quality of insulated cables. Background Art

[0002] Insulated cables are power transmission devices built around insulated cores. Their typical structure consists of one or more insulated cores, a sheath, and a protective layer. The production quality of insulated cables directly impacts the efficiency and safety of power transmission. The production of insulated cables requires the participation of multiple production equipment, each of which corresponds to at least one production parameter. These parameters directly impact the quality of the insulated cables.

[0003] At present, the patent application document with application publication number CN114399237A discloses a cable status assessment and early warning method and device, the method including: obtaining product structure data of a first production cable; performing process complexity analysis on the first production cable according to the product structure data to obtain a first production complexity; if the first production complexity is within the preset production complexity, dividing the production process flow of the cable into process nodes to generate multiple identification nodes, wherein the multiple identification nodes include multiple detection units, and the multiple identification nodes correspond one-to-one to the multiple detection units; constructing a quality detection model library, wherein the quality detection model library is connected to the multiple detection units; by performing process flow feature analysis on the first identification node, performing model matching from the quality detection model library to obtain a first matching detection model; the first detection unit calls the first matching detection model in the quality detection model library to obtain the quality detection information output by the first production cable at the first identification node; by performing quality qualification judgment on the quality detection information output by the first identification node, if qualified, the second identification node is detected until the detection of the multiple identification nodes is completed.

[0004] The above method divides the cable production process into multiple identification nodes, configures a detection unit for each identification node, and the detection unit needs to call the detection model from the quality detection model library to realize the quality detection of each identification node; the accuracy of quality detection is closely related to the detection model, and the construction of the quality detection model library requires a large amount of labeled training data, the training cost is high, and the accuracy of the detection model cannot be guaranteed under limited data, resulting in inaccurate cable detection results. Summary of the Invention

[0005] In order to solve the technical problem of inaccurate cable inspection results, the present application provides an insulated cable production quality inspection method and system, which can obtain accurate cable quality inspection results based on the production parameters of each production equipment in the production process.

[0006] In a first aspect, the present application provides a method for detecting the production quality of an insulated cable, the method comprising: constructing a plurality of real-time parameter features based on production parameter sequences of a plurality of production equipment; inputting the plurality of real-time features into an isolated tree in an isolation forest, weighting the output outliers according to the normalized preset weights of each isolated tree to obtain a comprehensive outlier; comparing the comprehensive outlier with the outlier threshold to obtain a detection result; a method for obtaining the preset weights of each isolated tree comprising: using an annotated historical sample as a trust sample set; obtaining the outlier value of any historical sample in the trust sample set in each isolated tree, comparing the outlier value with the outlier threshold to obtain the detection result of each isolated tree on the historical sample, and calculating the initial weight of each isolated tree based on the inter-class distance of the trust sample set; using the initial weight to screen the trust sample in the unlabeled historical sample, and calculating the initial weight again in the updated trust sample set; iteratively updating the initial weight until there is no unlabeled historical sample or the number of iterations is greater than the preset number or no trust sample can be screened, and obtaining the preset weight of each isolated tree.

[0007] Based on the time series of production parameters of various production equipment, multiple real-time parameter features at the current moment are constructed and input into the isolation forest. Combined with the preset weights of each isolation tree, the misjudgment caused by average voting is effectively avoided.

[0008] Preferably, constructing multiple real-time parameter features includes: after preprocessing the production parameter sequence, taking the mean and standard deviation of the production parameter sequence in the time window at the current moment as the characteristic values ​​of the production parameters; the characteristic value of each production parameter corresponds to the multiple real-time parameter features.

[0009] By using the mean and standard deviation within the time window to construct real-time parameter features, it can not only reflect the current device status, but also capture its fluctuation trend, which helps to improve the early perception capability of anomaly detection.

[0010] Preferably, the preprocessing includes a normalization process.

[0011] Preferably, the time window of the current moment is the current moment and multiple historical moments before the current moment.

[0012] Preferably, the isolated tree The initial weight for: ; and Isolation Tree The average outlier value for confidence samples with quality label normal and quality label anomaly.

[0013] The initial weight is constructed using the difference between the average outlier values ​​of normal and abnormal samples to quantify the discriminative ability of each tree. The initial weight of the tree that cannot effectively distinguish between categories (normal and abnormal) is reset to 0, and the tree that cannot effectively distinguish between categories is eliminated, thereby ensuring the accuracy and convergence of the comprehensive outlier value.

[0014] Preferably, using initial weights to screen trust samples from unlabeled historical samples includes: obtaining the outlier value of each isolated tree for the unlabeled historical samples, if the outlier value is greater than the outlier threshold, the detection result of the corresponding isolated tree is abnormal, otherwise, the detection result of the corresponding isolated tree is normal; taking the isolated trees with normal detection results as the first set, and taking the isolated trees with abnormal detection results as the second set; calculating the weighted average of the outliers in the first set and the second set respectively based on the normalized initial weights; in response to the absolute value of the difference between the weighted average of the outliers between the first set and the second set being greater than a preset value, marking the unlabeled historical samples as trust samples.

[0015] The isolated trees are grouped according to the detection results, and the weighted average outlier difference is used to measure the sample credibility. This can expand the trust sample with very small amounts of labeled data, and then calculate the initial weights on the expanded trust sample set, so that the isolation forest has self-learning capabilities and improves its adaptability.

[0016] Preferably, the weighted average of the outliers in the first set is: taking the ratio of the initial weight of any isolated tree in the first set to the sum of the initial weights of all isolated trees in the first set as the normalized weight of the isolated tree, and weighting the outliers in the first set according to the normalized weight to obtain the weighted average of the outliers in the first set.

[0017] Preferably, before using the updated trust sample set to update the initial weights, the method for obtaining the preset weights of each isolated tree also includes: for unlabeled historical samples in the trust sample set, in response to the sum of the initial weights of the isolated trees with normal detection results being greater than the sum of the initial weights of the isolated trees with abnormal detection results, the quality label of the unlabeled historical samples is normal; otherwise, the quality label of the unlabeled historical samples is abnormal.

[0018] The quality labels of unlabeled historical samples are indirectly constructed based on unsupervised methods to reduce the workload of data labeling.

[0019] Preferably, after obtaining the preset weight of each isolated tree, the method for obtaining the preset weight of each isolated tree also includes: in response to the number of unlabeled historical samples being greater than a minimum allowable value, increasing the number of isolated trees in the isolation forest; the minimum allowable value is a preset proportion of the total number of historical samples.

[0020] If there are still a large number of unlabeled historical samples after obtaining the preset weights of each isolated tree, it means that there are still a large number of historical samples that cannot accurately judge the production quality based on the preset weights of each isolated tree. At this time, increasing the number of isolated trees in the isolation forest algorithm, constructing multiple isolated trees again, and obtaining the preset weights of all isolated trees can improve the robustness of the isolation forest algorithm and ensure the accuracy of the quality inspection results.

[0021] In a second aspect of the present application, an insulated cable production quality inspection system is also provided, comprising a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, an insulated cable production quality inspection method according to the first aspect of the present application is implemented.

[0022] The technical solution of this application has the following beneficial technical effects: During the production process of insulated cables, multiple real-time parameter features at the current moment are constructed based on the time series of production parameters of various production equipment, and are input into the isolation forest. Combined with the preset weights of each isolated tree, the misjudgment caused by average voting is effectively avoided, and accurate quality inspection results of the insulated cables are obtained. At the same time, in the process of determining the preset weights of the isolation trees, inter-class differences are introduced to iteratively update the initial weights of each isolated tree, gradually expanding the trust sample set, so that the isolation forest has self-learning ability, thereby stably improving the judgment accuracy in the absence of manual labeling. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a flow chart of a method for detecting the production quality of insulated cables according to an embodiment of the present application.

[0024] Figure 2 This is a structural block diagram of an insulated cable production quality inspection system according to an embodiment of the present application. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the described embodiments are part of the embodiments of this application. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0026] According to the first aspect of the present application, the present application provides a method for detecting the production quality of an insulated cable, which is used to monitor the production parameters of each production equipment during the production process of the insulated cable, and then judge the production quality of the insulated cable. Figure 1 This is a flow chart of a method for detecting the production quality of an insulated cable according to an embodiment of the present application. Figure 1As shown, the insulated cable production quality inspection method includes steps S101 to S103, which are described in detail below.

[0027] S101, constructing multiple real-time parameter features based on production parameter sequences of multiple production equipment.

[0028] In one embodiment, the plurality of production equipment includes a stranding machine, a cross-linking machine, a cabling machine, and the like; each production equipment corresponds to at least one production parameter. Exemplarily, the production parameters of the cross-linking machine include nitrogen pressure, wire diameter along the X axis, wire diameter along the Y axis, and average insulation thickness; and the production parameters of the stranding machine include line speed.

[0029] During the insulated cable production process, production parameter sequences from various production equipment can be collected, and real-time parameter features at any given moment can be constructed based on these production parameter sequences. Specifically, constructing multiple real-time parameter features involves preprocessing the production parameter sequences and then using the mean and standard deviation of the production parameter sequences within the current time window as feature values ​​for the production parameters; the feature values ​​of each production parameter correspond to the multiple real-time parameter features.

[0030] Wherein, the preprocessing includes standardization processing; the standardization processing is used to eliminate the dimension of each production parameter.

[0031] Among them, the time window of the current moment is the current moment and multiple historical moments before the current moment; in this embodiment of the application, the number of historical moments in the time window of the current moment is 10.

[0032] In this way, multiple real-time parameter features at the current moment are obtained. The multiple real-time parameter features can characterize the working status of each production equipment at the current moment. The working status of each production equipment directly affects the production quality of the insulated cable, providing a data basis for subsequent test results.

[0033] S102, multiple real-time features are input into isolated trees in the isolation forest, and the output outliers are weighted according to the normalized preset weights of each isolated tree to obtain a comprehensive outlier value.

[0034] In one embodiment, the isolation forest algorithm is an existing anomaly detection algorithm. In the existing isolation forest algorithm, anomaly detection is achieved with the help of a preset number of isolated trees in the isolation forest. Each isolated tree can independently obtain an anomaly value, and the average value of the anomaly values ​​obtained by each isolated tree is used as the final anomaly detection result.

[0035] It should be noted that the process of constructing an isolation forest based on the parameter characteristics of any historical moment in the historical production process of the insulated cable and the quality label of the insulated cable at the historical moment is a well-known technology in the isolation forest algorithm and will not be repeated here; the quality label is normal or abnormal.

[0036] Among them, the preset number is artificially set. If the number of isolated trees in the isolation forest is too large, the calculation amount of the comprehensive outlier value will be too large, affecting the real-time performance of the quality detection; if the number of isolated trees in the isolation forest is too small, the isolated trees cannot accurately obtain the comprehensive outlier value, affecting the accuracy of the quality detection. Therefore, in the embodiment of the present application, the preset number is set to 100.

[0037] Since each isolated tree in the isolation forest splits the parameter features by randomly selecting multiple parameter features and randomly selecting a value between the maximum and minimum values ​​of the selected parameter features as the split point; these random operations will inevitably lead to different abilities of each isolated tree to identify anomalies. Therefore, the present application improves the isolation forest algorithm and sets a preset weight for each isolated tree according to its ability to identify anomalies, so as to accurately judge whether the production quality of the insulated cable is abnormal.

[0038] In one embodiment, a method for obtaining the preset weight of each isolated tree includes: taking the labeled historical samples as the trust sample set; obtaining the outlier value of any historical sample in the trust sample set in each isolated tree, comparing the outlier value with the outlier threshold, obtaining the detection result of each isolated tree on the historical sample, and calculating the initial weight of each isolated tree based on the inter-class distance of the trust sample set; using the initial weight to filter the trust sample in the unlabeled historical samples, and recalculating the initial weight in the updated trust sample set; iteratively updating the initial weight until there are no unlabeled historical samples or the number of iterations is greater than the preset number or no trust sample can be filtered, and the preset weight of each isolated tree is obtained.

[0039] The abnormal threshold value is 0.5.

[0040] Labeled historical samples are historical samples with quality labels. These historical samples consist of multiple parameter features collected at any historical moment. Since quality labels require manual annotation, the number of available labeled historical samples is limited. First, the labeled historical samples are used as the trust sample set, and the initial weights of each isolated tree are calculated using this limited set of labeled historical samples.

[0041] Specifically, the isolation tree The initial weight for: ; and Isolation Tree The average outlier value for confidence samples with quality label normal and quality label anomaly.

[0042] Understandably, all historical samples in the trust sample set are trust samples. is the difference between the average outlier values ​​of abnormal trust samples and normal trust samples, that is, the inter-class distance between abnormal trust samples and normal trust samples; if , which means that the average outlier value of the trust sample with normal quality label is greater than the average outlier value of the trust sample with abnormal quality label, which shows that the isolated tree There was an error in the quality inspection of the trust sample, and the isolated tree It is impossible to accurately distinguish whether the quality of the trust sample is abnormal; therefore, the isolated tree The inter-class distance of the trust sample set is set to 0, that is, the isolated tree The initial weight Set to 0, not to participate in the quality inspection of insulated cables; at the same time, if , representing an isolated tree It can distinguish whether the quality of the trust sample is abnormal. The larger the value of the inter-class distance, the more accurate the distinction. It should be an isolated tree. Set a larger initial weight.

[0043] In this way, the initial weight of each isolated tree is calculated based on the labeled historical samples.

[0044] In one embodiment, using initial weights to screen trust samples from unlabeled historical samples includes: obtaining the outlier value of each isolated tree for the unlabeled historical sample, if the outlier value is greater than the outlier threshold, the detection result of the corresponding isolated tree is abnormal, otherwise, the detection result of the corresponding isolated tree is normal; taking the isolated trees with normal detection results as the first set, and taking the isolated trees with abnormal detection results as the second set; calculating the weighted average of the outliers in the first set and the second set respectively based on the normalized initial weights; in response to the absolute value of the difference between the weighted average of the outliers between the first set and the second set being greater than a preset value, marking the unlabeled historical sample as a trust sample.

[0045] The weighted average of the outliers in the first set is calculated by taking the ratio of the initial weight of any isolated tree in the first set to the sum of the initial weights of all isolated trees in the first set as the normalized weight of the isolated tree, and weighting the outliers in the first set according to the normalized weight to obtain the weighted average of the outliers in the first set. The default value is 0.4.

[0046] It can be understood that the closer the absolute value of the difference between the weighted average values ​​of the outliers between the first set and the second set is to 0, the closer the average outlier value of the unlabeled historical sample for the isolation tree with a normal detection result is to the average outlier value of the unlabeled historical sample for the isolation tree with an abnormal detection result, indicating that the isolation forest cannot effectively distinguish whether the quality of the unlabeled historical sample is abnormal, that is, the isolation forest cannot accurately obtain the detection results of the unlabeled historical sample, and the unlabeled historical sample is an untrusted sample.

[0047] In this way, new trust samples are obtained, the trust sample set is updated, and the updated trust sample set is used to update the initial weights. At this time, the trust sample set includes labeled historical samples with quality labels and unlabeled historical samples without quality labels. In the process of updating the initial weights, it is necessary to obtain the quality labels of the unlabeled historical samples. Therefore, before using the updated trust sample set to update the initial weights, the method for obtaining the preset weights of each isolated tree also includes: for the unlabeled historical samples in the trust sample set, in response to the sum of the initial weights of the isolated trees with normal detection results being greater than the sum of the initial weights of the isolated trees with abnormal detection results, the quality label of the unlabeled historical samples is normal; otherwise, the quality label of the unlabeled historical samples is abnormal.

[0048] At this point, after generating quality labels for the unlabeled historical samples in the trust sample set, the initial weights are recalculated in the updated trust sample set to update the initial weights. This iterative update allows the isolation forest to accurately distinguish more historical samples, continuously improving the accuracy of the isolation forest. Iterations are terminated until there are no more unlabeled historical samples, the number of iterations exceeds the preset number, or no trust samples are found. The preset number of iterations is set to 100, resulting in the preset weights for each isolation tree.

[0049] In another embodiment, after obtaining the preset weights for each isolated tree, the number of isolated trees in the isolation forest is increased in response to the number of unlabeled historical samples exceeding a minimum allowable value; the minimum allowable value is a preset ratio of the total number of historical samples, and the preset ratio is 0.05. Furthermore, to prevent an excessive number of isolated trees in the isolation forest, an upper limit on the number of isolated trees is set, and the upper limit is set to 300.

[0050] It is understandable that after obtaining the preset weights of each isolated tree, if there are still a large number of unlabeled historical samples, it means that based on the preset weights of each isolated tree, there are still a large number of historical samples that cannot accurately judge the production quality. At this time, it is necessary to increase the number of isolated trees in the isolation forest algorithm, construct multiple isolated trees again, merge the new isolated trees and the original isolated trees into a larger isolated forest, obtain the preset weights of all isolated trees again, and improve the robustness of the isolation forest algorithm to ensure accurate quality detection results.

[0051] It should be noted that increasing the number of isolated trees in an isolation forest here means constructing new isolated trees. These new isolated trees and the original isolated trees form the isolation forest, and the preset weights of each isolated tree in each isolation forest are recalculated. For example, if the number of isolated trees added to the isolation forest is 20, the new isolated trees and the original 100 isolated trees form the isolation forest, which now contains 120 isolated trees.

[0052] In one embodiment, after determining the preset weight of each isolation tree based on historical samples, multiple real-time features are input into the isolation trees in the isolation forest, and the output outliers are weighted according to the normalized preset weights of each isolation tree to obtain a comprehensive outlier.

[0053] S103, comparing the comprehensive abnormal value and the abnormal threshold to obtain a detection result.

[0054] In one embodiment, the combined abnormality value is compared with the abnormality threshold. If the combined abnormality value is greater than the abnormality threshold, the detection result is abnormal; otherwise, the detection result is normal. The abnormality threshold is set to 0.5; implementers can also adjust the value of the abnormality threshold based on actual scenarios, and this application does not impose any restrictions.

[0055] In this way, the real-time detection of the production quality of the insulated cable is achieved during the production process of the insulated cable.

[0056] According to the second aspect of the present application, the present application also provides an insulated cable production quality detection system. Figure 2 This is a structural block diagram of an insulated cable production quality inspection system according to an embodiment of the present application. Figure 2 As shown, the system 50 includes a processor and a memory. The memory stores computer program instructions. When the computer program instructions are executed by the processor, the method for inspecting the production quality of insulated cables according to the first aspect of the present application is implemented. The system also includes other components well known to those skilled in the art, such as a communication bus and a communication interface. The configuration and functions of these components are well known in the art and are therefore not described in detail here.

[0057] It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present application, and these all fall within the scope of protection of the present application.

Claims

1. A method for detecting the production quality of an insulated cable, characterized in that: The detection method comprises: Construct multiple real-time parameter features based on the production parameter sequences of various production equipment; Multiple real-time features are input into the isolated trees in the isolation forest, and the output outliers are weighted according to the preset weights of the normalized isolated trees to obtain the comprehensive outlier value; Compare the comprehensive outlier value and the outlier threshold to obtain the detection result; The method for obtaining the preset weight of each isolated tree includes: taking the labeled historical samples as the trust sample set; obtaining the outlier value of any historical sample in the trust sample set in each isolated tree, comparing the outlier value and the outlier threshold, obtaining the detection result of each isolated tree on the historical sample, and calculating the initial weight of each isolated tree based on the inter-class distance of the trust sample set; using the initial weight to screen the trust samples in the unlabeled historical samples, and recalculating the initial weight in the updated trust sample set; iteratively updating the initial weight until there are no unlabeled historical samples or the number of iterations is greater than the preset number or no trust samples can be screened, and the preset weight of each isolated tree is obtained.

2. A method for detecting the production quality of an insulated cable according to claim 1, characterized in that: Constructing multiple real-time parameter features includes: After preprocessing the production parameter sequence, the mean and standard deviation of the production parameter sequence in the current time window are used as the characteristic values ​​of the production parameter; The characteristic value of each production parameter corresponds to the plurality of real-time parameter characteristics.

3. A method for detecting the production quality of an insulated cable according to claim 2, characterized in that: The preprocessing includes a standardization process.

4. A method for detecting the production quality of an insulated cable according to claim 2, characterized in that: The time window of the current moment is the current moment and multiple historical moments before the current moment.

5. The method for detecting the production quality of an insulated cable according to claim 1, wherein: Isolated tree The initial weight for: ; and Isolation Tree The average outlier value for confidence samples with quality label normal and quality label anomaly.

6. The method for detecting the production quality of an insulated cable according to claim 1, characterized in that: Using initial weights to screen trust samples from unlabeled historical samples includes: obtaining the outlier value of each isolated tree for the unlabeled historical samples, if the outlier value is greater than the outlier threshold, the detection result of the corresponding isolated tree is abnormal, otherwise, the detection result of the corresponding isolated tree is normal; taking the isolated trees with normal detection results as the first set, and taking the isolated trees with abnormal detection results as the second set; calculating the weighted average of the outliers in the first set and the second set respectively based on the normalized initial weights; in response to the absolute value of the difference between the weighted average of the outliers between the first set and the second set being greater than a preset value, marking the unlabeled historical samples as trust samples.

7. A method for detecting the production quality of an insulated cable according to claim 6, characterized in that: The weighted average of the outliers in the first set is: The ratio of the initial weight of any isolated tree in the first set to the sum of the initial weights of all isolated trees in the first set is used as the normalized weight of the isolated tree, and the outliers in the first set are weighted according to the normalized weight to obtain the weighted average of all outliers in the first set.

8. The method for detecting the production quality of an insulated cable according to claim 1, wherein: Before updating the initial weights using the updated trust sample set, the method for obtaining the preset weights of each isolated tree further includes: For unlabeled historical samples in the trust sample set, if the sum of the initial weights of the isolated trees whose detection results are normal is greater than the sum of the initial weights of the isolated trees whose detection results are abnormal, the quality label of the unlabeled historical samples is normal; otherwise, the quality label of the unlabeled historical samples is abnormal.

9. The method for detecting the production quality of an insulated cable according to claim 1, wherein: After obtaining the preset weight of each isolated tree, the method for obtaining the preset weight of each isolated tree further includes: In response to the number of unlabeled historical samples being greater than a minimum allowed value, the number of isolated trees in the isolation forest is increased; the minimum allowed value is a preset proportion of the total number of historical samples.

10. An insulated cable production quality inspection system, characterized in that: The method comprises a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, a method for detecting the production quality of an insulated cable according to any one of claims 1 to 9 is implemented.

Images