Shoe material qualification rate prediction system based on machine learning

Abstract

The invention relates to the technical field of shoe material qualification rate prediction, in particular to a shoe material qualification rate prediction system based on machine learning. The system comprises an influence trend simulation module and a qualified rate prediction module. According to the invention, the influence trend simulation module updates the monitoring frequency and the sample size of the monitoring equipment according to the influence trend, adaptively updates the sample size through the defect rate of small-batch samples obtained in real time, feeds back the defect severity of the current batch of shoe materials, and improves the detection accuracy. And the qualified rate prediction module is combined with the updated monitoring frequency and sample size of the monitoring equipment to predict the qualified rate of the rubber surface of the current batch of shoe materials, so that the prediction accuracy of the qualified rate of the rubber surface of the shoe materials is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of predicting the qualification rate of shoe materials, and more specifically, to a system for predicting the qualification rate of shoe materials based on machine learning. Background Art

[0002] During the production process of shoe materials, defects in the gluing process directly affect the product performance and appearance. According to industry data statistics, gluing-related defects account for about 30% of the total quality problems of shoes. Therefore, predicting the qualification rate of the glued surface of shoe materials is the key to evaluating whether the gluing process is abnormal. The traditional prediction process for the qualification rate of the glued surface of shoe materials mainly includes the following steps:

[0003] First step: Install monitoring equipment on the assembly line of the shoe material gluing process to obtain the state data of the glued surface of the shoe materials;

[0004] Second step: Compare with the qualified glued surface database to distinguish abnormal glued surfaces from normal glued surfaces;

[0005] Third step: Adopt the sample spot-check method to obtain the abnormality rate of real-time samples, that is, plan the sample quantity, formulate the spot-check order, obtain the proportion of abnormal glued surfaces in the samples, and finally predict the qualification rate of the glued surface of the current batch of shoe materials.

[0006] Since most of the defects in the glued surface of shoe materials in the gluing process are mechanical failures, for example, glue overflow, glue collapse, and glue indentation are all caused by abnormalities in the glue spraying area. And once an abnormality occurs and is not dealt with in time, corresponding defect situations will appear in clusters, that is, adjacent glued surfaces of shoe materials will all become abnormal. Therefore, the above sample spot-check method cannot reflect the actual defect proportion, resulting in an increase in the error of the predicted result of the glued surface qualification rate in the later stage.

[0007] In order to address the above problems, there is an urgent need for a system for predicting the qualification rate of shoe materials that can update the sampling frequency in real time according to the influence trend. Summary of the Invention

[0008] The purpose of the present invention is to provide a system for predicting the qualification rate of shoe materials based on machine learning to solve the problems raised in the above background art.

[0009] To achieve the above purpose, a system for predicting the qualification rate of shoe materials based on machine learning is provided, including a shoe material monitoring data acquisition module, an abnormal area marking module, a historical defect data feedback module, a defect comparison and identification module, an influence trend simulation module, and a qualification rate prediction module;

[0010] Among them, the shoe material monitoring data acquisition module is used to collect the monitoring data of the glued surface of shoe materials by combining the feedback information of the monitoring equipment on the assembly line;

[0011] The abnormal area marking module combines with the standard glue surface data set to mark abnormal shoe materials and abnormal glue surface areas;

[0012] The historical defect data feedback module combines with historical detection data to obtain a defective glue surface data set;

[0013] The defect comparison and identification module compares the abnormal glue surface area and the defective glue surface data set to predict the defective glue surface and the defect type;

[0014] The influence trend simulation module obtains the monitoring frequency, sample size, and the total amount of the current batch of shoe materials of the monitoring device, predicts the influence trend of shoe material defects, and updates the monitoring frequency and sample size of the monitoring device according to the influence trend;

[0015] The pass rate prediction module combines the updated monitoring frequency and sample size of the monitoring device to predict the pass rate of the glue surface of the current batch of shoe materials.

[0016] As a further improvement of this technical solution, the monitoring devices in the shoe material monitoring data acquisition module include camera monitoring devices, infrared imaging devices, and laser profile scanning devices;

[0017] Among them, the camera monitoring device is used to collect image data of the glue surface area;

[0018] The infrared imaging device is used to collect temperature change data of the glue surface area;

[0019] The laser profile scanning device is used to obtain the inner and outer contours of the glue surface area.

[0020] As a further improvement of this technical solution, the acquisition method of the standard glue surface data set in the abnormal area marking module includes the following steps:

[0021] S201. Collect historical monitoring data information to obtain various monitoring values of normal shoe materials;

[0022] S202. Divide the corresponding normal value ranges according to different types of monitoring data and summarize them into a standard glue surface data set.

[0023] As a further improvement of this technical solution, the marking of abnormal shoe materials and abnormal glue surface areas in the abnormal area marking module includes the following methods:

[0024] The abnormal shoe materials are marked by the serial numbers of the abnormal shoe materials;

[0025] The abnormal glue surface areas are marked by the glue surface positions, abnormal areas, and abnormal quantities.

[0026] As a further improvement of this technical solution, the method for obtaining the defective glue surface data set in the historical defect data feedback module includes the following steps:

[0027] S301. Collect historical monitoring data and obtain the monitoring values under various defect states;

[0028] S302. Define the defect scope under various defect states;

[0029] S303. Statistically analyze the defect scope under various defect states and generate a defect glue surface dataset.

[0030] As a further improvement of this technical solution, the method for defining the defect scope under various defect states in S302 includes the following steps:

[0031] S3021. Obtain the numerical distribution under various defect states, select the minimum value and the maximum value to form the defect scope;

[0032] S3022. Combine the later detection results, obtain the defect degree under different numerical states in the defect scope, and re-divide the defect intervals in the defect scope according to the defect degree;

[0033] S3023. Assign weights to the defect intervals according to the defect degree.

[0034] As a further improvement of this technical solution, the method for predicting the influence trend of shoe material defects in the influence trend simulation module includes the following steps:

[0035] S501. Calculate the real-time probability of each influencing defect according to the initial sampling data;

[0036] S502. Define different sampling stages, and different sampling stages correspond to probability thresholds and updated sampling frequencies;

[0037] S503. Compare the real-time probability with the probability threshold to determine whether to enter the next sampling stage;

[0038] When the real-time probability < the probability threshold, maintain the sampling frequency of the current stage;

[0039] When the real-time probability ≥ the probability threshold, automatically proceed to the next sampling stage and update the sampling frequency.

[0040] As a further improvement of this technical solution, the method for calculating the real-time probability of each influencing defect in S501 includes the following steps:

[0041] S5011. Define the adjacent unit selection quantity according to the serial number of the selected shoe material;

[0042] S5012. Calculate the probability of the adjacent unit selection quantity for each position and update the real-time probability according to the position ranking.

[0043] Beneficial effects of the present invention compared with the prior art:

[0044] In the shoe material qualification rate prediction system based on machine learning, the influence trend simulation module updates the monitoring frequency and sample size of the monitoring device according to the influence trend, adaptively updates the sampling sample size through the defect rate of the small batch samples obtained in real time, and feeds back the severity of the defects of the current batch of shoe materials. The qualification rate prediction module combines the updated monitoring frequency and sample size of the monitoring device to predict the qualification rate of the glue surface of the current batch of shoe materials, improving the prediction accuracy of the qualification rate of the glue surface of shoe materials. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 It is a block diagram of the overall structure system of the present invention;

[0046] Figure 2 It is a flowchart of the acquisition method of the standard glue surface data set of the present invention;

[0047] Figure 3 It is a flowchart of the method for obtaining the defective glue surface data set of the present invention;

[0048] Figure 4 It is a flowchart of the method for formulating the defect range under various defect states of the present invention;

[0049] Figure 5 It is a flowchart of the method for predicting the influence trend of shoe material defects of the present invention;

[0050] Figure 6 It is a flowchart of the method for calculating the real-time probability of each influencing defect of the present invention.

[0051] The meanings of the various labels in the figure are as follows:

[0052] 10. Shoe material monitoring data acquisition module;

[0053] 20. Abnormal area marking module;

[0054] 30. Historical defect data feedback module;

[0055] 40. Defect comparison and identification module;

[0056] 50. Influence trend simulation module;

[0057] 60. Qualification rate prediction module. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0058] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings in the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without creative efforts belong to the scope of protection of the present invention.

[0059] Please refer to Figure 1 as shown, a shoe material qualification rate prediction system based on machine learning is provided, including a shoe material monitoring data acquisition module 10, an abnormal area marking module 20, a historical defect data feedback module 30, a defect comparison and identification module 40, an influence trend simulation module 50, and a qualification rate prediction module 60;

[0060] Among them, the shoe material monitoring data acquisition module 10 is used to collect shoe material glue surface monitoring data in combination with the feedback information of the monitoring equipment on the production line;

[0061] The abnormal area marking module 20 marks abnormal shoe materials and abnormal glue surface areas in combination with the standard glue surface data set;

[0062] The historical defect data feedback module 30 obtains the defective glue surface data set in combination with the historical detection data;

[0063] The defect comparison and identification module 40 compares the abnormal glue surface area and the defective glue surface data set to predict the defective glue surface and the defect type;

[0064] The influence trend simulation module 50 obtains the monitoring frequency, sample size, and the total amount of the current batch of shoe materials of the monitoring equipment, predicts the influence trend of shoe material defects, and updates the monitoring frequency and sample size of the monitoring equipment according to the influence trend;

[0065] The qualification rate prediction module 60 predicts the qualification rate of the shoe material glue surface of the current batch in combination with the updated monitoring frequency and sample size of the monitoring equipment.

[0066] In order to address the above problems, this solution provides a shoe material qualification rate prediction system that can update the sampling frequency in real time according to the influence trend, and its specific content is as follows:

[0067] First, in order to obtain the initial sampling data, the shoe material monitoring data acquisition module 10 collects the shoe material glue surface monitoring data in combination with the feedback information of the monitoring equipment on the production line, that is, adopts the initial sampling method, formulates the sample size and the sampling interval, and indirectly obtains the shoe material glue surface monitoring data. This sampling method is mainly applied to the defect monitoring caused by non-mechanical failures;

[0068] In order to distinguish the normal rubber surface area from the defective rubber surface area, it is necessary to use the abnormal area marking module 20 to combine with the standard rubber surface data set to mark the abnormal shoe materials and the abnormal rubber surface area. At this time, the corresponding abnormal rubber surface area is only different from the standard rubber surface data set, and does not represent the defective rubber surface area. Further comparison is required. The defective comparison and identification module 40 compares the abnormal rubber surface area and the defective rubber surface data set to predict the defective rubber surface and the defect type;

[0069] Since most mechanical defects occur in batches, it is difficult to predict by the initial sampling method. It is necessary to use the impact trend simulation module 50 to obtain the monitoring frequency, sample size and the total amount of the current batch of shoe materials of the monitoring equipment, predict the impact trend of shoe material defects, and update the monitoring frequency and sample size of the monitoring equipment according to the impact trend, that is, obtain the defect rate of a small batch in real time. Through the analysis of the defect rate and the defect type, predict the impact trend of the current defect, such as the severity of the defect. Finally, the pass rate prediction module 60 combines the updated monitoring frequency and sample size of the monitoring equipment to predict the pass rate of the rubber surface of the current batch of shoe materials.

[0070] In the present invention, the impact trend simulation module 50 updates the monitoring frequency and sample size of the monitoring equipment according to the impact trend, adapts and updates the sampling sample size through the defect rate of the small batch samples obtained in real time, feeds back the severity of the defects of the current batch of shoe materials, and the pass rate prediction module 60 combines the updated monitoring frequency and sample size of the monitoring equipment to predict the pass rate of the rubber surface of the current batch of shoe materials, improving the prediction accuracy of the pass rate of the rubber surface of shoe materials.

[0071] In addition, the monitoring equipment in the shoe material monitoring data acquisition module 10 includes a camera monitoring device, an infrared imaging device and a laser profile scanning device;

[0072] Among them, the camera monitoring device is used to collect image data of the rubber surface area;

[0073] The infrared imaging device is used to collect temperature change data of the rubber surface area;

[0074] The laser profile scanning device is used to obtain the inner and outer contours of the rubber surface area. For example, when there is a hollow inside the rubber surface, the laser profile scanning is used for hollow positioning.

[0075] Furthermore, as Figure 2 shown, the acquisition method of the standard rubber surface data set in the abnormal area marking module 20 includes the following steps:

[0076] S201. Collect historical monitoring data information and obtain various monitoring values of normal shoe materials;

[0077] S202. Divide the corresponding normal value ranges according to different types of monitoring data and summarize them into a standard rubber surface data set.

[0078] During specific use, in order to divide the abnormal glue surface area in the later stage, by collecting historical monitoring data information, various monitoring values of normal shoe materials are obtained. For example, the glue surface overflow amount, the depth of the depression outside the glue surface, and the glue surface notch amount reflected in the image data can all be fed back through the glue surface image data. At the same time, according to different types of monitoring data, the corresponding normal value ranges are divided, that is, the value ranges within which the shoe materials meet the standards after detection. For example, in the glue surface temperature change data, the feedback that exceeds the normal temperature by more than 3°C is marked as the abnormal range. These corresponding normal value ranges will be summarized into a standard glue surface data set as the basis for evaluating the abnormal glue surface area.

[0079] Furthermore, in the abnormal area marking module 20, marking abnormal shoe materials and abnormal glue surface areas includes the following methods:

[0080] The abnormal shoe materials are marked by the serial numbers of the abnormal shoe materials;

[0081] The abnormal glue surface areas are marked by the glue surface positions, abnormal areas, and abnormal quantities. Among them, the abnormal area is the area occupied by the current abnormal area, and the abnormal quantity is the total number of abnormalities that appear at different positions on the same glue surface.

[0082] Specifically, as Figure 3 shown, the method for obtaining the defective glue surface data set in the historical defect data feedback module 30 includes the following steps:

[0083] S301. Collect historical monitoring data and obtain the monitoring values under various defective states;

[0084] S302. Define the defect ranges under various defective states;

[0085] S303. Statistically analyze the defect ranges under various defective states and generate a defective glue surface data set.

[0086] In addition, as Figure 4 shown, the method for defining the defect ranges under various defective states in S302 includes the following steps:

[0087] S3021. Obtain the numerical distributions under various defective states, select the minimum value and the maximum value to form the defect range;

[0088] S3022. Combine the later detection results, obtain the defect degrees under different numerical states in the defect range, and re-divide the defect intervals in the defect range according to the defect degrees;

[0089] S3023. Assign weights to the defect intervals according to the defect degrees.

[0090] During specific use, since the specific manifestations of different defects are different, for later comparison, historical monitoring data is collected to obtain the monitoring values under various defect states, and the defect ranges under various defect states are formulated. In the actual collection process, the monitoring values of the defect states are not directly proportional to the degree of influence. Therefore, in order to accurately divide the defect ranges, by obtaining the numerical distributions under various defect states, the minimum value and the maximum value are selected to form the defect ranges. During this process, combined with the later detection results, the defect degrees under different numerical states in the defect ranges are obtained, and the defect intervals in the defect ranges are re-divided according to the defect degrees, that is, according to the numerical references in the detection results, such as tensile test data and hardness test data, and the changes of these values are directly proportional to the detection results, which are used as the evaluation of the defect degrees. Finally, weight assignments are given to the defect intervals according to the defect degrees, that is, the higher the defect degree of the defect interval, the higher the corresponding weight assignment.

[0091] Further, as Figure 5 shown, the method for predicting the influence trend of shoe material defects in the influence trend simulation module 50 includes the following steps:

[0092] S501. Calculate the real-time probability of each influencing defect according to the initial spot-check data;

[0093] S502. Define different spot-check stages, and each spot-check stage corresponds to a probability threshold and an updated spot-check frequency;

[0094] S503. Compare the real-time probability with the probability threshold to determine whether to enter the next spot-check stage;

[0095] When the real-time probability < the probability threshold, maintain the spot-check frequency of the current stage;

[0096] When the real-time probability ≥ the probability threshold, automatically proceed to the next spot-check stage and update the spot-check frequency.

[0097] Still further, as Figure 6 shown, the method for calculating the real-time probability of each influencing defect in S501 includes the following steps:

[0098] S5011. Define the adjacent unit selection quantity according to the serial number of the selected shoe material;

[0099] S5012. Calculate the probability of the adjacent unit selection quantity of each order and update the real-time probability according to the order ranking.

[0100] In actual use, during the process of predicting the influence trend of shoe material defects, since the defect situations of shoe materials generally occur in piles, the conventional sampling method is not applicable. This solution calculates the real-time probability of each influencing defect according to the initial sampling data. For example, during the initial sampling, every 10 shoe materials are sampled once, and the labels of each shoe material are 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, , if the first one is the sampled shoe material, the corresponding labels 11, 21, 31, 41, 51, 61, 71, 81, and 91 are all the sampled shoe materials under the initial sampling method. To improve the accuracy of the real-time probability, according to the serial numbers of the selected shoe materials, the adjacent unit selection quantity is formulated. For example, the adjacent unit selection quantity is marked as 4 times, and the corresponding 11, 21, 31, and 41 are a group of adjacent unit selection quantities, and the probability of each adjacent unit selection quantity in each position is calculated, and the real-time probability is updated according to the position sorting, that is, updated according to the serial number order. Finally, different sampling stages are formulated, and different sampling stages correspond to probability thresholds and updated sampling frequencies. The real-time probability is compared with the probability threshold to determine whether to enter the next sampling stage;

[0101] When the real-time probability < the probability threshold, the sampling frequency of the current stage is maintained;

[0102] When the real-time probability ≥ the probability threshold, it automatically proceeds to the next sampling stage and updates the sampling frequency;

[0103] For example, when the adjacent unit selection quantities are 11, 21, 31, and 41, the corresponding defect probability is 50%. When the adjacent unit selection quantities are 21, 31, 41, and 51, the corresponding defect probability is 75%. And the current sampling stage is in the initial stage, and the probability threshold of this stage is 60%. When the adjacent unit selection quantities are 21, 31, 41, and 51, the corresponding defect probability exceeds the probability threshold of this stage. At this time, it is necessary to skip the initial stage and enter the next sampling stage. The corresponding sampling frequency is to sample once every 5 shoe materials. The initial adjacent unit selection quantities are 52, 57, 62, and 67. At this time, the defect probability in this sampling stage is calculated and compared with the probability threshold in the current sampling stage to determine whether to enter the next stage.

[0104] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and the descriptions in the specification are only preferred examples of the present invention and are not used to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. The shoe material qualification rate prediction system based on machine learning is characterized by: It comprises a shoe material monitoring data collection module (10), an abnormal area marking module (20), a historical defect data feedback module (30), a defect comparison and identification module (40), an impact trend simulation module (50) and a qualified rate prediction module (60); The shoe material monitoring data collection module (10) is used to collect shoe material rubber surface monitoring data in combination with feedback information from monitoring equipment on the production line; The abnormal area marking module (20) combines the standard rubber surface data set to mark abnormal shoe materials and abnormal rubber surface areas; The historical defect data feedback module (30) combines the historical inspection data to obtain a defective adhesive surface data set; The defect comparison and identification module (40) compares the abnormal adhesive surface area and the defective adhesive surface data set to predict the defective adhesive surface and the defect type; The impact trend simulation module (50) obtains the monitoring frequency, sample size and total amount of the current batch of shoe materials of the monitoring equipment, predicts the impact trend of shoe material defects, and updates the monitoring frequency and sample size of the monitoring equipment according to the impact trend; The qualified rate prediction module (60) predicts the qualified rate of the rubber surface of the shoe material of the current batch in combination with the monitoring frequency and sample size of the updated monitoring equipment.

2. The shoe material qualification rate prediction system based on machine learning according to claim 1 is characterized in that: The monitoring equipment in the shoe material monitoring data acquisition module (10) includes a camera monitoring device, an infrared imaging device and a laser profile scanning device; Among them, the camera monitoring equipment is used to collect image data of the rubber surface area; Infrared imaging equipment is used to collect temperature change data of the adhesive surface area; Laser profile scanning equipment is used to obtain the inner and outer contours of the adhesive surface area.

3. The shoe material qualification rate prediction system based on machine learning according to claim 1 is characterized in that: The method for collecting the standard adhesive surface data set in the abnormal area marking module (20) comprises the following steps: S201, collecting historical monitoring data information to obtain various monitoring values ​​of normal shoe materials; S202. According to different types of monitoring data, the corresponding normal value ranges are divided and summarized into a standard rubber surface data set.

4. The shoe material qualification rate prediction system based on machine learning according to claim 3 is characterized in that: The abnormal area marking module (20) marks abnormal shoe material and abnormal rubber surface areas in the following manners: Marking abnormal shoe materials is indicated by the serial number of the abnormal shoe materials; Mark the abnormal adhesive surface area by the adhesive surface location, abnormal area and abnormal number.

5. The shoe material qualification rate prediction system based on machine learning according to claim 1 is characterized in that: The method for obtaining the defective adhesive surface data set in the historical defect data feedback module (30) comprises the following steps: S301, collect historical monitoring data and obtain monitoring values ​​under various defect states; S302. Formulate the defect scope under each defect state; S303, counting the defect ranges under various defect states, and generating a defect adhesive surface data set.

6. The shoe material qualification rate prediction system based on machine learning according to claim 1 is characterized in that: The method for formulating the defect range under each defect state in S302 includes the following steps: S3021. Obtain the value distribution under each defect state, select the minimum value and the maximum value to form a defect range; S3022. In combination with the later detection results, the defect degrees in different numerical states in the defect range are obtained, and the defect intervals in the defect range are divided twice according to the defect degrees; S3023. Assign weights to defect intervals according to the degree of defects.

7. The shoe material qualification rate prediction system based on machine learning according to claim 1 is characterized in that: The method for predicting the impact trend of shoe material defects in the impact trend simulation module (50) comprises the following steps: S501. Calculate the real-time probability of each influencing defect according to the initial spot check data; S502. Formulate different sampling stages, and each sampling stage has a corresponding probability threshold and updates the sampling frequency; S503, comparing the real-time probability with the probability threshold, and determining whether to enter the next spot check stage; When the real-time probability is less than the probability threshold, the sampling frequency at the current stage is maintained; When the real-time probability ≥ the probability threshold, it will automatically proceed to the next sampling stage and update the sampling frequency.

8. The shoe material qualification rate prediction system based on machine learning according to claim 7 is characterized in that: The method for calculating the real-time probability of each influencing defect in S501 comprises the following steps: S5011. According to the serial number of the shoe materials to be selected, determine the adjacent unit frame selection quantity; S5012. Calculate the probability of the adjacent unit selection amount in each order, and update the probability in real time according to the order.

Images