Aluminum foil lunch box defect detection method and system

Abstract

The invention discloses an aluminum foil lunch box defect detection method and system, and relates to the technical field of defect detection. The method comprises: activating an aluminum foil raw material table of an aluminum foil lunch box, extracting a raw material type set and a raw material ratio set; receiving an aluminum foil mother roll preparation process, combining the raw material type set and the raw material ratio set, calibrating the mother roll content of a first heavy metal of the raw material type set, and generating a predicted content of the first heavy metal; extracting a first heavy metal standard content threshold from an aluminum foil lunch box metal raw material standard table; performing a trigger threshold evaluation according to a first heavy metal historical detection log, and calibrating a detection trigger content threshold; performing physical defect detection of the first heavy metal on the aluminum foil lunch box, thereby solving the problem in the prior art that the defect detection work is not rigorous enough and lacks completeness, resulting in low detection efficiency, and realizing reasonable and precise control of heavy metal detection on aluminum foil lunch boxes.

Description

Technical Field

[0001] The present invention relates to the technical field of defect detection, and in particular to a method and system for detecting defects in an aluminum foil lunch box. Background Art

[0002] Aluminum foil lunch boxes are widely used in catering, food packaging and other fields due to their lightness, durability and easy recycling. However, heavy metals in aluminum foil lunch boxes are directly related to human health, so the defect detection of heavy metal content is an important monitoring content, but the traditional detection method requires a full inspection of each heavy metal. This method not only leads to low detection efficiency of defective heavy metal content, but also easy to miss and misdetect. How to improve the detection efficiency of defective heavy metal content is of great significance to improving the production efficiency of aluminum foil lunch boxes and ensuring food safety.

[0003] The defect detection work existing in the existing technology is not rigorous enough and lacks completeness, resulting in low detection efficiency, which makes it impossible to rationally and accurately control the heavy metal detection of aluminum foil lunch boxes. Summary of the invention

[0004] The present application provides an aluminum foil lunch box defect detection method and system, which solves the problem of low detection efficiency due to lack of rigor and completeness in the defect detection work in the prior art, and realizes reasonable and precise control of heavy metal detection of aluminum foil lunch boxes.

[0005] In view of the above problems, the present application provides a method for detecting defects in aluminum foil lunch boxes.

[0006] In a first aspect, the present application provides a method for detecting defects in an aluminum foil lunch box, the method comprising: extracting a raw material type set and a raw material ratio set by activating an aluminum foil raw material table of an aluminum foil lunch box; receiving an aluminum foil mother roll preparation process, combining the raw material type set and the raw material ratio set, calibrating the mother roll content of a first heavy metal in the raw material type set, and generating a first heavy metal predicted content; extracting a first heavy metal standard content threshold from an aluminum foil lunch box metal raw material standard table; when the first heavy metal predicted content is greater than or equal to the first heavy metal standard content threshold, performing a physical defect detection of the first heavy metal on the aluminum foil lunch box.

[0007] In a second aspect, the present application provides an aluminum foil lunch box defect detection system, the system comprising:

[0008] Aluminum foil raw material table activation module: activates the aluminum foil raw material table of the aluminum foil lunch box, extracts the raw material type set and the raw material ratio set; mother roll content calibration module: receives the aluminum foil mother roll preparation process, combines the raw material type set and the raw material ratio set, calibrates the mother roll content of the first heavy metal in the raw material type set, and generates a first heavy metal predicted content; standard content threshold extraction module: extracts the first heavy metal standard content threshold from the aluminum foil lunch box metal raw material standard table; physical defect detection module: when the first heavy metal predicted content is greater than or equal to the first heavy metal standard content threshold, performs physical defect detection of the first heavy metal on the aluminum foil lunch box.

[0009] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0010] The aluminum foil lunch box defect detection method and system provided in the embodiment of the present application activates the aluminum foil raw material table of the aluminum foil lunch box, extracts the raw material type set and the raw material ratio set, then receives the aluminum foil mother roll preparation process, combines the raw material type set and the raw material ratio set, calibrates the mother roll content of the first heavy metal in the raw material type set, generates a first heavy metal predicted content, then extracts the first heavy metal standard content threshold from the aluminum foil lunch box metal raw material standard table, and finally, when the first heavy metal predicted content is greater than or equal to the first heavy metal standard content threshold, performs physical defect detection of the first heavy metal on the aluminum foil lunch box, thereby solving the problem of low detection efficiency due to insufficient rigor and completeness of defect detection work in the prior art, and realizes reasonable and precise control of heavy metal detection on aluminum foil lunch boxes. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 A schematic diagram of the process of detecting defects in aluminum foil lunch boxes is provided for this application;

[0012] Figure 2 A structural schematic diagram of an aluminum foil lunch box defect detection system is provided for this application.

[0013] Explanation of the reference numerals: aluminum foil raw material table activation module 11, mother roll content calibration module 12, standard content threshold extraction module 13, physical defect detection module 14. DETAILED DESCRIPTION

[0014] This application provides a defect detection method and system for aluminum foil lunch boxes. By activating the aluminum foil raw material table of the aluminum foil lunch box, extracting the raw material type set and the raw material ratio set, and then receiving the aluminum foil mother roll preparation process, combining the raw material type set and the raw material ratio set, the mother roll content of the first heavy metal in the raw material type set is calibrated to generate the first heavy metal predicted content, and then the first heavy metal standard content threshold is extracted from the aluminum foil lunch box metal raw material standard table. Finally, when the first heavy metal predicted content is greater than or equal to the first heavy metal standard content threshold, the aluminum foil lunch box is subjected to physical defect detection of the first heavy metal. The problem of low detection efficiency due to insufficient rigor and completeness of defect detection work in the prior art is solved, and the heavy metal detection of aluminum foil lunch boxes is rationally and accurately controlled.

[0015] Embodiment 1

[0016] like Figure 1 As shown, the present application provides a method and system for detecting defects in aluminum foil lunch boxes, the method comprising:

[0017] Activate the aluminum foil raw material table of the aluminum foil lunch box, and extract the raw material type set and raw material ratio set;

[0018] The aluminum foil raw material table of the aluminum foil lunch box refers to the raw material composition table for producing the aluminum foil lunch box. The raw materials for producing the aluminum foil lunch box include aluminum, iron, silicon, copper and other elements, which may vary according to different manufacturing processes and materials. Some aluminum foils may also contain materials such as plastics and fibers to form composite materials. Obtain through the aluminum foil raw material table of the aluminum foil lunch box, and obtain the material components that constitute the aluminum foil lunch box in the aluminum foil raw material table of the aluminum foil lunch box, and output the obtained material components that constitute the aluminum foil lunch box to obtain the raw material type set; then obtain the material additive components that constitute the aluminum foil lunch box through the aluminum foil raw material table of the aluminum foil lunch box, and output the material additive components that constitute the aluminum foil lunch box to obtain the raw material ratio set. By activating the aluminum foil raw material table of the aluminum foil lunch box, extracting the raw material type set and the raw material ratio set, the raw material type set and the raw material ratio set are combined for the subsequent reception of the aluminum foil mother roll preparation process, and the raw material type set and the raw material ratio set are combined to calibrate the mother roll content of the first heavy metal in the raw material type set, and generate the first heavy metal predicted content to provide a data basis.

[0019] receiving an aluminum foil mother roll preparation process, combining the raw material type set and the raw material ratio set, calibrating the mother roll content of a first heavy metal of the raw material type set, and generating a predicted first heavy metal content;

[0020] Due to the different preparation processes of aluminum foil mother rolls, the final heavy metal content of aluminum foil lunch boxes processed with the same raw material ratio will also be different, so the aluminum foil mother roll preparation process needs to be considered. The aluminum foil mother roll preparation process is obtained, and the raw material type set and raw material ratio set corresponding to the aluminum foil mother roll preparation process are combined to calibrate the heavy metal content of the mother roll for the first heavy metal in the raw material type set, that is, to calibrate the mother roll content of the first heavy metal, obtain the calibration result, and output the calibration result to obtain the predicted content of the first heavy metal. Among them, to calibrate the content of the first heavy metal in the mother roll, it is necessary to set the corresponding calibration function, and input the raw material type set and raw material ratio set corresponding to the aluminum foil mother roll preparation process into the calibration function according to the aluminum foil mother roll preparation process. The calibration function calculates according to the input value to obtain the corresponding heavy metal content value. The heavy metal content value obtained by the calibration function is the theoretical heavy metal content value. In the actual processing process, the heavy metal content of the aluminum foil lunch box will be affected by many factors, so the heavy metal content value obtained by the calibration function is the theoretical calculation value, that is, the predicted content. The calibration function needs to be further optimized and corrected in the future to make the obtained first heavy metal predicted content more accurate. The generation of the first heavy metal predicted content provides a data basis for the subsequent generation of the first residual function.

[0021] Extract the first heavy metal standard content threshold from the standard table of metal raw materials for aluminum foil lunch boxes;

[0022] The standard table of metal raw materials for aluminum foil lunch boxes is a list of standard values ​​of metal raw materials contained in aluminum foil lunch boxes, which specifies the metal content of aluminum foil lunch boxes and the heavy metal content contained in aluminum foil lunch boxes. In aluminum foil lunch boxes, according to the requirements of relevant food safety standards, a certain amount of heavy metals is allowed. For example, in the current regulations, the content of heavy metals shall not exceed 0.25 parts per million. The heavy metal content is the standard threshold of the heavy metal content. The value is output to obtain the first heavy metal standard content threshold, that is, the heavy metal content in the aluminum foil lunch boxes obtained by production and processing shall not exceed the first heavy metal standard content threshold. If it exceeds the first heavy metal standard content threshold, it is an unqualified aluminum foil lunch box. The acquisition of the first heavy metal standard content threshold provides a data basis for the subsequent physical defect detection of the first heavy metal on the aluminum foil lunch box when the first heavy metal standard content threshold is greater than or equal to the detection trigger content threshold.

[0023] When the predicted content of the first heavy metal is greater than or equal to the standard content threshold of the first heavy metal, the aluminum foil lunch box is subjected to a physical defect detection of the first heavy metal.

[0024] Physical defect detection refers to the actual heavy metal detection of aluminum foil lunch boxes, including atomic spectroscopy, spectrophotometry, electrochemistry and x-ray fluorescence spectroscopy. Different detection methods are selected according to different situations. When the standard content of the first heavy metal is greater than or equal to the standard content threshold of the first heavy metal, the aluminum foil lunch box is subjected to physical detection of the first heavy metal. When the standard content of the first heavy metal is greater than or equal to the standard content threshold of the first heavy metal, it means that the detection trigger content threshold is reached first. When the detection trigger content threshold is reached, the physical defect of the first heavy metal is detected to ensure that the heavy metal content of the aluminum foil lunch box is lower than the standard content threshold of the first heavy metal, and the heavy metal content in the aluminum foil lunch box is absolutely controlled within the safe range. Physical defect detection is performed on problematic aluminum foil lunch boxes to reduce the inefficiency caused by all detections and improve overall production efficiency.

[0025] Furthermore, the method further comprises:

[0026] Downloading a first heavy metal calibration master function from an aluminum foil lunch box production sharing cloud service platform, wherein the first heavy metal calibration master function is a machine learning model for predicting heavy metal content;

[0027] Based on the production line number of the aluminum foil lunch box, production backtracking is performed in a preset time zone to extract a memory value set of the aluminum foil mother roll preparation process, a memory value set of the raw material type, a memory value set of the raw material ratio, and a memory value set of the first heavy metal mother roll detection content;

[0028] The first heavy metal calibration master function is verified according to the aluminum foil mother roll preparation process memory value set, the raw material type memory value set, the raw material ratio memory value set and the first heavy metal mother roll detection content memory value set to generate a content prediction verification loss;

[0029] When the content prediction verification loss is greater than or equal to the content prediction verification loss threshold, the first heavy metal calibration mother function is subjected to residual training according to the aluminum foil mother roll preparation process memory value set, the raw material type memory value set, the raw material ratio memory value set and the first heavy metal mother roll detection content memory value set to generate a first heavy metal calibration sub-function, wherein the first heavy metal calibration sub-function is a machine learning model obtained by performing enhanced training based on the first heavy metal calibration mother function;

[0030] The first heavy metal calibration sub-function is activated, the aluminum foil mother roll preparation process, the raw material type set and the raw material ratio set are received, the mother roll content of the first heavy metal of the raw material type set is calibrated, and the first heavy metal predicted content is generated.

[0031] The aluminum foil lunch box production sharing cloud service platform refers to an online service platform for sharing the production data of aluminum foil lunch boxes. Through the signing of confidentiality agreements by multiple aluminum foil lunch box manufacturers, they jointly become members of the sharing service, and the production data of all sharing members is shared and obtained through the aluminum foil lunch box production sharing cloud service platform. The first metal calibration mother function is downloaded and obtained through the aluminum foil lunch box production sharing cloud service platform, where the first heavy metal calibration mother function is the aluminum foil mother roll preparation process memory value set, raw material type memory value set, raw material ratio memory value set and the first heavy metal mother roll detection content memory value set extracted from each factory, and the mother function is extracted from the aluminum foil mother roll preparation process memory value set, raw material type memory value set, raw material ratio memory value set and the first heavy metal mother roll detection content memory value set through a conventional neural network. By tracing back the production line number of the aluminum foil lunch box in a preset time zone, the production process information in the tail number of the aluminum foil lunch box production line within the preset time is obtained, and the historical aluminum foil mother roll preparation process information, historical raw material type information, historical raw material ratio information, and historical first heavy metal mother roll detection content information are obtained, and the aluminum foil mother roll preparation process memory value set, raw material type memory value set, raw material ratio memory value set, and first heavy metal mother roll detection content memory value set are obtained respectively. The first heavy metal calibration mother function is verified according to the previous data, wherein the first heavy metal calibration mother function is a machine learning model for predicting heavy metal content, and the first heavy metal mother function is corrected, and the difference between the first heavy metal calibration mother function and the previous first heavy metal mother roll detection content memory value set is first obtained as a loss value and output, and the content prediction verification loss is obtained. The content prediction verification loss is determined, and a content prediction verification loss threshold is set. When the content prediction verification loss is greater than or equal to the content prediction verification loss threshold, it indicates that the current first heavy metal calibration mother function has a large deviation and cannot be used directly, and further processing is required. The first heavy metal calibration mother function is subjected to residual training, and the first heavy metal calibration mother function is corrected and optimized to obtain a corrected optimization result, which is the first heavy metal calibration sub-function. That is, the first heavy metal calibration sub-function is a machine learning model obtained by performing enhanced training based on the first heavy metal calibration mother function, and the first heavy metal calibration sub-function is used to calibrate the mother roll content of the first heavy metal in the raw material type set of the aluminum foil mother roll preparation process, the raw material type set and the raw material ratio set to obtain a calibration result, which is the first heavy metal predicted content. When the content prediction verification loss is less than the content prediction verification loss threshold, the first heavy metal calibration mother function can be used directly, and the mother roll content of the first heavy metal in the raw material type set of the aluminum foil mother roll preparation process, the raw material type set and the raw material ratio set is calibrated to generate the first heavy metal predicted content. The generation of the predicted content of the first heavy metal provides a data basis for subsequent physical defect detection of the first heavy metal on the aluminum foil lunch box when the standard content threshold of the first heavy metal is greater than or equal to the detection trigger content threshold.

[0032] Furthermore, the method further comprises:

[0033] Extract 20% of the data from the aluminum foil mother roll preparation process memory value set, the raw material type memory value set, the raw material ratio memory value set, and the first heavy metal mother roll detection content memory value set, and set it as a verification memory value set;

[0034] Traversing the verification memory value set, activating the first heavy metal calibration master function, mapping the aluminum foil mother roll preparation process memory value, the raw material type memory value and the raw material ratio memory value, and generating a first heavy metal content calibration value;

[0035] Counting the absolute value of the deviation between the first heavy metal content memory value detected by the mother roll and the first heavy metal content calibration value;

[0036] A feature binarization process is performed on several absolute values ​​of deviations, and an inaccurate prediction ratio coefficient is calculated and set as the content prediction verification loss, wherein the inaccurate prediction ratio coefficient refers to the proportion of deviations with lower prediction accuracy in several absolute values ​​of deviations.

[0037] When generating the content verification loss, first extract some data from the aluminum foil mother roll preparation process memory value set, the raw material type memory value set, the raw material ratio memory value set and the first heavy metal mother roll detection content memory value set as the verification set for correcting the first heavy metal calibration function, extract 20% of the data, and use the extracted data as the verification memory value set. Traverse the verification memory value set to retrieve the first heavy metal content calibration value corresponding to the aluminum foil mother roll preparation process memory value, the raw material type memory value and the raw material ratio memory value. The first heavy metal content calibration value is the first heavy metal predicted content obtained by calibrating the aluminum foil mother roll preparation process memory value, the raw material type memory value and the raw material ratio memory value. By comparing the first heavy metal content calibration value with the first heavy metal mother roll detection content memory value, the deviation value is obtained, and the deviation value is absoluteized to obtain an absolute deviation value. The deviation value is then processed by feature binarization, and the percentage of the deviation number, that is, the inaccurate prediction ratio coefficient, is calculated. The inaccurate prediction ratio coefficient is the proportion of the number of deviations with lower prediction accuracy in several absolute values ​​of deviations, and the inaccurate prediction ratio coefficient is used as the content prediction verification loss. Feature binarization is a data preprocessing method that converts features from continuous numerical types to binary discrete types, and counts the number of 0s and 1s to facilitate percentage calculation of the data to obtain the inaccurate prediction ratio coefficient. The acquisition of content prediction verification loss provides a data basis for the subsequent residual training of the first heavy metal calibration mother function based on the aluminum foil mother roll preparation process memory value set, raw material type memory value set, raw material ratio memory value set and the first heavy metal mother roll detection content memory value set to generate the first heavy metal calibration sub-function when the content prediction verification loss is greater than or equal to the content prediction verification loss threshold.

[0038] Furthermore, the method further comprises:

[0039] When the absolute value of the deviation is greater than or equal to the deviation threshold, a 1 value flag is generated;

[0040] When the absolute value of the deviation is less than the deviation threshold, a zero value flag is generated;

[0041] The sum of the binary numbers of the number of 1-valued identifications and the number of 0-valued identifications is calculated, and the proportion of the number of 1-valued identifications in the binary number sum is calculated to generate the inaccurate prediction ratio coefficient, which is set as the content prediction verification loss.

[0042] The absolute value of the deviation is binarized, that is, the absolute value of the deviation is converted into 0 and 1. When the absolute value of the deviation is greater than or equal to the deviation threshold, a 1-value identifier is generated and the 1-value identifier is obtained; when the absolute value of the deviation is less than the deviation threshold, a 0-value identifier is generated and the generated 0-value identifier is obtained. 1 and 0 identifiers are generated for all absolute values ​​of the deviation, and one absolute value of the deviation is generated once. All generated identifiers are counted to obtain the sum of the binary numbers of the number of 1-value identifiers and the number of 0-value identifiers, and the number of 1-value identifiers and the number of 0-value identifiers are obtained. The proportion of the number of 1-value identifiers in the total number of binary numbers is calculated to obtain the count. The calculation result is output to obtain the inaccurate prediction ratio coefficient, which refers to the number of absolute values ​​of deviations greater than or equal to the deviation threshold in the total absolute values ​​of deviations, and the inaccurate prediction ratio coefficient is set as the content prediction verification loss. The content prediction verification loss is obtained for the subsequent when the content prediction verification loss is greater than or equal to the content prediction verification loss threshold. According to the aluminum foil mother roll preparation process memory value set, the raw material type memory value set, the raw material ratio memory value set and the first heavy metal mother roll detected content memory value set, the first heavy metal calibration mother function is residually trained to provide a data basis for generating the first heavy metal calibration sub-function.

[0043] Furthermore, the method further comprises:

[0044] The aluminum foil mother roll preparation process memory value set, the raw material type memory value set, the raw material ratio memory value set and the first heavy metal mother roll detection content memory value set are divided into k groups to generate a data grouping result

[0045] Training the first heavy metal calibration master function according to the first data grouping of the data grouping result to generate a first data grouping error set and a first heavy metal predicted content set;

[0046] Performing residual fitting training with the first data grouping error set, the first heavy metal predicted content set, and the first heavy metal mother roll detected content memory value set to generate a first residual function;

[0047] Based on the second data grouping of the data grouping result, the first heavy metal calibration mother function and the first residual function are jointly verified to generate a first joint verification loss;

[0048] When the first joint verification loss is less than the content prediction verification loss threshold, merging the first heavy metal calibration mother function and the first residual function to generate the first heavy metal calibration sub-function;

[0049] When the first joint verification loss is greater than or equal to the content prediction verification loss threshold, residual training is performed based on the second data group to generate a second residual function.

[0050] Residual training specifically refers to receiving the output value and loss value of the previous training, and using the output value and loss value of the previous training as the input of this training, taking the target prediction value as the supervision data, fitting the output error of the previous training, and each residual function includes a residual mapping chain, each residual chain has multiple residual blocks, and training is the process of constructing residual blocks. The aluminum foil mother roll preparation process memory value set, the raw material type memory value set, the raw material ratio memory value set and the first heavy metal mother roll detection content memory value set are divided to obtain a division result, and the division result is a data grouping result. The first data group of the data grouping result is obtained to obtain the first data group, and the first heavy metal calibration mother function is trained on the first data group. The first heavy metal calibration mother function is used to calibrate the first data group to obtain the corresponding calibration result, that is, the first heavy metal prediction content set, and the error analysis of the first heavy metal prediction content set is performed according to the first data group to obtain the first data group error set. Based on the neural network, residual fitting training is performed on the first data grouping error set, the first heavy metal predicted content set and the first heavy metal mother roll detected content memory value set to obtain the fitting training result. The fitting training result is the first residual function. The first residual function represents the residual mapping function obtained by the error set for the first heavy metal predicted content set and the first heavy metal mother roll detected content memory value set. The error of the first heavy metal predicted content set can be obtained through the function. The error is combined with the first heavy metal predicted content set to obtain the correct first heavy metal predicted content. The first residual function is also a residual block in the residual mapping chain. Then the second data group of the data grouping result is obtained, and the first heavy metal calibration mother function and the first residual function are jointly verified through the second data group. The error between the first heavy metal calibration mother function and the first residual function is output, and the error between the first residual function and the first heavy metal calibration mother function is obtained, and the error is added to the second data group, and the data is fitted with the second data group as the input value for the next training, that is, the first joint verification loss is generated. When the first joint verification loss is less than the content prediction verification loss threshold, it means that the first joint verification loss is qualified, and the first heavy metal calibration mother function and the first residual function are combined to generate the first heavy metal calibration sub-function; when the first joint verification loss is greater than or equal to the content prediction verification loss threshold, residual training is performed based on the second data grouping to generate a second residual function, and joint verification is performed until the generated joint verification loss is less than the content prediction verification loss threshold. By dividing the aluminum foil mother roll preparation process memory value set, the raw material type memory value set, the raw material ratio memory value set and the first heavy metal mother roll detection content memory value set into multiple groups and performing residual training, the training efficiency can be significantly improved. When the verification is accurate, there is no need to continue training with subsequent data, and it can be directly regarded as convergence, thereby improving the overall system efficiency.

[0051] Furthermore, the method further comprises:

[0052] The first predicted heavy metal content and the first data group average error are used as data to be fitted, and the first heavy metal mother roll detected content memory value is used as fitting identification data, and fitting is performed based on the neural network structure to generate a first residual block, wherein the first residual block has a first fitting error and a first fitting value;

[0053] When the absolute value of the first fitting error is greater than or equal to a preset characteristic value, the first fitting error and the first fitting value are used as data to be fitted, and the first heavy metal mother roll detection content memory value is used as fitting identification data to perform fitting based on the neural network structure to generate a second residual block. Training is performed based on the second residual block until the absolute value of the fitting error is less than the preset characteristic value to obtain the first residual function.

[0054] When the absolute value of the first fitting error is smaller than the preset characteristic value, the first residual block is set to the first residual function.

[0055] The first heavy metal predicted content and the first data group average error are used as the data to be fitted, and the first heavy metal mother roll detected content memory value is used as the fitting identification data, that is, the fitting target value, and fitting is performed based on the neural network structure to generate a first residual block, wherein the first residual block has a first fitting error and a first fitting value, and the first fitting error is the error between the first fitting value and the fitting identification data. When the absolute value of the first fitting error is greater than or equal to the preset characteristic value, the first fitting error and the first fitting value are used as the data to be fitted, and the first heavy metal mother roll detected content memory value is used as the fitting identification data to perform fitting based on the neural network structure to generate a second residual block, wherein the preset characteristic value is the tolerance range of the fitting identification data, indicating that when the closed result is within a certain range of the fitting identification data, it is qualified fitting data, which can reduce a large number of calculations generated by reaching a fixed fitting value and improve fitting efficiency. When the absolute value of the first fitting error is less than the preset characteristic value, it means that the first fitting error is qualified, and the first residual block is set to the first residual function. Each training data is trained based on the residual function of the previous training. If the result obtained based on the residual function training of the previous training converges, there is no need to add residual blocks. If the residual function of the previous training still has errors, it is necessary to add residual blocks and modify the residual function until convergence. Each input data value can be fitted in the residual function to ensure accuracy.

[0056] Embodiment 2

[0057] Based on the same inventive concept as the aluminum foil lunch box defect detection method in the aforementioned embodiment, Figure 2As shown, the present application provides an aluminum foil lunch box defect detection system, the system comprising:

[0058] Aluminum foil raw material table activation module 11: The aluminum foil raw material table activation module 11 is used to activate the aluminum foil raw material table of the aluminum foil lunch box and extract the raw material type set and the raw material ratio set;

[0059] The mother roll content calibration module 12 is used to receive the aluminum foil mother roll preparation process, combine the raw material type set and the raw material ratio set, perform mother roll content calibration on the first heavy metal of the raw material type set, and generate the predicted content of the first heavy metal;

[0060] Standard content threshold value extraction module 13: The standard content threshold value extraction module 13 is used to extract the first heavy metal standard content threshold value from the standard table of metal raw materials for aluminum foil lunch boxes;

[0061] Physical defect detection module 14: The physical defect detection module 14 is used to perform physical defect detection of the first heavy metal on the aluminum foil lunch box when the predicted content of the first heavy metal is greater than or equal to the standard content threshold of the first heavy metal.

[0062] Furthermore, the mother roll content calibration module 12 includes the following execution steps:

[0063] Downloading a first heavy metal calibration master function from an aluminum foil lunch box production sharing cloud service platform, wherein the first heavy metal calibration master function is a machine learning model for predicting heavy metal content;

[0064] Based on the production line number of the aluminum foil lunch box, production backtracking is performed in a preset time zone to extract a memory value set of the aluminum foil mother roll preparation process, a memory value set of the raw material type, a memory value set of the raw material ratio, and a memory value set of the first heavy metal mother roll detection content;

[0065] The first heavy metal calibration master function is verified according to the aluminum foil mother roll preparation process memory value set, the raw material type memory value set, the raw material ratio memory value set and the first heavy metal mother roll detection content memory value set to generate a content prediction verification loss;

[0066] When the content prediction verification loss is greater than or equal to the content prediction verification loss threshold, the first heavy metal calibration mother function is subjected to residual training according to the aluminum foil mother roll preparation process memory value set, the raw material type memory value set, the raw material ratio memory value set and the first heavy metal mother roll detection content memory value set to generate a first heavy metal calibration sub-function, wherein the first heavy metal calibration sub-function is a machine learning model obtained by performing enhanced training based on the first heavy metal calibration mother function;

[0067] The first heavy metal calibration sub-function is activated, the aluminum foil mother roll preparation process, the raw material type set and the raw material ratio set are received, the mother roll content of the first heavy metal of the raw material type set is calibrated, and the first heavy metal predicted content is generated.

[0068] Furthermore, the mother roll content calibration module 12 includes the following execution steps:

[0069] Extract 20% of the data from the aluminum foil mother roll preparation process memory value set, the raw material type memory value set, the raw material ratio memory value set, and the first heavy metal mother roll detection content memory value set, and set it as a verification memory value set;

[0070] Traversing the verification memory value set, activating the first heavy metal calibration master function, mapping the aluminum foil mother roll preparation process memory value, the raw material type memory value and the raw material ratio memory value, and generating a first heavy metal content calibration value;

[0071] Counting the absolute value of the deviation between the first heavy metal content memory value detected by the mother roll and the first heavy metal content calibration value;

[0072] A feature binarization process is performed on several absolute values ​​of deviations, and an inaccurate prediction ratio coefficient is calculated and set as the content prediction verification loss, wherein the inaccurate prediction ratio coefficient refers to the proportion of deviations with lower prediction accuracy in several absolute values ​​of deviations.

[0073] Furthermore, the mother roll content calibration module 12 includes the following execution steps:

[0074] When the absolute value of the deviation is greater than or equal to the deviation threshold, a 1 value flag is generated;

[0075] When the absolute value of the deviation is less than the deviation threshold, a zero value flag is generated;

[0076] The sum of the binary numbers of the number of 1-valued identifications and the number of 0-valued identifications is calculated, and the proportion of the number of 1-valued identifications in the binary number sum is calculated to generate the inaccurate prediction ratio coefficient, which is set as the content prediction verification loss.

[0077] Furthermore, the mother roll content calibration module 12 includes the following execution steps:

[0078] The aluminum foil mother roll preparation process memory value set, the raw material type memory value set, the raw material ratio memory value set and the first heavy metal mother roll detection content memory value set are divided into k groups to generate a data grouping result

[0079] Training the first heavy metal calibration master function according to the first data grouping of the data grouping result to generate a first data grouping error set and a first heavy metal predicted content set;

[0080] Performing residual fitting training with the first data grouping error set, the first heavy metal predicted content set, and the first heavy metal mother roll detected content memory value set to generate a first residual function;

[0081] Based on the second data grouping of the data grouping result, the first heavy metal calibration mother function and the first residual function are jointly verified to generate a first joint verification loss;

[0082] When the first joint verification loss is less than the content prediction verification loss threshold, merging the first heavy metal calibration mother function and the first residual function to generate the first heavy metal calibration sub-function;

[0083] When the first joint verification loss is greater than or equal to the content prediction verification loss threshold, residual training is performed based on the second data group to generate a second residual function.

[0084] Furthermore, the mother roll content calibration module 12 includes the following execution steps:

[0085] The first predicted heavy metal content and the first data group average error are used as data to be fitted, and the first heavy metal mother roll detected content memory value is used as fitting identification data, and fitting is performed based on the neural network structure to generate a first residual block, wherein the first residual block has a first fitting error and a first fitting value;

[0086] When the absolute value of the first fitting error is greater than or equal to a preset characteristic value, the first fitting error and the first fitting value are used as data to be fitted, and the first heavy metal mother roll detection content memory value is used as fitting identification data to perform fitting based on the neural network structure to generate a second residual block. Training is performed based on the second residual block until the absolute value of the fitting error is less than the preset characteristic value to obtain the first residual function.

[0087] Furthermore, the method further comprises:

[0088] When the absolute value of the first fitting error is smaller than the preset characteristic value, the first residual block is set to the first residual function.

[0089] Through the above detailed description of the aluminum foil lunch box defect detection method in this specification, those skilled in the art can clearly understand the aluminum foil lunch box defect detection method in this embodiment. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.

[0090] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. Aluminum foil lunch box defect detection method, characterized in that: The method comprises: Activate the aluminum foil raw material table of the aluminum foil lunch box, and extract the raw material type set and raw material ratio set; receiving an aluminum foil mother roll preparation process, combining the raw material type set and the raw material ratio set, calibrating the mother roll content of a first heavy metal of the raw material type set, and generating a predicted first heavy metal content; Extract the first heavy metal standard content threshold from the standard table of metal raw materials for aluminum foil lunch boxes; When the predicted content of the first heavy metal is greater than or equal to the standard content threshold of the first heavy metal, performing a physical defect detection of the first heavy metal on the aluminum foil lunch box; The aluminum foil mother roll preparation process is received, and the first heavy metal of the raw material type set and the raw material ratio set are combined to calibrate the mother roll content, and generate the predicted content of the first heavy metal, including: Downloading a first heavy metal calibration master function from an aluminum foil lunch box production sharing cloud service platform, wherein the first heavy metal calibration master function is a machine learning model for predicting heavy metal content; Based on the production line number of the aluminum foil lunch box, production backtracking is performed in a preset time zone to extract a memory value set of the aluminum foil mother roll preparation process, a memory value set of the raw material type, a memory value set of the raw material ratio, and a memory value set of the first heavy metal mother roll detection content; The first heavy metal calibration master function is verified according to the aluminum foil mother roll preparation process memory value set, the raw material type memory value set, the raw material ratio memory value set and the first heavy metal mother roll detection content memory value set to generate a content prediction verification loss; When the content prediction verification loss is greater than or equal to the content prediction verification loss threshold, the first heavy metal calibration mother function is subjected to residual training according to the aluminum foil mother roll preparation process memory value set, the raw material type memory value set, the raw material ratio memory value set and the first heavy metal mother roll detection content memory value set to generate a first heavy metal calibration sub-function, wherein the first heavy metal calibration sub-function is a machine learning model obtained by performing enhanced training based on the first heavy metal calibration mother function; The first heavy metal calibration sub-function is activated, the aluminum foil mother roll preparation process, the raw material type set and the raw material ratio set are received, the mother roll content of the first heavy metal of the raw material type set is calibrated, and the first heavy metal predicted content is generated.

2. The method according to claim 1, characterized in that The first heavy metal calibration master function is verified according to the aluminum foil mother roll preparation process memory value set, the raw material type memory value set, the raw material ratio memory value set and the first heavy metal mother roll detection content memory value set to generate a content prediction verification loss, including: Extract 20% of the data from the aluminum foil mother roll preparation process memory value set, the raw material type memory value set, the raw material ratio memory value set, and the first heavy metal mother roll detection content memory value set, and set it as a verification memory value set; Traversing the verification memory value set, activating the first heavy metal calibration master function, mapping the aluminum foil mother roll preparation process memory value, the raw material type memory value and the raw material ratio memory value, and generating a first heavy metal content calibration value; Counting the absolute value of the deviation between the first heavy metal content memory value detected by the mother roll and the first heavy metal content calibration value; A feature binarization process is performed on several absolute values ​​of deviations, and an inaccurate prediction ratio coefficient is calculated and set as the content prediction verification loss, wherein the inaccurate prediction ratio coefficient refers to the proportion of deviations with lower prediction accuracy in several absolute values ​​of deviations.

3. The method according to claim 2, characterized in that Performing feature binarization processing on a number of absolute values ​​of deviations, calculating the inaccurate prediction ratio coefficient, and setting it as the content prediction verification loss, including: When the absolute value of the deviation is greater than or equal to the deviation threshold, a 1 value flag is generated; When the absolute value of the deviation is less than the deviation threshold, a zero value flag is generated; The sum of the binary numbers of the number of 1-valued identifications and the number of 0-valued identifications is calculated, and the proportion of the number of 1-valued identifications in the binary number sum is calculated to generate the inaccurate prediction ratio coefficient, which is set as the content prediction verification loss.

4. The method according to claim 1, characterized in that When the content prediction verification loss is greater than or equal to the content prediction verification loss threshold, the first heavy metal calibration mother function is subjected to residual training according to the aluminum foil mother roll preparation process memory value set, the raw material type memory value set, the raw material ratio memory value set and the first heavy metal mother roll detection content memory value set to generate a first heavy metal calibration sub-function, including: Divide the aluminum foil mother roll preparation process memory value set, the raw material type memory value set, the raw material ratio memory value set, and the first heavy metal mother roll detection content memory value set into k groups to generate a data grouping result; Training the first heavy metal calibration master function according to the first data grouping of the data grouping result to generate a first data grouping error set and a first heavy metal predicted content set; Performing residual fitting training with the first data grouping error set, the first heavy metal predicted content set, and the first heavy metal mother roll detected content memory value set to generate a first residual function; Performing joint verification on the first heavy metal calibration mother function and the first residual function based on a second data grouping of the data grouping result to generate a first joint verification loss; When the first joint verification loss is less than the content prediction verification loss threshold, merging the first heavy metal calibration mother function and the first residual function to generate the first heavy metal calibration sub-function; When the first joint verification loss is greater than or equal to the content prediction verification loss threshold, residual training is performed based on the second data group to generate a second residual function.

5. The method according to claim 4, characterized in that Performing residual fitting training with the first data grouping error set, the first heavy metal predicted content set, and the first heavy metal mother roll detected content memory value set to generate a first residual function, including: The first predicted heavy metal content and the first data group average error are used as data to be fitted, and the first heavy metal mother roll detected content memory value is used as fitting identification data, and fitting is performed based on the neural network structure to generate a first residual block, wherein the first residual block has a first fitting error and a first fitting value; When the absolute value of the first fitting error is greater than or equal to a preset characteristic value, the first fitting error and the first fitting value are used as data to be fitted, and the first heavy metal mother roll detection content memory value is used as fitting identification data to perform fitting based on the neural network structure to generate a second residual block. Training is performed based on the second residual block until the absolute value of the fitting error is less than the preset characteristic value to obtain the first residual function.

6. The method according to claim 5, characterized in that Also includes: When the absolute value of the first fitting error is smaller than the preset characteristic value, the first residual block is set to the first residual function.

7. Aluminum foil lunch box defect detection system, characterized in that: The system is used to execute the method according to any one of claims 1 to 6, and the system comprises: Aluminum foil raw material table activation module: activates the aluminum foil raw material table of the aluminum foil lunch box, extracts the raw material type set and raw material ratio set; A mother roll content calibration module receives the aluminum foil mother roll preparation process, combines the raw material type set and the raw material ratio set, performs mother roll content calibration on the first heavy metal of the raw material type set, and generates a predicted content of the first heavy metal; Standard content threshold extraction module: extract the first heavy metal standard content threshold from the standard table of metal raw materials for aluminum foil lunch boxes; Physical defect detection module: when the predicted content of the first heavy metal is greater than or equal to the standard content threshold of the first heavy metal, the aluminum foil lunch box is subjected to physical defect detection of the first heavy metal.

Images