A method, system, readable storage medium and device for analyzing and inspecting a filled product

Abstract

The application provides a filling product analysis and quality inspection method based on a millimeter wave radar and a deep learning model, aiming to improve the detection accuracy of a can body tail folding and a cap locking. The millimeter wave radar is used to detect the can body tail folding, data is segmented based on a loss value, the hollow rate of the can body tail folding is calculated, and the can body with incomplete tail folding packaging is marked. An infrared thermal imaging technology is used to obtain a cap locking image, a deep learning model is used to analyze the integrity of the cap locking, whether the can body is completely packaged is judged, and the filling product with incomplete cap locking packaging is marked. In addition, the application also provides a quality inspection system, a readable storage medium and a device, so as to realize the automation and intelligentization of the above method.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of production control, and relates to a filling product analysis and quality inspection method, system, readable storage medium and device. BACKGROUND

[0002] In the large-scale production of filling products, the tail of the tank body is first folded, that is, the tail of the tank body is packaged. After the tail folding is completed, the filling process is performed, and after the filling is completed, the cover body is locked with the head of the tank body, and the locking operation is a packaging operation after the tank body is filled.

[0003] The traditional water injection test method takes a long time and has high cost when detecting the folded tail of the tank body. When detecting the integrity of the locking operation, it is difficult to accurately identify the state of the locking operation, which leads to the inability to effectively judge the sealing between the cover body and the tank body. When marking the products with incomplete packaging, it is difficult to judge in real time and accurately. SUMMARY

[0004] To solve the problems in the background art, the technical scheme adopted by the application is as follows:

[0005] A filling product analysis and quality inspection method comprises:

[0006] The folded tail of the tank body to be filled is detected by using a millimeter wave radar, and the millimeter wave radar detection data is segmented based on a loss value.

[0007] After the filling product is filled, the tank body is locked, the infrared thermal image of the tank body after the locking is obtained, and the integrity of the locking operation in the tank body infrared thermal image is analyzed by using a deep learning model.

[0008] The segmented millimeter wave radar detection data and the integrity of the locking operation are integrated and analyzed to obtain the tank body closure of the current filling product.

[0009] Based on the tank body closure, it is judged whether the tank body is packaged completely, and the tank body with incomplete packaging is marked.

[0010] Further, the specific method for detecting the folded tail of the tank body to be filled by using the millimeter wave radar and segmenting the millimeter wave radar detection data based on the loss value is as follows:

[0011] The echo data of the millimeter wave radar is obtained, including distance information, which is recorded in the form of time sequence.

[0012] The loss value of the radar signal echo is calculated, and the specific formula is as follows:

[0013]

[0014] Wherein r is the radar to the tank body tail folding distance, r0 is the radar to the tank body tail folding distance reference value, n is the radar wave transmission path loss index;

[0015] A sliding window is created in the echo data of the millimeter wave radar, the sliding window is used to separate multiple windows for the echo data of the millimeter wave radar, and the average value of the loss value is calculated in each window;

[0016] The clustering algorithm is used to classify the loss value, and each type of millimeter wave radar echo data is labeled with a loss value label, and the millimeter wave radar detection data segmentation is completed.

[0017] Further, the specific method for acquiring the tank infrared thermal image of the completed lock cover and analyzing the integrity of the lock cover operation corresponding to the tank infrared thermal image through the deep learning model is:

[0018] The position of the lock cover in the tank infrared thermal image is extracted by the Harris corner point detection method, and a feature set around the lock cover is extracted;

[0019] The information gain of each feature in the feature set around the lock cover is calculated, and the specific formula is:

[0020]

[0021] Wherein H(D) is the entropy of the feature set around the lock cover, H(D v ) is the conditional entropy based on the feature A, the influence of the unimportant feature is reduced through the penalty term, and the first N features are selected, wherein N is a positive integer;

[0022] Based on the first N features, the integrity of the lock cover operation corresponding to all tank infrared thermal images is judged through the deep learning model.

[0023] Further, the specific method for acquiring the tank infrared thermal image of the completed lock cover and analyzing the integrity of the lock cover operation corresponding to the tank infrared thermal image through the deep learning model is:

[0024] Each feature in the first N features is arranged according to the size of the specific value to generate an N feature set;

[0025] Randomly select a feature set M, mark the last feature value and the first feature value in M, randomly select a value between the last feature value and the first feature value to divide the feature set M, and mark the division value;

[0026] According to the above method, the median position of the feature set M is iterated until all features in the feature set M are labeled, the average of all division values is selected, the discrete threshold is set, and the corresponding discrete value of all feature values in the feature set M is obtained;

[0027] The discrete value of each feature value Mi in the feature set M is taken as the numerator, the difference between the maximum feature value and the minimum feature value is taken as the denominator, the value is recorded as Zi, and the value of the integrity of the locking cover operation is set to 1-Zi.

[0028] Further, the specific method for integrating and analyzing the segmented millimeter wave radar detection data and the integrity of the locking cover operation to obtain the specific sealing degree of the current filling product is as follows:

[0029] The gradient change threshold of each segment of millimeter wave radar exploration data is calculated, each segment of millimeter wave radar detection data is multiplied by the corresponding loss value ratio, and the hollow rate of the can tail is obtained.

[0030] The can tail hollow rate threshold and the locking cover operation integrity threshold are set.

[0031] The can tail hollow rate and the locking cover operation integrity are calculated, and if both conditions are met, it indicates that the current can is complete.

[0032] In another aspect, the present application provides a filling product analysis and quality inspection system, comprising:

[0033] A data acquisition module acquires millimeter wave radar detection data of the can tail to be filled, and acquires an infrared thermal image of the can body after the locking cover is completed.

[0034] A data processing module segments the millimeter wave radar detection data based on the loss value, analyzes the integrity of the locking cover operation corresponding to the can body infrared thermal image, and further comprises calculating the can body sealing degree of the filling product.

[0035] A control module marks the can body that is not completely sealed.

[0036] In another aspect, the present application provides a computer readable storage medium storing the above-mentioned system.

[0037] In another aspect, the present application provides a filling product analysis and quality inspection device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor executes the above-mentioned filling product analysis and quality inspection method through the computer program.

[0038] Compared with the prior art, the present application has the following beneficial effects:

[0039] By using millimeter wave radar technology to accurately detect the can tail, and segmenting the detection data based on the loss value, this method can more accurately identify the changes in the internal structure of the can, and improve the detection accuracy.

[0040] The infrared thermal imaging technology and the deep learning model are combined to analyze the integrity of the locking cover operation of the tank, which can quickly determine whether the locking cover operation is intact, reduces the need for manual inspection, and improves the efficiency of the entire detection process.

[0041] The present application not only considers the distortion of the detection data at the tail of the tank, but also combines the integrity of the locking cover operation. Through analysis of the data in these two aspects, the sealing performance of the tank can be comprehensively evaluated to ensure the sealing and safety of the product.

[0042] Based on the judgment standard of the tank sealing degree, the products with incomplete packaging can be effectively identified and marked to avoid unqualified products entering the market. BRIEF DESCRIPTION OF DRAWINGS

[0043] Figure 1 is a method flowchart of the present application;

[0044] Figure 2 is a system structure diagram of the present application. DETAILED DESCRIPTION

[0045] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0046] As shown in Figure 1 , a method for analyzing and inspecting filled products includes:

[0047] In the process of producing filled products, a millimeter wave radar is used to detect the tail folding condition of a to-be-filled tank. Through a specific algorithm, the data detected by the millimeter wave radar is segmented and processed according to the loss value, so as to more accurately analyze the data characteristics. The tail folding is a packaging operation of the tail of the tank. After the segmentation processing is completed, the hollow rate of the tank tail is calculated based on the segmentation result, and the tank with incomplete tail packaging is marked.

[0048] After the tank with complete tail packaging is filled, a locking cover operation is performed on the tank. The locking cover operation is a packaging operation after the tank is filled.

[0049] At this time, an infrared thermal image of the tank after the locking cover operation is obtained by using infrared thermal imaging technology, and the image is analyzed in depth by means of a deep learning model, so as to determine the integrity of the locking cover operation, such as detecting whether the locking cover is tight, whether there is deformation or gap, etc.

[0050] Then, based on the integrity of the locking cover operation, the filling product with incomplete locking cover packaging is marked. If the locking cover operation does not reach the set standard, it is determined that the can body is not completely packaged, and the corresponding filling product is marked, so as to ensure the product packaging quality and protect the safety and stability of the product in the storage, transportation and sales process.

[0051] Before the filling process, a millimeter wave radar is used to detect the fold tail condition of the to-be-filled can body. After the millimeter wave radar collects data, the radar data is segmented.

[0052] The echo data of the millimeter wave radar, including distance information, is recorded in time series. These information can be collected by multiple pulse echoes and recorded as a time series. Generally, the radar converts the echo signal into a distance map when receiving the echo.

[0053] The loss value can be defined in several ways, usually based on the amplitude attenuation of the echo signal, the decrease of signal-to-noise ratio or error, etc. The attenuation of the radar signal is usually inversely proportional to the square of the propagation distance, and the loss value can be calculated using the following formula:

[0054]

[0055] Where r is the radar distance to the can body fold tail, r0 is the reference value of the radar distance to the can body fold tail, and n is the radar wave transmission path loss index. The increase of the attenuation amplitude usually means the decrease of the quality of the radar detection.

[0056] A sliding window is constructed based on the echo data of the millimeter wave radar, and the echo data is divided into several independent windows by means of the sliding window. In the range of each window, the average value of the loss value is accurately calculated. Advanced clustering algorithm is used to classify the loss value, and each class of millimeter wave radar echo data is given a corresponding loss value label according to the classification result, so as to realize the effective segmentation of the millimeter wave radar detection data.

[0057] The gradient change threshold of each segment of millimeter wave radar exploration data is calculated. The gradient change of the radar data can reflect the difference of the objects detected by the radar, for example, the wave returned by the radar when detecting material defects has obvious density change.

[0058] Each segment of millimeter wave radar detection data is multiplied by the corresponding loss value proportion, and the loss proportion is the proportion of the loss wave of the radar data in the whole radar data. The hollow rate of the can body fold tail is obtained by the above calculation.

[0059] The fold tail hollow rate threshold is set.

[0060] The hollow rate of the tail of the can body is calculated, and if the tail hollow rate threshold condition is met, it indicates that the current can body tail packaging is complete. The can body that does not meet the tail hollow rate threshold is marked to avoid entering the subsequent production process.

[0061] After the can body with complete tail packaging is filled, the can body is immediately locked. The infrared thermal image of the can body after locking is obtained by using infrared thermal imaging technology, and the image is analyzed in depth by means of a deep learning model, so as to judge the completeness of the locking operation, such as detecting whether the locking is tight, whether there is deformation or gap, etc.

[0062] The infrared thermal image thereof is obtained. The harris corner point detection technology is used to accurately extract the position information of the cover in the infrared thermal image of the can body, and then a feature set around the cover is extracted.

[0063] The information gain of each feature in the feature set around the cover is calculated, and the specific formula is:

[0064]

[0065] Where H(D) is the entropy of the feature set around the cover, H(D v ) is the conditional entropy based on feature A, and the penalty term is used to reduce the influence of unimportant features.

[0066] Where the entropy of the feature set and the conditional entropy based on feature A are involved, and a penalty term is introduced, which aims to effectively reduce the interference and influence of unimportant features on the result. Finally, the top N features are selected according to the calculation results, and N is a positive integer, so as to provide accurate and key data support for the subsequent analysis of the completeness of the locking operation. N can be selected by repeated experiments.

[0067] Based on the top N features with the greatest influence, the completeness of the locking operation of all can bodies corresponding to the infrared thermal image is judged by a deep learning model.

[0068] The top N features are arranged in order according to the specific numerical value, thereby generating an N-feature set.

[0069] A feature set M is randomly selected from the N-feature set, and the last feature value and the first feature value in the set are clearly labeled. Then, values are randomly selected in the interval range of the last feature value and the first feature value, and these values are used to reasonably segment the feature set and label the segmentation values.

[0070] According to the above method, the median of the feature set M is iterated, and if the number of elements in the feature set M is odd, the median is not processed. Until all features in the feature set M are labeled, the average of all segmentation values is selected, the discrete threshold is set, and the corresponding discrete value of each feature value in the feature set M is obtained;

[0071] The discrete value of each feature value Mi in the feature set M is taken as the numerator, and the difference between the maximum feature value and the minimum feature value is taken as the denominator, and the value is recorded as Zi. The value of the integrity is set to 1-Zi.

[0072] The integrity threshold of the lock cover operation is set.

[0073] The integrity of the lock cover operation is calculated, and if the integrity of the lock cover operation meets the condition, it indicates that the current tank is intact. Thus, the comprehensive evaluation and judgment of the tank closure are completed.

[0074] If the tank closure does not meet the set standard, it is determined that the tank is not completely packaged, and the corresponding filled product is marked, so as to ensure the product packaging quality and protect the safety and stability of the product in the storage, transportation and sales process.

[0075] On the other hand, as shown in Figure 2 The present application provides a filled product analysis and quality inspection system, which mainly covers three key modules:

[0076] The core function of the data acquisition module is to accurately acquire the data information detected by the millimeter wave radar for the folded tail of the tank to be filled, and to efficiently acquire the infrared thermal image of the tank after the lock cover is completed;

[0077] The data processing module can finely segment the millimeter wave radar detection data according to the loss value, calculate the hollow rate of the tank folded tail, and accurately calculate the tank closure of the filled product.

[0078] The control module undertakes an extremely important task, that is, to mark those tanks that are not completely packaged, so as to ensure that the quality and quality of the entire filled product production line meet the high standard requirements. The tanks that are not completely packaged include the tanks with incomplete folded tail packaging and the tanks with incomplete lock cover packaging.

[0079] On the other hand, the present application also provides a computer readable storage medium, which stores the above filled product analysis and quality inspection system, wherein the filled product analysis and quality inspection system executes the above filled product analysis and quality inspection method. The computer readable medium includes permanent and non-permanent, removable and non-removable media, which can realize the storage of the filled product analysis and quality inspection system by any method or technology.

[0080] In another aspect, the present application provides a filling product analysis and quality inspection device, comprising a memory, a processor, and a filling product analysis and quality inspection system stored on the memory and executable on the processor, and the processor executes a filling product analysis and quality inspection method through the filling product analysis and quality inspection system. The filling product analysis and quality inspection method provided by the present application must be run with the filling product analysis and quality inspection system as a software platform, and the filling product analysis and quality inspection system must rely on the filling product analysis and quality inspection device hardware platform provided by the present application to run.

[0081] Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art can make modifications to the technical solutions recorded in the foregoing embodiments, or make equivalent replacements to part of the technical features, and any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for analyzing and inspecting the quality of filled products, characterized in that, Including: Millimeter-wave radar is used to detect the folded ends of the cans to be filled. The millimeter-wave radar detection data is segmented based on the loss value. The void rate of the folded ends is calculated based on the segmentation results. Cans with incomplete folded end sealing are marked. After filling the can with the folded tail completely sealed, the can is capped and an infrared thermal image of the capped can is obtained. The integrity of the capping operation in the infrared thermal image of the can is analyzed by a deep learning model. Based on the integrity of the locking operation, cans with incomplete locking seals are marked; The specific method for calculating the void ratio of the tank tail based on the segmentation results is as follows: Calculate the gradient change threshold of each segment of millimeter-wave radar detection data, and multiply each segment of millimeter-wave radar detection data by the corresponding loss value ratio to obtain the void ratio of the tank tail. Set a threshold for the porosity rate at the tail end; The void ratio at the fold end of the can is calculated. If the void ratio threshold condition is met, it indicates that the current fold end of the can is well sealed.

2. The method for analyzing and inspecting filled products according to claim 1, characterized in that, The specific method for using millimeter-wave radar to detect the tail of the can to be filled, and segmenting the millimeter-wave radar detection data based on the loss value, is as follows: Acquire echo data from millimeter-wave radar, including range information, and record it in a time-series manner; The specific formula for calculating the loss value of radar signal echo is as follows: ， in The distance from the radar to the tank's tail bend. This is a reference value for the distance from the radar to the tank's tail. It is the radar wave transmission path loss index; A sliding window is created in the echo data of the millimeter-wave radar. The millimeter-wave radar echo data is divided into multiple windows using the sliding window, and the average value of the loss value is calculated in each window. Clustering algorithms are used to classify the loss values, and loss value labels are assigned to the echo data of each type of millimeter-wave radar, thus completing the segmentation of millimeter-wave radar detection data.

3. The method for analyzing and inspecting filled products according to claim 1, characterized in that, The specific method for obtaining the infrared thermal image of the can after the lid is locked, and analyzing the integrity of the locking operation in the infrared thermal image of the can using a deep learning model, is as follows: Obtain an infrared thermal image of the can after the lid is locked, extract the position of the lid in the infrared thermal image of the can using Harris corner detection, and extract the feature set around the lid. The information gain of each feature in the feature set surrounding the cover is calculated using the following formula: ， in, It is the entropy of the feature set surrounding the cover. It is based on the conditional entropy of feature A, which reduces unimportant features through a penalty term and selects the top N features, where N is a positive integer; Based on the first N features, a deep learning model is used to determine the integrity of the locking operation corresponding to the infrared thermal images of all tanks.

4. The method for analyzing and inspecting filled products according to claim 3, characterized in that, The specific method for determining the integrity of the locking operation corresponding to the infrared thermal images of all tanks using a deep learning model is as follows: Arrange each of the first N features according to its specific value to generate a set of N features; Randomly select a feature set ,Will The last feature value and the first feature value are labeled, and a value between the last feature value and the first feature value is randomly selected to pair the feature set. Perform segmentation and label the segmentation values; The process iterates towards the median position in the feature set M using the method described above. If the number of elements in feature set M is odd, the median is not processed until the feature set is complete. After all features are labeled, the average of all segmentation values ​​is selected, and a discrete threshold is set for the feature set. All feature values ​​are used to obtain their corresponding discrete values; feature set Each eigenvalue The discrete values ​​are used as the numerator, and the difference between the largest and smallest eigenvalues ​​is used as the denominator. The values ​​are recorded as follows: Set the integrity value of the lock cover operation to 1. ; Set the integrity threshold for the cover lock operation; The integrity of the locking operation is calculated. If the integrity of the locking operation meets the conditions, it indicates that the tank is currently intact.

5. A quality inspection system for analyzed filled products, characterized in that, Including: The data acquisition module acquires the detection data of the folded end of the can to be filled by millimeter-wave radar, and acquires the infrared thermal image of the can after the cap is locked. The data processing module segments the millimeter-wave radar detection data based on the loss value, calculates the void ratio of the tank tail, and analyzes the integrity of the locking operation corresponding to the infrared thermal image of the tank. The control module marks tanks that are not fully sealed; The specific method for calculating the void ratio of the tank tail based on the segmentation results is as follows: Calculate the gradient change threshold of each segment of millimeter-wave radar detection data, and multiply each segment of millimeter-wave radar detection data by the corresponding loss value ratio to obtain the void ratio of the tank tail. Set a threshold for the porosity rate at the tail end; The void ratio at the fold end of the can is calculated. If the void ratio threshold condition is met, it indicates that the current fold end of the can is well sealed.

6. A computer-readable storage medium, characterized in that, The system described in claim 5 is stored.

7. A quality inspection device for analyzing filled products, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the filling product analysis and quality inspection method according to any one of claims 1 to 4 through the computer program.

Images