A quality tracing method based on AI image recognition

Abstract

The present invention relates to the technical field of quality tracing. The present invention relates to a quality tracing method based on AI image recognition. It comprises the following steps: S1, obtaining the total sub-production process of the product, and simultaneously obtaining the working parameters of the production equipment corresponding to each sub-production process; S2, extracting the production records of historical products, extracting the initial image data and processed image data of the product in each sub-production process from the production records, and then using AI to extract processing features; the present invention constructs a data set covering all links of production by collecting the working parameters of the total sub-production process of the product and the corresponding equipment, combining the initial image and processed image data of each sub-process, realizing full-chain traceability from raw material input to finished product output, avoiding the one-sidedness of single-dimensional data, and being able to quickly locate the specific sub-process corresponding to the quality problem. By comparing the initial image and the processed image of each sub-process, the quality feature changes of each link can be accurately located.

Description

Technical Field

[0001] The present invention relates to the field of quality tracing technology, and in particular to a quality tracing method based on AI image recognition. Background Art

[0002] In the manufacturing industry, product quality control and traceability are core links to ensuring production efficiency, reducing costs, and enhancing customer trust. Existing technologies primarily implement quality control through manual spot checks, equipment parameter monitoring, and traditional image recognition technology. For example, traditional solutions set up quality inspection points at key processes, using manual visual inspection or fixed-rule image inspection to identify product surface defects, while also recording equipment operating parameters for process traceability.

[0003] However, manual sampling relies on the experience of quality inspectors and is subject to strong subjectivity, low efficiency, and a high rate of missed detection. It is particularly difficult to identify small size changes or complex texture defects. At the same time, traditional image recognition technology is based on edge detection or template matching, which is not adaptable to feature changes in multi-variety and small-batch production scenarios, and cannot dynamically capture subtle processing feature differences under different process parameters. Therefore, a quality traceability method based on AI image recognition is proposed. Summary of the Invention

[0004] The purpose of the present invention is to provide a quality tracing method based on AI image recognition to solve the problems raised in the above background technology.

[0005] To achieve the above objectives, a quality tracing method based on AI image recognition is provided, which includes the following steps:

[0006] S1. Obtain the total sub-production process of the product and the working parameters of the production equipment corresponding to each sub-production process;

[0007] S2. Extract historical product production records, extract the initial image data and processed image data of the product in each sub-production process from the production records, and then use AI to extract processing features;

[0008] S3. Classify historical products according to test results and working parameters. Then, extract the processing feature library corresponding to each sub-production process based on the classified historical products. Combine the processing feature libraries with different test results to perform risk processing feature analysis on each sub-production process.

[0009] S4. Extract the real-time processing features of the real-time product and, before entering the next sub-production process, perform a processing feature increase prediction based on the real-time processing features combined with the working parameters corresponding to the sub-production process;

[0010] S5. Compare the predicted processing features with the risk processing features corresponding to the sub-production process. When the predicted processing features are the same as the risk processing features, stop the sub-production process of the real-time product and remind the production end.

[0011] As a further improvement of the present technical solution, the step S1 obtains the total process required for product production through the product production end, and divides the total process into multiple sub-production processes according to the stages of the process; and sets an image acquisition device at the starting position and the end position of each sub-production process;

[0012] Get the production equipment corresponding to each sub-production process.

[0013] As a further improvement of this technical solution, the steps of S2 are as follows:

[0014] S2.1. On the production side, save production records for products after production testing, then obtain the current product type based on the working parameters, and extract historical products and production records of the same product type;

[0015] S2.2. Extract the initial image data and processed image data of the product in each sub-production process from the production records extracted in S2.1. Then, use AI to extract processing features based on the initial image data and the processed image data. Based on the extraction results, obtain the processing features added to the product after passing through each sub-production process.

[0016] As a further improvement of the present technical solution, S2.2 accumulates the processing features added to the product in each sub-production process during the calculation process until the last sub-production process is completed, thereby obtaining all the processing features corresponding to the product.

[0017] As a further improvement of the present technical solution, the initial image data in S2 is the image data of the product before entering the sub-production process, and the processed image data is the image data after passing through the sub-production process;

[0018] When a product needs to enter a sub-production process, the processed image data of the previous sub-production process is used as the initial image data until all sub-production processes are completed.

[0019] As a further improvement of this technical solution, the steps of S3 are as follows:

[0020] S3.1. Extract the test results of each historical product, classify them according to the test results, and obtain the historical products for each test result. At the same time, classify the operating parameters corresponding to the historical products and obtain the historical products corresponding to each operating parameter.

[0021] The test results are divided into qualified test results and unqualified test results;

[0022] S3.2. Extract processing features from historical products. Then, combine these features with the operating parameters and sub-production processes to create a processing feature library. The library contains the processing features added for different products under the different operating parameters of each sub-production process.

[0023] S3.3. Conduct risk analysis on the processing features in the processing feature library based on the unqualified test results to obtain risk processing features corresponding to different working parameters of each sub-production process.

[0024] As a further improvement of this technical solution, the steps of performing risk feature analysis in S3.3 are as follows:

[0025] S3.3.1. Analyze the reasons for failure of each historical product with unqualified test results and obtain the risk sub-production process corresponding to the failure reason;

[0026] S3.3.2. Obtain the sum of the added processing features of historical products after they have undergone the risk sub-production process, then extract the processing parameters corresponding to the risk sub-production process, and use the sum of the added processing features as the risk processing features of the processing parameters corresponding to the sub-production process.

[0027] As a further improvement of this technical solution, the steps of S4 are as follows:

[0028] S4.1. Extract real-time processing features of real-time products;

[0029] S4.2. Before the real-time product enters the next sub-production process, the real-time processing features corresponding to the real-time product are combined with the working parameters corresponding to the next sub-production process to perform a processing feature addition prediction, and obtain the predicted processing features added after the real-time product enters the next sub-production process.

[0030] As a further improvement of this technical solution, the steps of S5 are as follows:

[0031] S5.1. Compare the predicted processing characteristics with the risk processing characteristics corresponding to the sub-production process;

[0032] S5.2: If the predicted processing characteristics are the same as the risk processing characteristics, the sub-production process of the real-time product is stopped and the production end is notified. After the production end adjusts the working parameters, S4.2 is repeated to regenerate the predicted processing characteristics.

[0033] S5.3. When the predicted processing characteristics are different from the risk processing characteristics, the sub-production process will be processed normally.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] 1. This quality traceability method based on AI image recognition collects the operating parameters of the product's overall production process and the corresponding equipment, combines the initial images and processed image data of each sub-process, and constructs a data set covering all production links. This enables full-chain traceability from raw material input to finished product output, avoids the one-sidedness of single-dimensional data, and can quickly locate the specific sub-processes corresponding to quality issues. By comparing the initial image and processed image of each sub-process, the quality feature changes of each link can be accurately located.

[0036] 2. This quality traceability method based on AI image recognition predicts the processing feature increment based on the real-time product processing features and the working parameters of the next sub-process through a prediction model, and simulates the processing effect in advance before the product enters the next process, so as to avoid the flow of unqualified products into subsequent processes and reduce the waste of materials and working hours. By comparing the predicted features with the risk feature library, high-risk production paths are blocked in real time, and reminders are given to adjust the working parameters, reducing the reliance on manual inspections, reducing the quality risks caused by human omissions, and improving the level of production automation.

[0037] 3. In this quality traceability method based on AI image recognition, historical products are doubly classified according to test results and working parameters, and correlation relationships are established to distinguish the differences in processing characteristics between qualified and unqualified products, identify key parameters affecting quality, and customize quality control standards for different product types. At the same time, a risk feature library is incrementally constructed through the processing features of unqualified products, and risk parameters associated with the causes of unqualified products in each sub-process are identified, providing data support for process optimization, such as adjusting equipment parameters to avoid the generation of risk features and reduce the unqualified rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is the overall flow chart of the present invention;

[0039] Figure 2 A flowchart for extracting historical product and production records of the same product type for the present invention;

[0040] Figure 3 A flowchart of the process of obtaining historical products corresponding to each working parameter for the present invention;

[0041] Figure 4 A flowchart of extracting real-time processing features of real-time products according to the present invention;

[0042] Figure 5 This is a flowchart of the present invention for comparing predicted processing features with risk processing features corresponding to sub-production processes. DETAILED DESCRIPTION

[0043] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0044] See also Figure 1-Figure 5 As shown, the purpose of this embodiment is to provide a quality tracing method based on AI image recognition, including the following steps:

[0045] S1. Obtain the total sub-production process of the product and the working parameters of the production equipment corresponding to each sub-production process;

[0046] S1 obtains the total process required for product production through the product production end, and divides the total process into multiple sub-production processes according to the stages of the process; and sets image acquisition devices at the starting position and end position of each sub-production process;

[0047] Get the production equipment corresponding to each sub-production process.

[0048] S2. Extract historical product production records, extract the initial image data and processed image data of the product in each sub-production process from the production records, and then use AI to extract processing features;

[0049] The steps of S2 are as follows:

[0050] S2.1. On the production side, save production records for products after production testing, then obtain the current product type based on the working parameters, and extract historical products and production records of the same product type;

[0051] During the production process, after each product undergoes production testing, relevant production records are saved. These records include image data, working parameters, test results, and other information for each sub-production process. The initial image data (images of the product before entering the sub-production process) and processed image data (images of the product after passing through the sub-production process) of the product in each sub-production process are then extracted from the historical production records.

[0052] The initial image data in S2 is the image data of the product before entering the sub-production process, and the processed image data is the image data after passing through the sub-production process;

[0053] S2.2. Extract the initial image data and processed image data of the product in each sub-production process from the production records extracted in S2.1. Then, use AI to extract processing features based on the initial image data and the processed image data. Based on the extraction results, obtain the processing features added to the product after passing through each sub-production process.

[0054] S2.2 During the calculation process, the processing features added to the product in each sub-production process are accumulated until the last sub-production process is completed, thereby obtaining all the processing features corresponding to the product.

[0055] The initial image data is combined with the processed image data, and AI is used for feature extraction. The AI ​​model automatically identifies key features of the product during processing (such as size changes, shape changes, color changes, etc.) based on the input image data. With each sub-production process, the product will gain some new processing features. Therefore, it is necessary to accumulate the processing features of each process until a complete processing feature set is finally obtained;

[0056] When a product needs to enter a sub-production process, the processed image data of the previous sub-production process is used as the initial image data until all sub-production processes are completed. The formula is as follows:

[0057] ;

[0058] Among them, F processed (t) is the extracted processing feature of the t-th sub-production process, AI model For AI models, I init (t) is the initial image data of the t-th sub-production process, I processed (t) is the processed image data of the t-th sub-production process;

[0059] ;

[0060] Among them, F total It is the sum of the final processing characteristics of the product after going through all sub-production processes, and n is the total number of sub-production processes that the product goes through.

[0061] S3. Classify historical products according to test results and working parameters. Then, extract the processing feature library corresponding to each sub-production process based on the classified historical products. Combine the processing feature libraries with different test results to perform risk processing feature analysis on each sub-production process.

[0062] The steps for S3 are as follows:

[0063] S3.1. Extract the test results of each historical product, classify them according to the test results, and obtain the historical products for each test result. At the same time, classify the operating parameters corresponding to the historical products and obtain the historical products corresponding to each operating parameter.

[0064] The test results are divided into qualified test results and unqualified test results. The specific steps are as follows:

[0065] Extracting the test results of each historical product: Each product undergoes quality inspection during the production process, and generates a test result. The test result can be qualified or unqualified (for example, dimensional discrepancy, surface defects, etc.). Then, the test results of each product are extracted from the historical production records and classified into qualified and unqualified test results.

[0066] Classify by test results: All historical product test results are classified according to their qualification. For qualified products, their historical records are stored; for unqualified products, their historical records are also stored, and the specific reasons for failure can be further marked;

[0067] Obtain historical products for each test result: Extract historical records of qualified and unqualified products based on the classification results. This helps in subsequent analysis of similarities and differences in the production processes of qualified and unqualified products.

[0068] Classification by working parameters: historical products can be classified according to working parameters. Qualified products and unqualified products can be classified according to different working parameters to view the impact of different working parameters on product quality.

[0069] Obtain historical product records corresponding to each operating parameter: Under each test result category, extract the corresponding historical product records based on the operating parameter classification. This can help production managers understand whether specific operating parameters are directly related to the qualification or failure of products.

[0070] S3.2. Extract processing features from historical products (using the same extraction method as above). Then, combine these processing features with the working parameters and sub-production processes to create a processing feature library. The processing feature library contains the processing features added to different products under the different working parameters of each sub-production process. The specific steps are as follows:

[0071] According to the processing characteristics, working parameters and sub-production processes of each product, a processing feature library is established. This library will contain the corresponding processing features in each production process, and record different processing features according to different working parameters;

[0072] The relationship between processing features and working parameters is classified and associated in the feature library, and a connection is established between the different working parameters under each sub-production process and its corresponding processing features.

[0073] S3.3. Conduct risk analysis on the processing features in the processing feature library based on the unqualified test results to obtain risk processing features corresponding to different working parameters of each sub-production process.

[0074] S3.3 The steps for risk characterization are as follows:

[0075] S3.3.1. Analyze the reasons for failure of each historical product with unqualified test results and obtain the risk sub-production process corresponding to the failure reason;

[0076] S3.3.2. Obtain the sum of the added processing characteristics of historical products after undergoing the risk sub-production process, then extract the processing parameters corresponding to the risk sub-production process, and use the sum of the added processing characteristics as the risk processing characteristics of the processing parameters corresponding to the sub-production process;

[0077] First, it is necessary to determine the reasons for failure based on the test results of each historical product. Each historical product test result will lead to a failure reason, and each failure reason corresponds to one or more risk sub-production processes. The mapping relationship between the failure reason and the risk sub-production process can be determined through historical data or empirical rules. After going through the risk sub-production process, the processing characteristics of the historical product will change. The change in processing characteristics can be calculated by comparing its processing characteristics before and after the risk sub-production process. The specific formula is as follows:

[0078] ;

[0079] in, To represent the historical product i undergoing the risk sub-production process P r The subsequent machining feature increment, For the risk sub-production process P r After processing features, is the initial processing characteristics of historical product i before the risk sub-production process;

[0080] ;

[0081] Among them, F r is the risk processing characteristic, P r The risk sub-production process is the step in product production that may cause risk or non-conformity. s is the set of risk sub-production steps, which is the set of all possible risk sub-production processes, W ris the processing parameter, i.e., various processing control parameters in the risk sub-production process, X s is a set of processing parameters, is a set of machining feature increments, representing all possible machining feature increments.

[0082] S4. Extract the real-time processing features of the real-time product and, before entering the next sub-production process, perform a processing feature increase prediction based on the real-time processing features combined with the working parameters corresponding to the sub-production process;

[0083] The steps for S4 are as follows:

[0084] S4.1. Extract real-time processing features of real-time products;

[0085] Before the product enters the next sub-production process, the processing features of the current real-time product are first extracted. This feature contains the status reached by the current product during the production process;

[0086] S4.2. Before the real-time product enters the next sub-production process, the real-time processing features corresponding to the real-time product are combined with the working parameters corresponding to the next sub-production process to predict the processing features to be added. This method obtains the predicted processing features that will be added to the real-time product after it enters the next sub-production process.

[0087] Combine the real-time product processing features with the working parameters of the next sub-production process to form a new processing prediction model input. Through models (such as regression models, machine learning models, etc.), based on the existing real-time processing features and the working parameters of the next sub-production process, predict the addition of processing features. According to the prediction results, obtain the processing feature increments that the product should have after entering the next sub-production process.

[0088] S5. Compare the predicted processing features with the risk processing features corresponding to the sub-production process. When the predicted processing features are the same as the risk processing features, stop the sub-production process of the real-time product and remind the production end.

[0089] The steps for S5 are as follows:

[0090] S5.1. Compare the predicted processing characteristics with the risk processing characteristics corresponding to the sub-production process;

[0091] S5.2: If the predicted processing characteristics are the same as the risk processing characteristics, the sub-production process of the real-time product is stopped and the production end is notified. After the production end adjusts the working parameters, S4.2 is repeated to regenerate the predicted processing characteristics.

[0092] S5.3. When the predicted processing characteristics are different from the risk processing characteristics, the sub-production process will be processed normally.

[0093] 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 to the above embodiments. The above embodiments and descriptions are merely preferred examples of the present invention and are not intended to limit the present invention. Various changes and improvements may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and improvements fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A quality tracing method based on AI image recognition, characterized by: The following steps are involved: S1. Obtain the total sub-production process of the product and the working parameters of the production equipment corresponding to each sub-production process; S2. Extract historical product production records, extract the initial image data and processed image data of the product in each sub-production process from the production records, and then use AI to extract processing features; The steps of S2 are as follows: S2.

1. On the production side, save production records for products after production testing, then obtain the current product type based on the working parameters, and extract historical products and production records of the same product type; S2.

2. Extract the initial image data and processed image data of the product in each sub-production process from the production records extracted in S2.

1. Then, use AI to extract processing features from the initial image data combined with the processed image data. Based on the extraction results, obtain the processing features added to the product as it passes through each sub-production process. During the calculation process of S2.2, the processing features added to the product in each sub-production process are accumulated until the last sub-production process is completed, thereby obtaining all the processing features corresponding to the product; S3. Classify historical products according to test results and working parameters. Then, extract the processing feature library corresponding to each sub-production process based on the classified historical products. Combine the processing feature libraries with different test results to perform risk processing feature analysis on each sub-production process. The steps of S3 are as follows: S3.

1. Extract the test results of each historical product, classify them according to the test results, and obtain the historical products for each test result. At the same time, classify the operating parameters corresponding to the historical products and obtain the historical products corresponding to each operating parameter. The test results are divided into qualified test results and unqualified test results; S3.

2. Extract processing features from historical products. Then, combine these features with the operating parameters and sub-production processes to create a processing feature library. The library contains the processing features added for different products under the different operating parameters of each sub-production process. S3.

3. Conduct risk analysis on the processing features in the processing feature library based on the unqualified test results to obtain risk processing features corresponding to different working parameters for each sub-production process; The steps for risk characterization analysis in S3.3 are as follows: S3.3.

1. Analyze the reasons for failure of each historical product with unqualified test results and obtain the risk sub-production process corresponding to the failure reason; S3.3.

2. Obtain the sum of the added processing characteristics of historical products after undergoing the risk sub-production process, then extract the processing parameters corresponding to the risk sub-production process, and use the sum of the added processing characteristics as the risk processing characteristics of the processing parameters corresponding to the sub-production process; S4. Extract the real-time processing features of the real-time product and, before entering the next sub-production process, perform a processing feature increase prediction based on the real-time processing features combined with the working parameters corresponding to the sub-production process; S5. Compare the predicted processing features with the risk processing features corresponding to the sub-production process. When the predicted processing features are the same as the risk processing features, stop the sub-production process of the real-time product and remind the production end.

2. The quality tracing method based on AI image recognition according to claim 1, characterized in that: The S1 obtains the total process required for product production through the product production end, and divides the total process into multiple sub-production processes according to the stages of the process; An image acquisition device is set at the starting position and the ending position of each sub-production process; Get the production equipment corresponding to each sub-production process.

3. The quality tracing method based on AI image recognition according to claim 1, characterized in that: The initial image data in S2 is the image data of the product before entering the sub-production process, and the processed image data is the image data after passing through the sub-production process; When a product needs to enter a sub-production process, the processed image data of the previous sub-production process is used as the initial image data until all sub-production processes are completed.

4. The quality tracing method based on AI image recognition according to claim 1, characterized in that: The steps of S4 are as follows: S4.

1. Extract real-time processing features of real-time products; S4.

2. Before the real-time product enters the next sub-production process, the real-time processing features corresponding to the real-time product are combined with the working parameters corresponding to the next sub-production process to perform a processing feature addition prediction, and obtain the predicted processing features added after the real-time product enters the next sub-production process.

5. The quality tracing method based on AI image recognition according to claim 1, characterized in that: The steps of S5 are as follows: S5.

1. Compare the predicted processing characteristics with the risk processing characteristics corresponding to the sub-production process; S5.2: If the predicted processing characteristics are the same as the risk processing characteristics, the sub-production process of the real-time product is stopped and the production end is notified. After the production end adjusts the working parameters, S4.2 is repeated to regenerate the predicted processing characteristics. S5.

3. When the predicted processing characteristics are different from the risk processing characteristics, the sub-production process will be processed normally.

Images