Quality data acquisition and anomaly prediction method and system in processing process

Abstract

The invention discloses a quality data acquisition and anomaly prediction method and system in a processing process, and relates to the technical field of data processing. Comprising the steps of historical data collection, data processing, feature extraction, collection area distinguishing, partition calculation based on unqualified information and features of a target product in a reference data set, and distinguishing of a first area and a second area; real-time data acquisition: based on the first area and the second area, performing accurate acquisition and fuzzy acquisition on the target product; performing real-time data comparison and variable data induction, and collecting variable data of a target product to form a variable data set; and time prediction: predicting the time when the quality of the target product falls to the qualified line. According to the method, the collection areas of the target product can be distinguished, the real-time performance and accuracy of the collected data can be considered through targeted sampling adjustment, and meanwhile, the time when the product quality falls to the qualified line can be predicted, so that an enterprise can find and process potential product quality problems in time.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and specifically provides a method and system for collecting quality data and predicting anomalies during the processing Background Art

[0002] In industrial production, quality control is a multi-dimensional and systematic project. It starts with the strict screening and inspection of raw materials to ensure that all basic materials meet the established quality standards and requirements. Then, it delves into each key link of the production process. By introducing advanced production technologies and equipment, combined with precision monitoring instruments and strict process control, it ensures that each process operation can reach the established quality standards, thereby effectively reducing variations and errors in the production process.

[0003] A method and system for predicting abnormal operation of industrial production equipment with the application publication number CN110879971A includes the following steps: Step 1: Collect equipment-related data; Step 2: Preprocess the collected related data; Step 3: Perform sample annotation on the preprocessed data; Step 4: Extract features from the data after sample annotation to form feature set data; Step 5: Input the feature set data into the Gaussian mixture model algorithm for prediction model training to obtain a prediction result; Step 6: Analyze the obtained prediction result by experts; the analysis result that conforms to the actual situation is normally output, and at the same time, enterprise production users are reminded; the analysis result that does not conform to the actual situation is recommended by experts and fed back to optimize the model algorithm.

[0004] In the existing industrial sheet production process, with the continuous progress of manufacturing technology and the increasing expansion of production scale, quality control in the factory production process has become increasingly important. Most existing industrial sheet quality detection methods collect production quality data such as appearance defects, dimensions, spacing, and holes of industrial sheets in real time by installing sensors on the production line. When existing sensors collect quality data of industrial sheets, there are a large number of production quality data such as appearance defects, dimensions, spacing, and solder joints on the industrial sheets, resulting in an excessive amount of data during collection and transmission, making it difficult to transmit immediately. Rapidly collecting the quality data of industrial sheets can achieve rapid transmission, but it is difficult to balance the real-time and accuracy of the collected data. Existing detection systems cannot capture abnormal changes in the production process in real time, resulting in difficulties in timely discovery and handling of product quality problems. Summary of the Invention

[0005] The purpose of the present invention is to provide a method and system for collecting quality data and predicting anomalies during the processing to solve the problems raised in the above background art.

[0006] To achieve the above object, the present invention provides the following technical solutions: A method and system for collecting quality data and predicting anomalies during the processing, the method comprising:

[0007] Collecting historical data, collecting the quality data of the target product and the operation data of the production equipment;

[0008] Data processing, preprocessing based on the collected historical data, and summarizing the preprocessed data to form a reference data set;

[0009] Extracting features, extracting the unqualified information and features of the target product based on the reference data set to form a feature data set;

[0010] Distinguishing the collection area, performing partition calculation based on the unqualified information and features of the target product in the reference data set, and distinguishing the first area and the second area;

[0011] Collecting real-time data, based on the first area and the second area, performing precise collection and fuzzy collection on the target product;

[0012] Comparing real-time data, comparing the real-time data with the preset quality detection standard, and obtaining variable data according to the data change direction of the target product;

[0013] Inducing variable data, collecting the variable data of the target product to form a variable data set;

[0014] Prediction time, extracting the unqualified information features of the target product based on the variable data set, and predicting the time when the quality of the target product drops below the qualified line, the steps of the prediction method comprising:

[0015] S1: Analyzing the variable data set to obtain the target variable, and extracting the information feature variables of the unqualified target product, the calculation formula is as follows:

[0016]

[0017] Where: L is the observed value of the i-th feature in the original data set, Li is the sample average value of the i-th feature, σi is the standard deviation of the i-th feature L, and Xi is the observed value of the standardized i-th feature;

[0018] S2: Calculating the product quality index Y based on the information features of the collected unqualified target product, and the calculation data is as follows:

[0019]

[0020] Where: Y is the target variable, β0 is the intercept term, β1, β2, β3, β4 are all regression coefficients, X1, X2, X3, X4 are all independent variables, and ϵ is the error term;

[0021] S3: Predict the time when the product quality drops below the qualified line based on the product quality indicators. The prediction formula is as follows:

[0022]

[0023] Where: t is the time required from the current time point until the quality indicator is expected to drop below the qualified line, Q is the qualified line, ΔY is the average rate of change of the quality indicator over time, and Y is the quality indicator value at the current time point.

[0024] Furthermore, for the data processing, clean the collected historical data to ensure data accuracy and consistency. Merge and organize the processed data, and classify and archive it according to time sequence, product type, equipment type, etc. to form a unified reference data set.

[0025] Furthermore, the calculation formula for extracting features is as follows:

[0026]

[0027] Where: H² is the square of the original H statistic, X 不合格 and X 合格 are the average values of defective products and qualified products on feature X respectively, S² 不合格 and S² 合格 are the variances of defective products and qualified products on feature S respectively, N 不合格 and N 合格 are the sample sizes of defective products and qualified products respectively.

[0028] Furthermore, for the collection area differentiation, calculate the defect frequency, defect severity, and area of the target product based on the reference data set. The calculation formula for the area score is as follows:

[0029]

[0030] Where: i is the area index, Fi is the defect frequency in the i-th area, Si is the defect severity in the i-th area, Ai is the area of the i-th area, and Score(i) is the score of the i-th area.

[0031] Furthermore, for the collection area differentiation, differentiate the defect frequency, defect severity, and area of the target product based on the reference data set. The calculation formula for the area differentiation is as follows:

[0032]

[0033] Where: Score(i) is the score of the i-th region, Threshold is the score boundary for screening the first region, and R is the set of indices of all first regions. When the score of the target region is greater than or equal to the score boundary, it is the first region; when the score of the target region is lower than the score boundary, it is the second region.

[0034] Further, the real-time data collection includes the following steps:

[0035] M1: Identify the first region and the second region in the image;

[0036] M2: Increase the sampling frequency within the determined first region to ensure obtaining high-quality data and improving the data accuracy rate, and decrease the sampling frequency within the determined second region to reduce the amount of collected data;

[0037] M3: Perform lossless compression on the collected data through a compression program to achieve fast transmission, taking into account the real-time nature and accuracy of the collected data.

[0038] Further, for the real-time data comparison, compare the real-time data with the quality inspection standard, obtain variable data according to the data change direction of the target product, and supplement the variable data of the unqualified target product to the feature dataset. The steps of the prediction method are as follows:

[0039] S1: Analyze the variable dataset to obtain the target variable, and extract the information feature variables of the unqualified target product. The calculation formula is as follows:

[0040]

[0041] Where: L is the observed value of the i-th feature in the original dataset, Li is the sample average of the i-th feature, that is, the average of the values of all samples on the feature Li, σi is the standard deviation of the i-th feature L, which measures the degree of dispersion of the distribution of this feature data, and Xi is the observed value of the i-th feature after standardization;

[0042] S2: Calculate the product quality index Y based on the information features of the collected unqualified target products. The calculation data is as follows:

[0043]

[0044] Where: Y is the target variable, representing the product quality index; β0 is the intercept term, which is the expected value of Y when all independent variables Xi are 0; β1, β2, β3, β4 are all regression coefficients, respectively representing the influence degrees of the independent variables X1, X2, X3, X4 on Y; X1, X2, X3, X4 are all independent variables; ϵ is the error term, representing the part of the variation that the model fails to explain;

[0045] S3: Predict the time when the product quality drops below the qualified line based on the product quality indicators. The prediction formula is as follows:

[0046]

[0047] Where: t is the time required from the current time point to when the quality indicator is expected to drop below the qualified line, Q is the qualified line, that is, the minimum quality standard allowed for the target product, Y is the quality indicator value at the current time point, (Q - Y) is the difference between the current quality indicator and the qualified line, and ΔY is the average rate of change of the quality indicator over time.

[0048] Compared with the prior art, the beneficial effects of the present invention are:

[0049] In the quality data collection and anomaly prediction method and system during the processing, the collection area of the target product is distinguished and calculated through an algorithm, and it is divided into a first area and a second area. A differential sampling strategy is adopted. A higher sampling frequency is applied to the first area for more intensive data collection to ensure that all details that may affect product quality can be captured, thereby improving the accuracy and reliability of detection. On the contrary, for the second area, a lower sampling frequency is used to reduce unnecessary data volume, save storage space and transmission time, and at the same time will not significantly affect the overall quality assessment result. Through this targeted sampling adjustment, the real-time nature and accuracy of the collected data can be balanced;

[0050] At the same time, by comparing the real-time data with the quality detection standard, variable data is obtained according to the data change direction of the target product, the information feature variables of unqualified target products are extracted, the product quality indicator Y is calculated based on the information features of the collected unqualified target products, and the time when the product quality drops below the qualified line is predicted, enabling enterprises to timely discover and handle potential product quality problems and avoid the generation of defective products. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is a schematic diagram of the principle structure of the quality detection standard of the present invention;

[0052] Figure 2 It is a schematic diagram of the principle structure of the real-time data optimization of the present invention;

[0053] Figure 3 It is a schematic diagram of the principle structure of the collection area differentiation of the present invention;

[0054] Figure 4 It is a schematic diagram of the optimized structure of the feature data set of the present invention;

[0055] Figure 5 It is a schematic diagram of the principle structure of the variable data induction of the present invention;

[0056] Figure 6 Structural schematic diagram of the time prediction method of the present invention. Specific implementation mode

[0057] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0058] In this application, for the convenience of understanding, the method steps used do not necessarily need to be executed in the order of the steps in this embodiment during actual operation. In some other embodiments, these steps can be carried out simultaneously or in a changed order.

[0059] Embodiment 1:

[0060] As Figures 1-6 shown, the present invention provides a technical solution: a method and system for collecting quality data and predicting anomalies during the processing process, and the method includes:

[0061] Collecting historical data, collecting the quality data of the target product and the operation data of the production equipment;

[0062] It should be noted that: the quality data of the target product includes production quality data such as appearance defects, dimensions, spacing, and solder joints on industrial plates. Collecting appearance defects involves identifying various abnormal conditions on the surface of industrial plates, such as scratches, depressions, cracks, foreign objects, etc. Dimension inspection mainly includes whether the physical dimensions of the industrial plates themselves meet the design specifications. Spacing detection refers to checking the distance between each hole on the industrial plates.

[0063] Data processing, preprocessing based on the collected historical data, and summarizing the preprocessed data to form a reference data set;

[0064] Extracting features, extracting the unqualified information and features of the target product based on the reference data set to form a feature data set;

[0065] Distinguishing the collection area, performing zoning calculations based on the unqualified information and features of the target product in the reference data set, and distinguishing the first area and the second area;

[0066] Collecting real-time data, accurately collecting and fuzzily collecting the target product based on the first area and the second area;

[0067] Comparing real-time data, comparing the real-time data with the preset quality detection standard, and obtaining variable data according to the data change direction of the target product;

[0068] Variable data induction, collecting variable data of the target product to form a variable data set;

[0069] Prediction time, extracting the unqualified information features of the target product based on the variable data set, and predicting the time when the quality of the target product drops below the qualified line. The steps of the prediction method are as follows:

[0070] S1: Analyze the variable data set to obtain the target variable, and extract the information feature variables of the unqualified target product. The calculation formula is as follows:

[0071]

[0072] Where: L is the observed value of the i-th feature in the original data set, Li is the sample average value of the i-th feature, σi is the standard deviation of the i-th feature L, and Xi is the observed value of the i-th feature after standardization;

[0073] S2: Calculate the product quality index Y based on the information features of the collected unqualified target products. The calculation data is as follows:

[0074]

[0075] Where: Y is the target variable, β0 is the intercept term, β1, β2, β3, β4 are all regression coefficients, X1, X2, X3, X4 are all independent variables, and ϵ is the error term;

[0076] S3: Predict the time when the product quality drops below the qualified line based on the product quality index. The prediction formula is as follows:

[0077]

[0078] Where: t is the time required from the current time point to when the quality index is expected to drop below the qualified line, Q is the qualified line, ΔY is the average rate of change of the quality index over time, and Y is the quality index value at the current time point.

[0079] For the data processing, clean the collected historical data to ensure the accuracy and consistency of the data. Merge and organize the processed data, and classify and file it according to time sequence, product type, equipment type, etc. to form a unified reference data set.

[0080] It should be noted that:

[0081] The calculation formula for extracting features is as follows:

[0082]

[0083] Substitute the data,

[0084] The sample quantity N of unqualified products 不合格= 5N,

[0085] Sample values: 10, 12, 14, 15, 16,

[0086] Average value X 不合格 = (10 + 12 + 14 + 15 + 16) / 5 = 13.4,

[0087] S² 不合格 = [(10 - 13.4)² + (12 - 13.4)² + (14 - 13.4)² + (15 - 13.4)² + (16 - 13.4)²] / 4,

[0088] ≈ 4.3,

[0089] Sample quantity N of qualified products 合格 = 5,

[0090] Sample values: 8, 9, 10, 11, 128, 9, 10, 11, 12,

[0091] Average value X 合格 = (8 + 9 + 10 + 11 + 12) / 5,

[0092] = 10X,

[0093] S² 合格 = [(8 - 10)² + (9 - 10)² + (10 - 10)² + (11 - 10)² + (12 - 10)²] / 4,

[0094] = 2,

[0095] Substitute the known values:

[0096] X 不合格 = 13.4,

[0097] X 合格 = 10,

[0098] S 不合格 2≈ 4.3,

[0099] S 合格 = 2,

[0100] N 不合格 = 5,

[0101] N 合格 = 5,

[0102] (X 不合格 −X 合格 )² = (13.4 - 10)²,

[0103] = 3.4²,

[0104] = 11.56,

[0105] S² 不合格 / N 不合格 +S² 合格 / N 合格 =4.3 / 5 + 2 / 5,

[0106] =0.86 + 0.4,

[0107] =1.26,

[0108] Therefore,

[0109] H² = 11.56 / 1.26,

[0110] ≈9.17,

[0111] where: H² is the square of the original H statistic, X 不合格 and X 合格 are the average values of defective and non - defective products on feature X respectively, S² 不合格 and S² 合格 are the variances of defective and non - defective products on feature S respectively, N 不合格 and N 合格 are the sample sizes of defective and non - defective products respectively.

[0112] The acquisition area is distinguished by calculating the defect frequency, defect severity and area of the target product based on the reference data set. The formula for calculating the area score is as follows:

[0113]

[0114] Substitute the data,

[0115] Area 1:

[0116] Defect frequency F1 = 4 (the number of defects found in this area),

[0117] Defect severity S1 = 7 (this is a relative value representing the average severity of all defects),

[0118] Area A1 = 20 square units,

[0119] Area 2:

[0120] Defect frequency F2 = 2,

[0121] Defect severity S2 = 5,

[0122] Area A2 = 15 square units,

[0123] Area 3:

[0124] The defect frequency F3 = 6,

[0125] The defect severity S3 = 8,

[0126] The area of the region A3 = 25 square units,

[0127] Calculate the score Score(i),

[0128] According to the provided formula , we can calculate the score for each region,

[0129] For region 1,

[0130] Score(1) = 28 / 20,

[0131] = 1.4,

[0132] For region 2,

[0133] Score(2) = 2 × 5 / 15,

[0134] = 10 / 15,

[0135] ≈ 0.67,

[0136] For region 3,

[0137] Score(3) = 6 × 8 / 25,

[0138] = 48 / 25,

[0139] = 1.92,

[0140] As a result,

[0141] The score of region 1 is 1.4;

[0142] The score of region 2 is 0.67;

[0143] The score of region 3 is 1.92;

[0144] Where: i is the region index, Fi is the defect frequency in the i-th region, Si is the defect severity in the i-th region, Ai is the area of the i-th region, and Score(i) is the score of the i-th region.

[0145] The said acquisition region differentiation is based on the reference data set to differentiate the defect frequency, defect severity and region area on the target product. The region differentiation calculation formula is as follows:

[0146] , where: Score(i) is the score of the i-th region, Threshold is the score boundary for screening the first region, and R is the set of all indices of the first regions. When the score of the target region is greater than or equal to the score boundary, it is the first region; when the score of the target region is lower than the score boundary, it is the second region.

[0147] Substitute the data, and the score boundary Threshold = 1.0. Based on the scores of each region calculated previously, we can determine which regions are considered "first regions".

[0148] The scores have been calculated.

[0149] The score of Region 1 is Score(1) = 1.4.

[0150] The score of Region 2 is Score(2) ≈ 0.67.

[0151] The score of Region 3 is Score(3) = 1.92.

[0152] Screen the first regions.

[0153] According to , we will compare the scores of each region with the set threshold (Threshold = 1.0):

[0154] For Region 1, Score(1) = 1.4. Since 1.4 ≥ 1.0, it is a "first region";

[0155] For Region 2, Score(2) ≈ 0.67. Since 0.67 < 1.0, it is a "second region";

[0156] For Region 3, Score(3) = 1.92. Since 1.92 ≥ 1.0, it is also a "first region";

[0157] Therefore, the index set RR of the "first regions" contains: R = {1, 3},

[0158] This means that Region 1 and Region 3 are the first regions that need special attention. Their scores are both higher than or equal to the set score boundary Threshold = 1.0. The score of Region 2 is lower than this boundary, so it is classified as the second region, indicating that its quality status is relatively good or it does not require as urgent attention as the first regions.

[0159] The real-time data acquisition described above includes the steps of:

[0160] M1: Identify the first regions and the second regions in the image;

[0161] It should be noted that, in the form of coordinates or templates, the recognition process starts with dividing or matching the newly acquired image into regions of the same type as before, and directly locates the corresponding regions by using the predefined definition information (such as position coordinates or shape templates) of the first and second regions.

[0162] M2: Increase the sampling frequency within the determined first region to ensure obtaining high-quality data and improve the accuracy of the data. Decrease the sampling frequency within the determined second region to reduce the amount of collected data.

[0163] It should be noted that when processing product images, for the determined first region (the region that requires higher attention) and the second region (the region with relatively stable quality), a differential sampling strategy can be adopted to optimize the data acquisition efficiency and quality. The specific process is as follows: First, during the image acquisition stage, identify the positions of the first region and the second region according to the predefined coordinates or templates. Then, during the subsequent data acquisition process, apply a higher sampling frequency to the first region, which means more intensive data acquisition will be carried out in these key regions to ensure that all details that may affect product quality can be captured, thereby improving the accuracy and reliability of detection. On the contrary, for the second region, a lower sampling frequency is adopted to reduce unnecessary data volume, which helps save storage space and processing time, and at the same time will not significantly affect the overall quality assessment result. Through this targeted sampling adjustment, while ensuring the monitoring requirements of key regions, the operating efficiency of the entire system can be improved.

[0164] M3: Perform lossless compression on the collected data through a compression program to achieve fast transmission, taking into account the real-time nature and accuracy of the collected data.

[0165] It should be noted that after data acquisition is completed, in order to achieve fast and reliable transmission while maintaining the integrity and accuracy of the data, lossless compression technology can be used to process the data. The specific process is as follows: First, identify and mark the key data that requires high-fidelity transmission (such as the data in the first region) to ensure that no details of this information are lost due to compression. Then, use an efficient lossless compression algorithm (such as arithmetic coding or LZW algorithm) to compress the entire data set or the data in a specific region. This step can significantly reduce the data volume without losing any information, enabling fast transmission even when the network bandwidth is limited. At the receiving end, use the corresponding decompression program to restore the original data, ensuring that the real-time nature and accuracy of the data are not affected. In this way, both the requirements for high-quality transmission of key data can be met, and the data volume can be effectively managed, improving the overall transmission efficiency.

[0166] The real-time data comparison is to compare the real-time data with the quality inspection standards, obtain variable data according to the data change direction of the target product, and supplement the variable data of the unqualified target products to the feature dataset. The steps of the prediction method are as follows:

[0167] S1: Analyze the variable dataset to obtain the target variable, and extract the information feature variables of the unqualified target products. The calculation formula is as follows:

[0168]

[0169] Substitute the data:

[0170] L1 = 10,

[0171] L2 = 12,

[0172] L3 = 14,

[0173] L4 = 15,

[0174] L5 = 16,

[0175] Calculate the sample mean Li and the standard deviation σi of each feature.

[0176] Li = 1 / n * ∑² i=1 Li,

[0177] = (10 + 12 + 14 + 15 + 16) / 5,

[0178] = 13.4,

[0179] σi² = [(10 - 13.4)² + (12 - 13.4)² + (14 - 13.4)² + (15 - 13.4)² + (16 - 13.4)²] / 4,

[0180] σi ≈ 2.0736,

[0181] Apply the standardization formula , to calculate the standardized value of each observation;

[0182] L1 = 10, X1 = (10 - 13.4) / 2.0736 ≈ -1.64,

[0183] L2 = 12, X2 = (12 - 13.4) / 2.0736 ≈ -0.67,

[0184] L3 = 14, X3 = (14 - 13.4) / 2.0736 ≈ 0.29,

[0185] L4 = 15, X4 = (15 - 13.4) / 2.0736 ≈ 0.77,

[0186] L5 = 16, X5 = (16 - 13.4) / 2.0736 ≈ 1.25,

[0187] Where: L is the observed value of the i-th feature in the original dataset, Li is the sample mean of the i-th feature, that is, the average of the values of all samples on the feature Li, σi is the standard deviation of the i-th feature L, which measures the degree of dispersion of the data distribution of this feature, and Xi is the observed value of the i-th feature after standardization;

[0188] S2: Calculate the product quality index Y based on the information characteristics of the collected unqualified target products. The calculation data is as follows:

[0189]

[0190] Substitute the data,

[0191] X1 = -1.64,

[0192] X2 = -0.67,

[0193] X3 = 0.29,

[0194] X4 = 0.77,

[0195] β0 = 1 (intercept term),

[0196] β1 = 0.5,

[0197] β2 = 0.3,

[0198] β3 = -0.4,

[0199] β4 = 0.8,

[0200] ϵ = 0.1,

[0201] Substitute the above data into the formula:

[0202] Y = 1 + (0.5 × -1.64) + (0.3 × -0.67) + (-0.4 × 0.29) + (0.8 × 0.77) + 0.1,

[0203] Y = 1 - 0.82 - 0.201 - 0.116 + 0.616 + 0.1,

[0204] Y = 0.589,

[0205] Where: Y is the target variable, representing the product quality index; β0 is the intercept term, which is the expected value of Y when all independent variables Xi are 0; β1, β2, β3, β4 are all regression coefficients, representing the influence degrees of the independent variables X1, X2, X3, X4 on Y respectively; X1, X2, X3, X4 are all independent variables; ϵ is the error term, representing the part of the variation that the model fails to explain;

[0206] S3: Predict the time when the product quality drops below the qualified line based on the product quality indicators. The prediction formula is as follows:

[0207]

[0208] Substitute the data,

[0209] The qualified line Q = 0.6,

[0210] Y = 0.589,

[0211] Substitute the known values for calculation:

[0212] t > (0.6 - 0.589) / -0.02t,

[0213] t > 0.011 / -0.02t,

[0214] The obtained result is t > -0.55,

[0215] Where: t is the time required for the quality indicator to be expected to drop below the qualified line from the current time point, Q is the qualified line, that is, the minimum quality standard allowed for the target product, Y is the quality indicator value at the current time point, (Q - Y) is the difference between the current quality indicator and the qualified line, and ΔY is the average rate of change of the quality indicator over time; it means that under the current conditions, the quality indicator is expected to drop below the qualified line within less than 0.55 unit time.

[0216] Embodiment 2:

[0217] As Figures 1-6 shown, the present invention provides a technical solution: a method and system for collecting quality data and predicting anomalies during a processing process, and the method includes:

[0218] Collect historical data, and collect the quality data of the target product and the operation data of the production equipment;

[0219] It should be noted that: the quality data of the target product includes production quality data such as appearance defects, dimensions, spacing, and solder joints on industrial boards. Collecting appearance defects involves identifying various abnormal conditions on the surface of industrial boards, such as scratches, depressions, cracks, foreign objects, etc. Dimension inspection mainly includes whether the physical dimensions of the industrial board itself meet the design specifications, and spacing detection refers to checking the distance between each hole on the industrial board.

[0220] Data processing, perform preprocessing based on the collected historical data, and summarize the preprocessed data to form a reference data set;

[0221] Extract features, extract the unqualified information and features of the target product based on the reference data set to form a feature data set;

[0222] Real-time data comparison: Compare the real-time data with the preset quality inspection standards, and obtain variable data according to the data change direction of the target product.

[0223] Variable data induction: Collect the variable data of the target product to form a variable data set.

[0224] Prediction time: Based on the variable data set, extract the unqualified information features of the target product, and predict the time when the quality of the target product drops below the qualified line. The steps of the prediction method are as follows:

[0225] S1: Analyze the variable data set to obtain the target variable, and extract the information feature variables of the unqualified target products. The calculation formula is as follows:

[0226]

[0227] Where: L is the observed value of the i-th feature in the original data set, Li is the sample average value of the i-th feature, σi is the standard deviation of the i-th feature L, and Xi is the observed value of the i-th feature after standardization.

[0228] S2: Calculate the product quality index Y based on the information features of the collected unqualified target products. The calculation data is as follows:

[0229]

[0230] Where: Y is the target variable, β0 is the intercept term, β1, β2, β3, β4 are all regression coefficients, X1, X2, X3, X4 are all independent variables, and ϵ is the error term.

[0231] S3: Predict the time when the product quality drops below the qualified line based on the product quality index. The prediction formula is as follows:

[0232]

[0233] Where: t is the time required from the current time point to when the quality index is expected to drop below the qualified line, Q is the qualified line, ΔY is the average rate of change of the quality index over time, and Y is the quality index value at the current time point.

[0234] For the data processing, clean the collected historical data to ensure the accuracy and consistency of the data. Merge and organize the processed data, and classify and file it according to time sequence, product type, equipment type, etc. to form a unified reference data set.

[0235] It should be noted that:

[0236] The calculation formula for extracting features is as follows:

[0237]

[0238] Substitute the data,

[0239] The sample quantity N of unqualified products 不合格 = 5N,

[0240] Sample values: 10, 12, 14, 15, 16,

[0241] The average value X 不合格 = (10 + 12 + 14 + 15 + 16) / 5 = 13.4,

[0242] S² 不合格 = [(10 - 13.4)² + (12 - 13.4)² + (14 - 13.4)² + (15 - 13.4)² + (16 - 13.4)²] / 4,

[0243] ≈ 4.3,

[0244] The sample quantity N of qualified products 合格 = 5,

[0245] Sample values: 8, 9, 10, 11, 128, 9, 10, 11, 12,

[0246] The average value X 合格 = (8 + 9 + 10 + 11 + 12) / 5,

[0247] = 10X,

[0248] S² 合格 = [(8 - 10)² + (9 - 10)² + (10 - 10)² + (11 - 10)² + (12 - 10)²] / 4,

[0249] = 2,

[0250] Substitute the known values:

[0251] X 不合格 = 13.4,

[0252] X 合格 = 10,

[0253] S 不合格 2≈ 4.3,

[0254] S 合格 2 = 2,

[0255] N 不合格 = 5,

[0256] N 合格 = 5,

[0257] (X不合格 −X 合格 )²=(13.4−10)²,

[0258] =3.4²,

[0259] =11.56,

[0260] S² 不合格 / N 不合格 +S² 合格 / N 合格 =4.3 / 5+2 / 5,

[0261] =0.86+0.4,

[0262] =1.26,

[0263] Therefore,

[0264] H²=11.56 / 1.26,

[0265] ≈9.17,

[0266] where: H² is the square of the original H statistic, X 不合格 and X 合格 are the average values of non - conforming products and conforming products on feature X respectively, S² 不合格 and S² 合格 are the variances of non - conforming products and conforming products on feature S respectively, N 不合格 and N 合格 are the sample sizes of non - conforming products and conforming products respectively.

[0267] The real - time data acquisition includes the following steps:

[0268] M1: Identify the first region and the second region in the image;

[0269] It should be noted that, in the form of coordinates or templates, the recognition process starts with dividing or matching the newly acquired image into regions of the same type as before, and directly locates the corresponding regions using the predefined definition information (such as position coordinates or shape templates) of the first and second regions.

[0270] M2: Increase the sampling frequency in the determined first region to ensure obtaining high - quality data and improving the data accuracy rate, and decrease the sampling frequency in the determined second region to reduce the amount of acquired data;

[0271] It should be noted that when processing product images, for the determined first region (the region that requires higher attention) and the second region (the region with relatively stable quality), a differential sampling strategy can be adopted to optimize the data acquisition efficiency and quality. The specific process is as follows: First, in the image acquisition stage, the positions of the first region and the second region are identified according to predefined coordinates or templates. Then, in the subsequent data acquisition process, a higher sampling frequency is applied to the first region, which means that more intensive data acquisition will be carried out in these key regions to ensure that all details that may affect product quality can be captured, thereby improving the accuracy and reliability of detection. On the contrary, for the second region, a lower sampling frequency is adopted to reduce unnecessary data volume, which helps to save storage space and processing time without significantly affecting the overall quality assessment result. Through this targeted sampling adjustment, while ensuring the monitoring requirements of key regions, the operating efficiency of the entire system can be improved;

[0272] M3: The collected data is losslessly compressed through a compression program to achieve fast transmission, taking into account the real-time nature and accuracy of the collected data.

[0273] It should be noted that after the data acquisition is completed, in order to achieve fast and reliable transmission while maintaining the integrity and accuracy of the data, lossless compression technology can be used to process the data. The specific process is as follows: First, identify and mark the key data that needs to be transmitted with high fidelity (such as the data in the first region) to ensure that no details of this information are lost due to compression. Then, use an efficient lossless compression algorithm (such as arithmetic coding or LZW algorithm) to compress the entire data set or the data in a specific region. This step can significantly reduce the data volume without losing any information, enabling fast transmission even under limited network bandwidth. At the receiving end, the original data is restored using the corresponding decompression program, ensuring that the real-time nature and accuracy of the data are not affected. In this way, both the requirements for high-quality transmission of key data can be met, and the data volume can be effectively managed, improving the overall transmission efficiency.

[0274] For the real-time data comparison, the real-time data is compared with the quality detection standard. According to the data change direction of the target product, variable data is obtained, and the variable data of unqualified target products is supplemented to the feature data set. The steps of the prediction method are as follows:

[0275] S1: Analyze the variable data set to obtain the target variable, and extract the information feature variables of unqualified target products. The calculation formula is as follows:

[0276]

[0277] Substitute the data:

[0278] L1 = 11,

[0279] L2 = 12,

[0280] L3 = 13,

[0281] L4 = 14,

[0282] L5 = 15,

[0283] Calculate the sample mean Li and the standard deviation σi of each feature.

[0284] Li = 1 / n * ∑² i=1 Li,

[0285] = (11 + 12 + 13 + 14 + 15) / 5,

[0286] = 13,

[0287] σi² = [(11 - 13)² + (12 - 13)² + (13 - 13)² + (14 - 13)² + (15 - 13)²] / 4,

[0288] σi = 2.5,

[0289] Apply the normalization formula XXi = (L - Li) / σi to calculate the normalized value of each observation;

[0290] L1 = 11, X1 = (11 - 13) / 2.5 = -0.8,

[0291] L2 = 12, X2 = (12 - 13) / 2.5 = -0.4,

[0292] L3 = 13, X3 = (13 - 13) / 2.5 = 0,

[0293] L4 = 14, X4 = (14 - 13) / 2.5 = 0.25,

[0294] L5 = 15, X5 = (15 - 13) / 2.5 = 0.8,

[0295] Where: L is the observation value of the i-th feature in the original dataset, Li is the sample mean of the i-th feature, that is, the average of the values of all samples on the feature Li, σi is the standard deviation of the i-th feature L, measuring the degree of dispersion of the data distribution of this feature, and Xi is the observation value of the i-th feature after normalization;

[0296] S2: Calculate the product quality index Y based on the information characteristics of the collected non-conforming target products. The calculation data is as follows:

[0297]

[0298] Substitute the data,

[0299] X1 = -0.8,

[0300] X2 = -0.4,

[0301] X3 = 0,

[0302] X4 = 0.25,

[0303] β0 = 1 (intercept term),

[0304] β1 = 0.5,

[0305] β2 = 0.3,

[0306] β3 = -0.4,

[0307] β4 = 0.8,

[0308] ϵ = 0.1,

[0309] Substitute the above data into the formula:

[0310] Y = 1 + (0.5 × -0.8) + (0.3 × -0.4) + (-0.4 × 0) + (0.8 × 0.25) + 0.1,

[0311] Y = 1 - 0.4 - 0.12 + 0 + 0.2 + 0.1,

[0312] Y = 0.78,

[0313] where: Y is the target variable, representing the product quality index; β0 is the intercept term, which is the expected value of Y when all independent variables Xi are 0; β1, β2, β3, β4 are all regression coefficients, representing the influence degrees of independent variables X1, X2, X3, X4 on Y respectively; X1, X2, X3, X4 are all independent variables; ϵ is the error term, representing the part of variation that the model fails to explain;

[0314] S3: Predict the time when the product quality drops below the qualified line based on the product quality index. The prediction formula is as follows:

[0315]

[0316] Substitute the data,

[0317] The qualified line Q = 0.8,

[0318] Y = 0.78,

[0319] Substitute the known values for calculation:

[0320] t > (0.8 - 0.78) / -0.02t,

[0321] t > 0.02 / -0.02t,

[0322] The obtained result is t > -1,

[0323] where: t is the time required from the current time point until the quality index is expected to fall below the qualified line, Q is the qualified line, that is, the lowest quality standard allowed for the target product, Y is the value of the quality index at the current time point, (Q - Y) is the difference between the current quality index and the qualified line, and ΔY is the average rate of change of the quality index over time, indicating that under the current conditions, the quality index is expected to fall below the qualified line in less than one unit of time.

[0324] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended embodiments and their equivalents.

Claims

1. A method for quality data collection and abnormality prediction during processing, characterized in that: The method comprises: Historical data collection: collecting quality data of target products and operating data of production equipment; Data processing: preprocessing based on the collected historical data, summarizing the preprocessed data to form a reference data set; Extract features, extract the unqualified information and features of the target product based on the reference data set to form a feature data set; Acquisition area differentiation: based on the unqualified information and characteristics of the target product in the reference data set, partition calculation is performed to distinguish the first area and the second area; Real-time data collection, based on the first area and the second area, accurate collection and fuzzy collection of target products; Real-time data comparison: compare the real-time data with the preset quality inspection standards, and obtain variable data according to the data change direction of the target product; Variable data induction, collecting variable data of target products to form variable data sets; Predicting time: extracting the unqualified information features of the target product based on the variable data set, and predicting the time when the quality of the target product falls below the qualified line. The prediction method steps include: S1: Analyze the variable data set, obtain the target variable, and extract the information feature variables of the unqualified target products; S2: Calculate product quality index Y based on the collected information features of unqualified target products; S3: Based on product quality indicators, predict the time when product quality will fall below the qualified line.

2. The method for quality data collection and abnormality prediction during processing according to claim 1, characterized in that: The data processing includes cleaning the collected historical data to ensure the accuracy and consistency of the data, merging and sorting the processed data, and classifying and archiving the data according to chronological order, product type, equipment type, etc. to form a unified reference data set.

3. The method for quality data collection and abnormality prediction during processing according to claim 1, characterized in that: The calculation formula for extracting features is as follows: , where: H² is the square of the original H statistic, X 不合格 and X 合格 are the average values ​​of unqualified products and qualified products on feature X, S² 不合格 and S² 合格 are the variances of unqualified products and qualified products on feature S, N 不合格 and N 合格 are the sample numbers of unqualified products and qualified products respectively.

4. The method for quality data collection and abnormality prediction during processing according to claim 1, characterized in that: The acquisition area distinction is based on the reference data set to calculate the defect frequency, defect severity and area of ​​the target product. The regional score calculation formula is as follows: , where: i is the region index, Fi is the defect frequency in the i-th region, Si is the defect severity in the i-th region, Ai is the area of ​​the i-th region, and Score(i) is the score of the i-th region.

5. The method for quality data collection and abnormality prediction during processing according to claim 1, characterized in that: The acquisition area differentiation is based on the reference data set to differentiate the defect frequency, defect severity and area of ​​the target product. The regional differentiation calculation formula is as follows: , where: Score(i) is the score of the ith region, Threshold is the score limit for screening the first region, R is the set of all first region indexes, when the target region score is greater than or equal to the score limit, it is the first region, when the target region score is lower than the score limit, it is the second region.

6. The method for quality data collection and abnormality prediction during processing according to claim 1, characterized in that: The real-time data collection steps include: M1: Identify the first area and the second area in the image; M2: increasing the sampling frequency within the determined first area to ensure the acquisition of high-quality data and improve the accuracy of the data, and reducing the sampling frequency within the determined second area to reduce the amount of collected data; M3: The collected data is losslessly compressed through the compression program to achieve fast transmission, taking into account the real-time and accuracy of the collected data.

7. The method for quality data collection and abnormality prediction during processing according to claim 1, characterized in that: The real-time data comparison compares the real-time data with the quality inspection standard, obtains variable data according to the data change direction of the target product, and supplements the variable data of the unqualified target product to the feature data set.

8. The method for quality data collection and abnormality prediction during processing according to claim 1, characterized in that: The prediction method steps are as follows: S1: Analyze the variable data set, obtain the target variable, and extract the information characteristic variables of the unqualified target products. The calculation formula is as follows: , where: L is the observed value of the i-th feature in the original data set, Li is the sample average of the i-th feature, that is, the average of the values ​​of all samples on the feature Li, σi is the standard deviation of the i-th feature L, which measures the degree of dispersion of the feature data distribution, and Xi is the observed value of the i-th feature after standardization; S2: Calculate the product quality index Y based on the collected information characteristics of the unqualified target products. The calculation data is as follows: , where: Y is the target variable, representing the product quality index; β0 is the intercept term, which is the expected value of Y when all independent variables Xi are 0; β1, β2, β3, β4 are regression coefficients, which respectively represent the influence of independent variables X1, X2, X3, X4 on Y; X1, X2, X3, X4 are all independent variables; ϵ is the error term, which represents the part of the variation that the model cannot explain; S3: Based on the product quality indicators, the time when the product quality falls below the qualified line is predicted. The prediction formula is as follows: , where: t is the time from the current time point to the time when the quality indicator is expected to fall below the qualified line, Q is the qualified line, that is, the minimum quality standard allowed for the target product, Y is the quality indicator value at the current time point, (Q−Y) is the difference between the current quality indicator and the qualified line, and ΔY is the average rate of change of the quality indicator over time.

9. The quality data collection and abnormality prediction system during processing according to claim 1, characterized in that: The method for collecting quality data and predicting abnormalities during processing described in any one of claims 1 to 8 is used.

Images