Staged quality detection method and system for SMT (Surface Mount Technology) semi-finished product

Abstract

The invention discloses a staged quality detection method and system for SMT (Surface Mount Technology) semi-finished products, and particularly relates to the technical field of electronic manufacturing quality management.The staged quality detection method comprises the following steps: performing quality detection on each stage of a production line, recording detection results and forming a detection result set, identifying early signs of production rhythm imbalance by using time interval distribution analysis; if an imbalance sign exists, further calculating a defect frequency fluctuation index and a production line dynamic disorder index, and quantifying dynamic fluctuation and systematicness problems of an unqualified phenomenon; inputting the two indexes into a pre-trained machine learning model, and distinguishing whether the quality state of the production line is high quality or low quality through model output; a low-quality production line is optimized according to a preset adjustment strategy; the method can monitor the quality state of the production line in real time, accurately distinguish accidental problems from non-accidental problems, improve the quality management efficiency, guarantee the stable operation of the production line, and reduce the proportion of unqualified products.

Description

Technical Field

[0001] The present invention relates to the technical field of electronic manufacturing quality management, and more particularly to a staged quality detection method and system for SMT patch semi-finished products. Background Art

[0002] As electronic products demand ever-increasing performance and shrink in size, surface mount technology (SMT) has become a critical process in modern electronics manufacturing. In SMT production lines, automated equipment completes a series of steps, from solder paste printing and component placement to soldering. However, quality issues (such as defective products) can frequently occur due to the complexity of the production process and unstable equipment operation.

[0003] The main issues facing current production lines include: Difficulty in dynamic quality management: SMT patch production lines require continuous, high-volume production. Product quality is affected by a variety of factors, including equipment status, production rhythm, and material properties. Fluctuations in production rhythm and declines in equipment performance can easily lead to an increase in defective products. Unable to accurately determine the nature of quality issues: Defective products can be accidental (such as due to transient errors) or non-accidental (such as due to systemic defects). Traditional inspection methods struggle to distinguish between the two, easily leading to over- or under-adjustment of the production line. Therefore, a staged quality inspection method and system for semi-finished SMT patch products is proposed to address these issues. Summary of the Invention

[0004] To achieve the above object, the present invention provides the following technical solutions:

[0005] A staged quality inspection method for semi-finished SMT patches includes the following steps:

[0006] Perform quality inspections on each production stage of SMT patch semi-finished products, and record the quality inspection results of each production stage to obtain a set of inspection results for each production stage;

[0007] Perform time interval distribution analysis on the test result set, and identify whether there are early signs of production rhythm imbalance in the corresponding production stage based on the time interval distribution analysis results;

[0008] If there are early signs of imbalance in the production rhythm at the corresponding production stage, defect analysis and disorder analysis are performed on the production line to generate the defect frequency fluctuation index and the production line dynamic disorder index respectively;

[0009] The defect frequency fluctuation index and the production line dynamic disorder index are substituted into the pre-trained machine learning model. According to the output results, the corresponding production stage is divided into high-quality production lines and low-quality production lines, and the low-quality production lines are adjusted according to the preset adjustment strategy.

[0010] The production stages of SMT patch semi-finished products include: printing stage, mounting stage, and welding stage. Quality inspection is carried out at each production stage in accordance with the preset quality inspection strategy of the corresponding production stage.

[0011] In a preferred embodiment, performing time interval distribution analysis on the detection result set refers to:

[0012] During the production line inspection process, obtain the set of inspection time points where all quality inspection results are marked as unqualified products, arrange them in chronological order, and obtain the inspection time series of unqualified products;

[0013] The time intervals of adjacent time points in the detection time series are calculated to generate a time interval distribution dataset. The time interval represents the length of the time interval between two consecutive detections. Feature extraction is then performed on the time interval distribution dataset to obtain statistical features including the following:

[0014] The average value of the time interval is used to reflect the average inspection rhythm of the production line. The average value of the time interval refers to the average value of all time intervals.

[0015] The fluctuation range of the time interval is used to reflect the stability of the production rhythm. The fluctuation range of the time interval refers to the maximum time interval minus the minimum time interval;

[0016] The time interval anomaly ratio is used to identify the frequency of abnormal rhythmic behavior. The time interval anomaly ratio refers to the ratio of the number of time intervals that are not within the preset time standard range to the total number of time intervals.

[0017] In a preferred embodiment, identifying whether there is an early sign of production rhythm imbalance in the corresponding production stage based on the time interval distribution analysis result refers to:

[0018] When the average value of the time interval is not within the preset normal rhythm range, an abnormal signal is generated;

[0019] When the fluctuation range of the time interval is less than the preset threshold, an abnormal signal is generated;

[0020] When the ratio of abnormal points in a time interval exceeds a preset limit, an abnormal signal is generated;

[0021] When an abnormal signal is generated, it indicates that there are early signs of production rhythm imbalance in the corresponding production stage. When no abnormal signal is generated, it indicates that there are no early signs of production rhythm imbalance in the corresponding production stage.

[0022] In a preferred embodiment, the logic for obtaining the defect frequency fluctuation index is:

[0023] In a time interval consisting of N time windows, obtain the unqualified frequency fi in time window i, calculate the mean of the unqualified frequencies of all time windows as μDF, and calculate the standard deviation of the unqualified frequencies of all time windows as σDF;

[0024] Calculate the rate of change of the unqualified frequency fi', fi' = fi-fj; j = i-1; then calculate the average absolute value of the rate of change fi' Then substitute it into the following calculation formula:

[0025] C is a non-negative constant, γ is a preset non-zero adjustment factor, and IDF is the defect frequency fluctuation index.

[0026] In a preferred embodiment, the logic for obtaining the dynamic disorder index of the production line is:

[0027] Within a time interval consisting of N time windows, the start time of each test in the test time series of all products, the actual number of tests within a preset standard unit time, and the test results, i.e., pass and fail marks, are obtained from the test result set. The time interval between the start times of each test is then calculated to obtain a time interval test data set. The standard deviation of the time interval test data set is calculated, as well as the standard deviation reflecting the fluctuation of the actual number of tests within the standard unit time.

[0028] Calculate the ratio of qualified products to the actual number of tested products in each time window i, obtain the window qualified rate of time window i, and calculate the standard deviation of the qualified rates of all windows;

[0029] Perform the following calculations:

[0030] μΔt is the standard deviation of the time interval detection data set, σHG is the standard deviation of the qualified rate of all windows, σQ is the standard deviation reflecting the fluctuation of the actual number of detections per standard unit time, and ILD is the dynamic disorder index of the production line.

[0031] In a preferred embodiment, the usage logic of the machine learning model is:

[0032] The defect frequency fluctuation index and the production line dynamic disorder index are substituted into the pre-trained convolutional neural network model. The convolutional neural network model outputs a result of 1 or 0. The output result 1 indicates that the corresponding production stage is divided into a high-quality production line, indicating that the unqualified production situation of the production line corresponding to this production stage is accidental. The output result 0 indicates that the corresponding production stage is divided into a low-quality production line, indicating that the unqualified production situation of the production line corresponding to this production stage is not accidental.

[0033] In a preferred embodiment, a staged quality inspection system for semi-finished SMT patches includes:

[0034] The quality inspection module performs quality inspection on each production stage of the SMT patch semi-finished product, and records the quality inspection results of each production stage to obtain the inspection result set of each production stage;

[0035] The early signs identification module performs time interval distribution analysis on the detection result set and identifies whether there are early signs of production rhythm imbalance in the corresponding production stage based on the time interval distribution analysis results;

[0036] The production line analysis module performs defect analysis and disorder analysis on the production line if there are early signs of imbalance in the production rhythm at the corresponding production stage, generating a defect frequency fluctuation index and a production line dynamic disorder index respectively;

[0037] The production line classification module feeds the defect frequency fluctuation index and the production line dynamic disorder index into a pre-trained machine learning model and classifies the corresponding production stages into high-quality and low-quality production lines based on the output results.

[0038] The production line adjustment module adjusts low-quality production lines according to preset adjustment strategies.

[0039] The technical effects and advantages of the present invention are as follows:

[0040] Through the analysis of time interval distribution, calculation and judgment of defect frequency fluctuation index and production line dynamic disorder index, the present invention can distinguish whether the unqualified phenomenon is accidental or a systemic problem. Accidental problem: avoid excessive intervention, maintain normal operation of the production line, and improve production efficiency. Non-accidental problem: timely identify systemic hidden dangers, locate the root cause, and reduce unnecessary quality losses. By real-time monitoring of the unqualified frequency and rhythm fluctuations of the production line, the present invention can capture potential problems in the operation of the production line in real time, form early warning signals, improve the dynamic response capability of the production line, avoid large-scale shutdowns or rework caused by the accumulation of problems, and realize full intelligent monitoring of the quality status of the production line.

[0041] By substituting the defect frequency fluctuation index and the dynamic disorder index into the machine learning model, the present invention can intelligently judge the quality status of the production stage without the need for human participation in the analysis, thereby reducing the subjectivity and errors of manual judgment, reducing the cost of manual intervention, and improving detection efficiency. The present invention accurately classifies high-quality and low-quality production lines: For high-quality production lines: no excessive adjustments are required, reducing unnecessary waste of resources. For low-quality production lines: start preset optimization strategies (such as shutdown maintenance, process adjustment, etc.), concentrate resources to solve quality problems, realize scientific allocation of resources, and avoid losses caused by shutdown of the entire line.

[0042] The present invention can significantly reduce the proportion of unqualified products by accurately monitoring and optimizing the production line quality management process: improve the overall product qualification rate of the production line, ensure that the quality of the final product meets the design requirements, reduce the rework rate and the waste of raw materials, and improve production efficiency. The present invention covers multiple stages of the SMT patch production line, including solder paste printing, component placement and welding, etc. The quality inspection data of each stage are recorded and analyzed to ensure the quality controllability of the entire production process, and can quickly locate the source of unqualified phenomena and reduce the time for problem investigation. The present invention is applicable to a variety of electronic product production scenarios (such as mobile phones, automotive electronics, smart devices, etc.). Through intelligent management, it reduces downtime for maintenance, reduces the rework cost and material loss caused by unqualified products, improves production efficiency and yield rate, and reduces the operating cost of the production line as a whole. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings;

[0044] Figure 1 The figure is a schematic diagram of a staged quality inspection method for semi-finished SMT patches in the present invention.

[0045] Figure 2 The figure is a schematic diagram of a stage-by-stage quality inspection system for semi-finished SMT patches according to the present invention. DETAILED DESCRIPTION

[0046] 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.

[0047] Reference Figure 1 - Figure 2 The following examples were obtained:

[0048] Example 1: A staged quality inspection method for a semi-finished SMT patch product, comprising the following steps:

[0049] Quality inspections are conducted at every stage of production for semi-finished SMT products, and the results are recorded at each stage, resulting in a comprehensive set of inspection results for each stage. Independent quality inspections at all stages, including printing, placement, and soldering, pinpoint potential production issues and ensure that every stage of the production line meets quality requirements. Quality inspection results at each stage are systematically recorded, forming a comprehensive set of inspection results that provides a reliable data foundation for subsequent analysis and potential hazard identification. This allows for early detection and resolution of potential production issues, reducing finished product scrap and rework.

[0050] The time interval distribution analysis of the test result set can be used to identify early signs of an imbalance in the production rhythm during the corresponding production phase. By analyzing the test time series, the changing characteristics of the production rhythm can be identified to determine whether there is a trend of rhythm imbalance. The statistical characteristics of the test intervals (such as average value, fluctuation range, and outlier ratio) can promptly reveal potential equipment problems, manual errors, or unreasonable process design, providing a quantitative basis for production line adjustments and preventing further problems from escalating.

[0051] If there are early signs of an imbalance in the production rhythm during a specific production phase, defect and disorder analyses are performed on the production line, generating a defect frequency fluctuation index and a line dynamic disorder index, respectively. The defect frequency fluctuation index reflects the frequency and fluctuation of defective products, helping to assess whether defects are accidental or systematic. The dynamic disorder index reflects the overall operational stability of the production line through a comprehensive analysis of time intervals, fluctuations in inspection quantities, and changes in pass rates. The combination of these two indices provides a more comprehensive picture of the current level of abnormality and whether it is sporadic, providing a basis for subsequent decision-making.

[0052] The defect frequency fluctuation index and the production line dynamic disorder index are fed into a pre-trained machine learning model. Based on the output, the corresponding production stage is divided into high-quality and low-quality production lines. Low-quality production lines are then adjusted according to pre-set adjustment strategies. The machine learning model enables rapid and accurate identification of the production line's quality status, avoiding errors caused by subjective judgment or manual analysis. Classification standardization: The output is either 1 (high quality) or 0 (low quality), providing clear categorization for production line status assessment. A high-quality production line indicates that the nonconformity is sporadic and does not require extensive intervention. A low-quality production line indicates that the nonconformity is systemic and requires focused adjustment. Implementing adjustment measures for low-quality production lines can effectively reduce the risk of large-scale production errors. Efficient problem resolution: Pre-set adjustment strategies (such as downtime for maintenance, equipment calibration, and personnel operational improvements) can quickly resolve identified issues, ensuring production efficiency and quality. Improving production line stability: Implementing adjustment measures ensures that the production line continues to operate in a stable state, thereby improving overall production efficiency.

[0053] The production stages of SMT patch semi-finished products include: printing stage, mounting stage, and welding stage. Quality inspection is carried out at each production stage in accordance with the preset quality inspection strategy of the corresponding production stage. The processing links of semi-finished products include the following:

[0054] Printing stage (solder paste printing): The status is semi-finished product. The product at this stage is only a bare board PCB printed with solder paste, and components have not yet been mounted. It is a typical semi-finished product.

[0055] Mounting stage (component mounting): The state is semi-finished product. At this stage, the components have been mounted on the PCB, but have not yet been soldered and fixed. The solder paste is in an unmelted state and the electrical function has not yet formed a complete circuit.

[0056] Soldering (reflow) stage: After soldering is complete, the product is considered nearly finished. After reflow soldering, the components are firmly soldered to the PCB, and the circuit has basic electrical functionality. However, the product may still have defects (such as cold solder joints and solder bridging) and require further testing and verification.

[0057] Semi-finished product status: Semi-finished products refer to products that have completed some processes, such as accurately placing components on the PCB through a placement machine, but the product has not yet fully reached the finished product standard.

[0058] Key characteristics: Components have been mounted on the PCB but have not yet been soldered and secured via the reflow process. Alternatively, reflow has been completed but functional testing or other final inspections have not yet been performed. Before mounting or soldering is complete, semi-finished products typically cannot achieve complete electrical functionality because circuit connections are not yet fully formed.

[0059] Performing time interval distribution analysis on the test result set refers to:

[0060] During the production line inspection process, a set of inspection time points where all quality inspection results are marked as unqualified products is obtained and arranged in chronological order to obtain the inspection time series of unqualified products. By extracting the inspection time points of unqualified products, we focus on analyzing the temporal distribution patterns of abnormal phenomena to avoid loss of analysis focus due to interference from qualified product data.

[0061] The time intervals of adjacent time points in the detection time series are calculated to generate a time interval distribution dataset. The time interval represents the length of the time interval between two consecutive detections. Feature extraction is then performed on the time interval distribution dataset to obtain statistical features including the following:

[0062] The average time interval reflects the average inspection rhythm of the production line. The average time interval is the average of all time intervals. By calculating the time intervals between adjacent time points, we can reveal whether the production line's rhythm is regular and consistent with expectations. Sudden changes in time intervals may indicate unstable equipment operation, operational deviations, or abnormal production processes. Constructing a time interval distribution dataset provides the data foundation for subsequent statistical analysis.

[0063] The fluctuation range of the time interval is used to reflect the stability of the production rhythm. The fluctuation range of the time interval refers to the maximum time interval minus the minimum time interval. The larger the fluctuation range, the more unstable the production rhythm, which may be due to fluctuations in equipment performance or inconsistencies in human operations. The difference between the maximum and minimum time intervals can intuitively identify whether there are sudden abnormalities in the production rhythm. If the fluctuation range is too large, it may be necessary to subsequently inspect key equipment or adjust the operating procedures.

[0064] The time interval anomaly ratio is used to identify the frequency of abnormal rhythm behavior. It refers to the ratio of time intervals that fall outside the preset time standard to the total number of time intervals. By counting the anomaly ratio, you can quickly determine whether the production rhythm imbalance is systemic. A low anomaly ratio indicates that the anomaly detection is a rare occurrence and the overall production rhythm is relatively stable. A high anomaly ratio indicates a serious rhythm imbalance on the production line, which may require downtime and inspection.

[0065] Identifying early signs of production rhythm imbalance in the corresponding production phase based on the time interval distribution analysis results refers to:

[0066] When the average value of the time intervals falls outside the preset normal rhythm range, an abnormal signal is generated; the average value of the time intervals reflects the frequency of defective products. If the average value falls outside the preset normal rhythm range, it may indicate systemic problems in certain links of the production line. For example, improper process adjustments may lead to the accumulation of problems, or resource scheduling errors in a certain stage of the operation may lead to frequent defective products.

[0067] When the fluctuation range of the time interval is less than the preset threshold, an abnormal signal is generated; the fluctuation range of the time interval can reveal whether the appearance of unqualified products has periodic fluctuation characteristics. A fluctuation range that is too large may indicate instability in the production process, leading to increased quality fluctuations.

[0068] When the ratio of abnormal points in the time interval exceeds the preset limit, an abnormal signal is generated; the ratio of abnormal points in the time interval reflects whether the occurrence of unqualified products is systematic. If the ratio of abnormal points is high, it may indicate that the quality problem is not a random phenomenon, but a common problem in the production stage.

[0069] The generation of an abnormal signal indicates early signs of an imbalance in the production rhythm during the corresponding production phase. The absence of an abnormal signal indicates no such imbalance. Comprehensive analysis of average values, fluctuation ranges, and outlier ratios can capture quality fluctuations during the production process. The generation of abnormal signals directly indicates early signs of quality issues, paving the way for precise intervention. Dynamic analysis of time interval characteristics provides timely warnings and identification of systemic quality risks, preventing them from escalating to the entire production line and ensuring quality controllability and consistency throughout the production process.

[0070] The logic for obtaining the defect frequency fluctuation index is:

[0071] In a time interval consisting of N time windows, obtain the unqualified frequency fi in time window i, calculate the mean of the unqualified frequencies of all time windows as μDF, and calculate the standard deviation of the unqualified frequencies of all time windows as σDF;

[0072] Calculate the rate of change of the unqualified frequency fi', fi' = fi-fj; j = i-1; then calculate the average absolute value of the rate of change fi' Then substitute it into the following calculation formula:

[0073] C is a non-negative constant, γ is a preset non-zero adjustment factor, and IDF is the defect frequency fluctuation index.

[0074] The mean of the failure frequency represents the average frequency of defective products across all time windows, reflecting the overall level of nonconformity. The standard deviation of the failure frequency measures the degree of fluctuation in the failure frequency across all time windows, indicating the stability or discreteness of quality issues during the production process. The failure frequency change rate represents the magnitude of the change in the failure frequency between adjacent time windows, reflecting dynamic volatility. The average absolute value of the change rate quantifies the overall trend of these fluctuations.

[0075] The adjustment factor γ controls the weight of the rate of change, increasing or decreasing the impact of dynamic fluctuations in the rejection frequency. C prevents the denominator from reaching zero while smoothing the output of the overall formula to ensure index stability. The defect frequency fluctuation index (IDF) comprehensively measures the overall level, fluctuation amplitude, and dynamic trend of the rejection frequency, reflecting the dynamic behavior of the rejection phenomenon. A small IDF indicates a stable rejection frequency, potentially indicating a sporadic problem. For example, the occurrence of rejections within a certain time window may be due to a temporary operational error or equipment fluctuation. A large IDF indicates significant fluctuations in the rejection frequency, with a systematic trend in frequency change, indicating a strong likelihood of non-random anomalies. This often indicates quality issues within the production line caused by non-random factors, such as aging equipment, raw material batch issues, or unstable process parameters. The IDF value can be used as input to a machine learning model and combined with other indicators (such as the production line dynamic disorder index) to further determine the nature of the rejection phenomenon through the model output. Random anomalies indicate that the production line is generally stable and does not require extensive intervention. Non-random anomalies require downtime for maintenance or adjustments to the production process to prevent the problem from spreading.

[0076] The logic for obtaining the dynamic disorder index of the production line is:

[0077] Within a time interval consisting of N time windows, the start time of each test in the test time series of all products, the actual number of tests within a preset standard unit time, and the test results, i.e., pass and fail marks, are obtained from the test result set. The time interval between the start times of each test is then calculated to obtain a time interval test data set. The standard deviation of the time interval test data set is calculated, as well as the standard deviation reflecting the fluctuation of the actual number of tests within the standard unit time.

[0078] Calculate the ratio of qualified products to the actual number of tested products in each time window i, obtain the window qualified rate of time window i, and calculate the standard deviation of the qualified rates of all windows;

[0079] Perform the following calculations:

[0080] μΔt is the standard deviation of the time interval detection data set, σHG is the standard deviation of the qualified rate of all windows, σQ is the standard deviation reflecting the fluctuation of the actual number of detections within the standard unit time, ILD is the dynamic disorder index of the production line, and α and β are the preset adjustment coefficients respectively.

[0081] The standard deviation of the time interval detection data set indicates the degree of fluctuation in the intervals between adjacent detection time points. If the fluctuation is small, it means that the detection time is relatively uniform and the production rhythm is stable. If the fluctuation is large, it may indicate an uneven production process. The standard deviation of the fluctuation in the number of detections per unit time measures the degree of fluctuation in the number of products detected per unit time. Small fluctuations in the number indicate that the production line load is relatively stable. Large fluctuations in the number indicate that there may be process adjustments, equipment bottlenecks, or unstable material supply issues in the production process. The standard deviation of the pass rate measures the volatility of the pass rate within the time window. Small fluctuations in the pass rate indicate stable production line quality. Large fluctuations in the pass rate indicate that there may be periodic quality problems or systemic failures in certain links in the production line. α and β respectively adjust the influence weights of the fluctuations in the pass rate and the fluctuations in the number of detections in the final index.

[0082] If the ILD is small, it means that the dynamic changes of the production line are small, and the fluctuations in the detection time interval, pass rate and detection quantity are all within a reasonable range. The emergence of unqualified products is more likely to be accidental local problems. The index is large: the anomaly is a non-accidental phenomenon. If the ILD is large, it means that there is a systematic dynamic disorder problem in the production line, which may be caused by long-term unstable operation of certain links (such as equipment, materials or processes). These problems are more likely to be non-accidental. Combining the ILD with the defect frequency fluctuation index, the machine learning model is used to determine whether the anomaly is accidental or non-accidental: Accidental anomaly: the production process is stable as a whole, and only local problems exist; non-accidental anomaly: the dynamic fluctuations are significant, and the production line needs to be adjusted to prevent the problem from expanding.

[0083] The usage logic of the machine learning model is:

[0084] The defect frequency fluctuation index and the production line dynamic disorder index are substituted into the pre-trained convolutional neural network model. The convolutional neural network model outputs a result of 1 or 0. The output result 1 indicates that the corresponding production stage is divided into a high-quality production line, indicating that the unqualified production situation of the production line corresponding to this production stage is accidental. The output result 0 indicates that the corresponding production stage is divided into a low-quality production line, indicating that the unqualified production situation of the production line corresponding to this production stage is not accidental.

[0085] Using data from the production line quality inspection process, two core indices are generated:

[0086] Defect frequency fluctuation index: used to reflect the frequency of occurrence of unqualified products and its dynamic fluctuation characteristics.

[0087] Production Line Dynamic Disturbance Index: This index measures fluctuations in the production line's operating time intervals, pass rate, and number of inspections, reflecting overall dynamic stability. These indices serve as input features and are fed into a machine learning model for comprehensive analysis.

[0088] Convolutional neural network model processing logic: The machine learning model extracts and analyzes the characteristics of the input defect frequency fluctuation index and production line dynamic disorder index to capture the regularity of the production line's quality status. The pre-training process of the convolutional neural network model has learned the quality anomaly patterns and regularities in a large amount of production line operation data. Through model calculations, it outputs a judgment result indicating whether the current production stage is in a high or low quality state.

[0089] Output: "High Quality": When the model output is 1, the production line is performing at a high quality level. At this point, the occurrence of defective products is considered accidental, a random, non-systemic problem, and the overall production line is operating stably.

[0090] "Low Quality" Output: When the model output is 0, it indicates that the production line is in a low-quality state during the current production phase. In this case, the occurrence of substandard products is considered a systemic problem, not an accidental one, and suggests that the production line may have long-term issues such as equipment anomalies, process problems, or operational errors.

[0091] Adjustment Mechanism: For low-quality production lines, pre-set adjustment strategies are initiated, which may include shutting down for maintenance, optimizing process parameters, or checking equipment status. For high-quality production lines, the status quo is maintained without excessive intervention.

[0092] By comprehensively analyzing two core indicators—dynamic disturbances and defect fluctuations—the model can accurately determine whether the production line's operating quality meets expectations and improve intelligent production management. The machine learning model leverages patterns in historical data to intelligently classify the production line's quality status, reducing subjective errors and inefficient analysis processes often associated with manual judgment. It effectively distinguishes between accidental and non-accidental anomalies: Defective products on high-quality production lines are often accidental. The model can help rule out these low-probability events and avoid unnecessary line adjustments. Defective products on low-quality production lines are often systemic issues. The model promptly identifies these non-accidental risks, providing a basis for line optimization. It improves production line quality control capabilities: Using the model's classification results, managers can optimize low-quality production lines to reduce the systemic occurrence of defective products. It reduces waste and downtime: The machine learning model's precise classification allows resources to be focused on repairing and optimizing low-quality production lines without interfering with the normal operation of high-quality lines, thereby reducing unnecessary downtime and production waste.

[0093] Example 2: A staged quality inspection system for semi-finished SMT patches, comprising:

[0094] The quality inspection module performs quality inspection on each production stage of the SMT patch semi-finished product, and records the quality inspection results of each production stage to obtain the inspection result set of each production stage;

[0095] The early signs identification module performs time interval distribution analysis on the detection result set and identifies whether there are early signs of production rhythm imbalance in the corresponding production stage based on the time interval distribution analysis results;

[0096] The production line analysis module performs defect analysis and disorder analysis on the production line if there are early signs of imbalance in the production rhythm at the corresponding production stage, generating a defect frequency fluctuation index and a production line dynamic disorder index respectively;

[0097] The production line classification module feeds the defect frequency fluctuation index and the production line dynamic disorder index into a pre-trained machine learning model and classifies the corresponding production stages into high-quality and low-quality production lines based on the output results.

[0098] The production line adjustment module adjusts low-quality production lines according to preset adjustment strategies.

[0099] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0100] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0101] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0102] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0103] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A staged quality inspection method for SMT patch semi-finished products, characterized in that: The following steps are involved: Perform quality inspection on each production stage of SMT patch semi-finished products, and record the quality inspection results of each production stage to obtain the inspection result set of each production stage; Perform time interval distribution analysis on the detection result set, and identify whether there are early signs of production rhythm imbalance in the corresponding production stage based on the time interval distribution analysis results; If there are early signs of imbalance in the production rhythm at the corresponding production stage, defect analysis and disorder analysis are performed on the production line to generate a defect frequency fluctuation index and a production line dynamic disorder index, respectively; The defect frequency fluctuation index and the production line dynamic disorder index are substituted into the pre-trained machine learning model. According to the output results, the corresponding production stage is divided into high-quality production lines and low-quality production lines. The low-quality production line is adjusted according to the preset adjustment strategy.

2. A staged quality inspection method for SMT patch semi-finished products according to claim 1, characterized in that: The production stages of SMT patch semi-finished products include: printing stage, mounting stage, and welding stage. Quality inspection is carried out at each production stage in accordance with the preset quality inspection strategy of the corresponding production stage.

3. A staged quality inspection method for SMT patch semi-finished products according to claim 2, characterized in that: Performing time interval distribution analysis on the test result set refers to: During the production line inspection process, obtain the inspection time points of all quality inspection results marked as unqualified products, arrange them in chronological order, and obtain the inspection time series of unqualified products; The time intervals of adjacent time points in the detection time series are calculated to generate a time interval distribution data set. The time interval represents the length of the time interval between two consecutive detections. Then, the time interval distribution data set is subjected to feature extraction to obtain statistical features including the following: The average value of the time interval is used to reflect the average detection rhythm of the production line. The average value of the time interval refers to the average value of all time intervals. The fluctuation range of the time interval is used to reflect the stability of the production rhythm. The fluctuation range of the time interval refers to the maximum time interval minus the minimum time interval; The time interval anomaly point ratio is used to identify the frequency of abnormal rhythmic behavior. The time interval anomaly point ratio refers to the ratio of the number of time intervals that are not within the preset time standard range to the total number of time intervals among all time intervals.

4. The staged quality inspection method for SMT patch semi-finished products according to claim 3 is characterized in that: Based on the time interval distribution analysis results, early signs of imbalance in production rhythm can be identified in the corresponding production phase: When the average value of the time interval is not within the preset normal rhythm range, an abnormal signal is generated; When the fluctuation range of the time interval is less than the preset threshold, an abnormal signal is generated; When the ratio of abnormal points in the time interval exceeds the preset limit, an abnormal signal is generated; When an abnormal signal is generated, it indicates that there are early signs of an imbalance in the production rhythm in the corresponding production stage. When no abnormal signal is generated, it indicates that there are no early signs of an imbalance in the production rhythm in the corresponding production stage.

5. The staged quality inspection method for SMT patch semi-finished products according to claim 4 is characterized in that: The logic for obtaining the defect frequency fluctuation index is: In the time interval consisting of N time windows, obtain the unqualified frequency fi in time window i, calculate the mean of the unqualified frequencies of all time windows as μDF, and calculate the standard deviation of the unqualified frequencies of all time windows as σDF; Calculate the rate of change of the unqualified frequency fi', fi' = fi-fj; j = i-1; then calculate the average absolute value of the rate of change fi' Then substitute it into the following calculation formula: C is a non-negative constant, γ is a preset non-zero adjustment factor, and IDF is the defect frequency fluctuation index.

6. A staged quality inspection method for SMT patch semi-finished products according to claim 5, characterized in that: The logic for obtaining the dynamic disorder index of the production line is: In a time interval consisting of N time windows, the start time of each test in the test time series of all products, the actual number of tests within a preset standard unit time, and the test results, i.e., qualified and unqualified marks, are obtained from the test result set. Then, the time interval of the start time of each test is calculated to obtain a time interval test data set, and the standard deviation of the time interval test data set and the standard deviation reflecting the fluctuation of the actual number of tests within the standard unit time are calculated. Calculate the ratio of qualified products to the actual number of tests in each time window i, obtain the window qualified rate of time window i, and calculate the standard deviation of all window qualified rates; Perform the following calculations: μΔt is the standard deviation of the time interval detection data set, HG is the standard deviation of the pass rate of all windows, σQ is the standard deviation reflecting the actual number fluctuations detected within the standard unit time, α and β are the preset adjustment coefficients, and ILD is the dynamic disorder index of the production line.

7. A staged quality inspection method for SMT patch semi-finished products according to claim 6, characterized in that: The usage logic of the machine learning model is: The defect frequency fluctuation index and the production line dynamic disorder index are substituted into the pre-trained convolutional neural network model. The convolutional neural network model outputs a result of 1 or 0. The output result 1 indicates that the corresponding production stage is divided into a high-quality production line, indicating that the unqualified production situation of the production line corresponding to this production stage is accidental. The output result 0 indicates that the corresponding production stage is divided into a low-quality production line, indicating that the unqualified production situation of the production line corresponding to this production stage is not accidental.

8. A staged quality inspection system for SMT patch semi-finished products, implemented based on a staged quality inspection method for SMT patch semi-finished products according to any one of claims 1 to 7, characterized in that: include: The quality inspection module performs quality inspection on each production stage of the SMT patch semi-finished products, and records the quality inspection results of each production stage to obtain the inspection result set of each production stage; The early sign identification module performs time interval distribution analysis on the detection result set, and identifies whether there are early signs of production rhythm imbalance in the corresponding production stage based on the time interval distribution analysis results; The production line analysis module performs defect analysis and disorder analysis on the production line if there are early signs of imbalance in the production rhythm in the corresponding production stage, and generates a defect frequency fluctuation index and a production line dynamic disorder index respectively; The production line classification module substitutes the defect frequency fluctuation index and the production line dynamic disorder index into the pre-trained machine learning model, and divides the corresponding production stage into high-quality production lines and low-quality production lines according to the output results; The production line adjustment module adjusts low-quality production lines according to preset adjustment strategies.

Images