System and method for risk assessment of product quality inspection based on measurement uncertainty

Abstract

The application discloses a product quality inspection risk assessment system and method based on measurement uncertainty, and particularly relates to the field of computer-aided decision-making in product quality control, and is used to solve the problem of inaccurate decision risk and wasted inspection resources caused by the existing sequential inspection model ignoring the dynamic change of measurement uncertainty; by synchronously obtaining the measurement value and the current measurement uncertainty evaluation value of each sample in the sequential sampling process, when the conventional sequential decision cannot make a judgment, firstly, the statistical control state of the measurement process is analyzed, after confirming that the process is controlled, the membership value of each measurement value relative to the qualified interval is calculated by using the uncertainty, then the membership information of all samples is fused to dynamically update the batch overall risk estimation value, finally, the judgment of continuing sampling, accepting or rejecting the batch is adaptively made according to the comparison result of the risk estimation value and the threshold value, so that more accurate risk control and inspection resource saving are realized.

Description

Technical Field

[0001] This invention relates to the field of computer-aided decision-making technology in product quality control, and more specifically, to a product quality inspection risk assessment system and method based on measurement uncertainty. Background Technology

[0002] In scenarios such as the Industrial Internet, intelligent manufacturing service platforms, and industrial production big data resource services, product quality inspection, especially for products with high inspection costs or destructive potential, such as performance tests for material lifespan and strength, has seen widespread research and application of sequential sampling inspection methods to conserve inspection resources while controlling risks. The core of this method lies in dynamically deciding whether to continue sampling, accept the batch, or reject it based on the cumulative inspection results of successively drawn samples. Classical sequential decision theory provides the mathematical model foundation for this process, aiming to find the minimum average sample size under a set level of misjudgment risk.

[0003] However, the existing sequential testing decision models mentioned above typically implicitly assume, in constructing their mathematical foundations, that the test results obtained from each measurement are deterministic, or that the accompanying measurement error is a constant quantity. This assumption simplifies the theoretical derivation but deviates from actual metrological practice. This contradiction is even more pronounced in practical applications such as industrial data processing and data integration services, and industrial cloud database services. In real measurement activities, influenced by factors such as equipment status fluctuations, changes in environmental conditions, sample differences, and operational reproducibility, the measurement uncertainty of each specific measurement is often dynamically changing rather than a fixed value. Ignoring this objectively existing dynamic change and directly applying decision models based on the assumption of fixed error to practice will lead to biases in the assessment of the current decision risk. Consequently, the decision to continue sampling, accept, or reject may be made at a suboptimal time, failing to truly achieve the fundamental goal of balancing risk control and testing resource conservation. Summary of the Invention

[0004] To overcome the aforementioned deficiencies of the prior art, the present invention provides a product quality inspection risk assessment system and method based on measurement uncertainty to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: Product quality inspection risk assessment methods based on measurement uncertainty include: S1: Obtain the measurement value of each sample drawn in the sequential sampling inspection process of the product batch to be evaluated, and the current measurement uncertainty assessment value corresponding to each measurement value; S2: Based on the measurements of all extracted samples, determine whether the current test status meets the preset sequential decision boundary conditions; S3: If the test state does not meet the sequential decision boundary conditions, then perform pattern recognition analysis on the current measurement uncertainty assessment value sequence based on the statistical process control chart to determine whether the measurement process is under statistical control. S4: If the measurement process is under statistical control, then for each sample, calculate the membership value of the measured value to the preset qualified interval based on its measured value and the corresponding current measurement uncertainty assessment value. S5: Update the overall risk estimate of the product batch quality level based on the membership values ​​of all extracted samples; S6: Compare the updated overall risk estimate with the preset risk threshold, and make a decision on whether to continue sampling, accept the batch, or reject the batch based on the comparison result.

[0006] Furthermore, S1 includes: Receive the execution measurement command for the latest sample, and then control the measuring equipment to detect the latest sample according to the execution measurement command to obtain the raw detection data; The raw detection data is processed to obtain the measured values; The current measurement uncertainty assessment value corresponding to the measured value is obtained simultaneously based on the measurement condition parameters during the measurement and the current calibration status information of the measuring equipment.

[0007] Furthermore, S2 includes: Calculate the current sequential probability ratio statistic based on the measurements of all extracted samples; The sequential probability ratio statistic is compared with the preset acceptance and rejection thresholds; If the sequential probability ratio statistic is less than or equal to the acceptance threshold, it is determined that the acceptance decision boundary condition is met. If the sequential probability ratio statistic is greater than or equal to the rejection threshold, it is determined that the rejection decision boundary condition is met; otherwise, it is determined that the test state does not meet the sequential decision boundary condition.

[0008] Furthermore, S3 includes: Construct a sequence of current measurement uncertainty assessment values ​​based on the current measurement uncertainty assessment values; Based on the current measurement uncertainty assessment value sequence, construct single-value control charts and moving range control charts, and apply the preset statistical process control discrepancy rules to analyze the data point distribution patterns in the single-value control charts and moving range control charts; If the data point distribution pattern does not trigger any outlier detection rule, the measurement process is determined to be in statistical control; otherwise, the measurement process is determined not to be in statistical control.

[0009] Furthermore, S4 includes: Determine the lower and upper limits of the preset qualified range; Determine the width of the transition zone around the lower and upper limits based on the current measurement uncertainty assessment value; Based on the position of the measured value within the segmented interval formed by the lower limit, upper limit, and transition zone width, the membership value of the measured value to the preset qualified interval is calculated.

[0010] Furthermore, based on the position of the measured value within the segmented interval formed by the lower limit, upper limit, and transition zone width, the membership value of the measured value to the preset qualified interval is calculated as follows: when the measured value is less than or equal to the lower limit minus the transition zone width, the membership value is 0; when the measured value is greater than the lower limit minus the transition zone width and less than the lower limit, the membership value increases linearly from 0 to 1; when the measured value is between the lower limit and the upper limit, the membership value is 1; when the measured value is greater than the upper limit and less than the upper limit plus the transition zone width, the membership value decreases linearly from 1 to 0; when the measured value is greater than or equal to the upper limit plus the transition zone width, the membership value is 0.

[0011] Furthermore, S5 includes: Calculate the average membership values ​​of all extracted samples to characterize the central tendency of batch quality; Simultaneously, the standard deviation of the membership values ​​of all extracted samples is calculated to characterize the dispersion of batch quality; Based on the combination relationship between the mean and the standard deviation, the corresponding target risk mapping mode is selected from multiple preset risk mapping modes; By using the target risk mapping model to process the mean and standard deviation, an updated overall risk estimate of the product batch quality level is generated.

[0012] Furthermore, by using the target risk mapping model to process the mean and standard deviation, the updated overall risk estimate of the product batch quality level is generated by: taking the ratio of the mean to the standard deviation as input and querying a pre-established risk mapping table to obtain the updated overall risk estimate.

[0013] Furthermore, S6 includes: Calculate the absolute difference between the updated overall risk estimate and the preset risk threshold; The absolute difference is compared with a dynamically adjusted decision margin; If the updated overall risk estimate is less than the risk threshold and the absolute difference is greater than or equal to the decision margin, the batch is determined to be accepted. If the updated overall risk estimate is greater than the risk threshold and the absolute difference is greater than or equal to the decision margin, the batch is rejected; otherwise, sampling continues.

[0014] On the other hand, the present invention provides a product quality inspection risk assessment system based on measurement uncertainty, comprising: The data acquisition module is used to acquire the measurement value of each sample drawn in the sequential sampling inspection process of the product batch to be evaluated, as well as the current measurement uncertainty assessment value corresponding to each measurement value. The sequential boundary module is used to determine whether the current test status meets the preset sequential decision boundary conditions based on the measurement values ​​of all extracted samples. The control state identification module is used to perform pattern recognition analysis on the current measurement uncertainty assessment value sequence based on the statistical process control chart if the test state does not meet the sequential decision boundary conditions, in order to determine whether the measurement process is in a statistical control state. The membership calculation module is used to calculate the membership value of the measured value to the preset qualified interval for each sample, based on its measured value and the corresponding current measurement uncertainty assessment value, if the measurement process is under statistical control. The risk update module is used to update the overall risk estimate of the product batch quality level based on the membership values ​​of all extracted samples. The risk assessment module is used to compare the updated overall risk estimate with the preset risk threshold, and make a decision on whether to continue sampling, accept the batch, or reject the batch based on the comparison result.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. By acquiring and dynamically evaluating the uncertainty corresponding to each measurement and introducing it as a key input into the subsequent decision-making process, this method effectively overcomes the problem of inaccurate decision-making risk caused by the neglect of dynamic changes in measurement uncertainty in traditional sequential testing models. It establishes an evaluation framework that deeply integrates dynamic measurement uncertainty with sequential testing logic. By quantifying the reliability of each measurement in real time and using this as a basis to soften sample observations and dynamically estimate the overall risk of the batch, the judgment of the batch quality level at each decision point is no longer based on an idealized, fixed-error simple statistic, but on a comprehensive risk index that is closer to the actual measurement scenario and includes reliability weights. This significantly improves the authenticity of risk estimation and the rationality of decision-making at the model level.

[0016] 2. By performing statistical control state analysis on the measurement uncertainty sequence, the stability and reliability of the measurement data sequence used for risk assessment are first ensured from the data source, providing quality-controlled input for subsequent calculations. Membership values ​​are constructed using dynamic uncertainty, transforming the traditional hard binary decision of pass / fail into a continuous soft index reflecting the degree of compliance. This processing achieves a mathematical representation of the inherent fuzziness of measurement values ​​at the computer algorithm level, enabling more reasonable evaluation of sample data near specification boundaries. Overall risk estimation is performed based on the soft index of all samples, and a final decision is made through comparison with dynamic decision logic. This process essentially constructs an intelligent decision function that adapts to the inspection process and measurement data quality, systematically embedding dynamically changing measurement reliability information into each iteration of the sequential decision-making process. This allows the decision point for continuing sampling, acceptance, or rejection to be dynamically adjusted based on more comprehensive and realistic risk assessment results. Thus, at the level of computer-executed inspection decision-making procedures, the two major goals of misjudgment risk control and inspection resource conservation are more accurately balanced, improving the accuracy and robustness of the automated inspection decision-making system. Attached Figure Description

[0017] Figure 1 This is a flowchart of the product quality inspection risk assessment method based on measurement uncertainty according to the present invention; Figure 2 This is a schematic diagram of the product quality inspection risk assessment system based on measurement uncertainty according to the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Example 1: Figure 1 This invention presents a product quality inspection risk assessment method based on measurement uncertainty, comprising: S1: Obtain the measurement value of each sample drawn in the sequential sampling inspection process of the product batch to be evaluated, and the current measurement uncertainty assessment value corresponding to each measurement value; S2: Based on the measurements of all extracted samples, determine whether the current test status meets the preset sequential decision boundary conditions; S3: If the test state does not meet the sequential decision boundary conditions, then perform pattern recognition analysis on the current measurement uncertainty assessment value sequence based on the statistical process control chart to determine whether the measurement process is under statistical control. S4: If the measurement process is under statistical control, then for each sample, calculate the membership value of the measured value to the preset qualified interval based on its measured value and the corresponding current measurement uncertainty assessment value. S5: Update the overall risk estimate of the product batch quality level based on the membership values ​​of all extracted samples; S6: Compare the updated overall risk estimate with the preset risk threshold, and make a decision on whether to continue sampling, accept the batch, or reject the batch based on the comparison result.

[0020] S1: Obtain the measurement value of each sample drawn during the sequential sampling inspection of the product batch to be evaluated, as well as the current measurement uncertainty assessment value corresponding to each measurement value. Specifically, this is implemented as follows: In the sequential sampling inspection process, when it is necessary to test the latest sample in the batch of products to be evaluated, a measurement execution instruction for the latest sample will be received first. The measurement execution instruction can be manually triggered by the inspector or automatically generated by the automated inspection program after the sample is drawn according to the preset sampling plan. The measurement execution instruction includes at least the unique identification information of the sample and the quality characteristic item to be measured. The quality characteristic item is, for example, the fracture strength of the material or the life of the electronic component. After receiving the measurement execution instruction, the system or operator will start the measurement process for the latest sample.

[0021] Upon receiving a measurement execution command, the system or operator controls the corresponding measuring equipment to test the latest sample according to the content of the command, in order to obtain raw test data reflecting its quality characteristics. The type of measuring equipment depends on the quality characteristics to be measured. For example, a universal testing machine is used to measure strength, a high-precision caliper is used to measure dimensions, and a spectrometer is used to measure component content. The process of controlling the measuring equipment includes correctly installing or placing the latest sample in the designated position of the measuring equipment, setting the operating parameters of the measuring equipment, such as the loading rate of the testing machine and the integration time of the spectrometer, and then starting the measuring equipment. During operation, the measuring equipment will generate raw, unprocessed signals or readings. These signals or readings constitute the raw test data. Raw test data may be, for example, a curve of voltage changing over time, a spectrum of spectral intensity changing over wavelength, or a series of instantaneous force readings.

[0022] After obtaining the raw test data, specific data processing is required to transform it into final, interpretable measurements. The data processing method depends on the nature of the raw test data and the measurement principle. For example, if the raw test data is a stress-strain curve, the data processing may include identifying the maximum force point on the curve, dividing the maximum force by the original cross-sectional area of ​​the sample, and thus calculating the tensile strength as the measurement value. If the raw test data is a spectral signal, the data processing may include comparing the spectral intensity of the sample with the spectral intensity of a set of standard substances, and calculating the concentration of a specific component in the sample as the measurement value using a pre-established calibration curve or model. This process may involve various algorithms such as signal smoothing, peak finding, background subtraction, and curve fitting to ensure that the measurement value extracted from the raw test data accurately reflects the true characteristics of the sample.

[0023] In the process of acquiring the measured value, it is also necessary to evaluate the current measurement uncertainty corresponding to the measured value. This evaluation is based on the specific measurement condition parameters during the measurement and the current calibration status information of the measuring equipment. The measurement condition parameters refer to the quantified values ​​of various environmental and operational factors that may affect the measurement results at the moment or within a certain period of time when the latest sampled raw test data is obtained. The measurement condition parameters include at least the temperature value, relative humidity value, vibration level value of the measurement environment, and sample clamping position information recorded by the operator or automatically recorded by the sensor. The current calibration status information of the measuring equipment refers to the metrological characteristics of the measuring equipment itself at the time of this measurement. The current calibration status information of the measuring equipment includes at least the calibration correction value and its corresponding calibration uncertainty given on the most recent certified calibration certificate of the measuring equipment at the relevant measurement points, as well as the key performance indicators of the measuring equipment since the last calibration. When monitoring historical data and assessing key performance indicators such as resolution, repeatability, and stability, the evaluation of the current measurement uncertainty must comprehensively consider the impact of deviations from standard conditions in measurement parameters, as well as the uncertainty introduced by the measurement equipment itself. This should be done using general uncertainty assessment methods. For example, by analyzing the sensitivity coefficient of temperature fluctuations on material size measurement, combined with the temperature change range recorded during the measurement, the standard uncertainty component introduced by temperature is calculated. This temperature-introduced standard uncertainty component is then combined with the standard uncertainty component introduced by the equipment indication error obtained from the calibration certificate, and the component introduced by the equipment resolution, according to the variance synthesis principle. The resulting expanded uncertainty of the current measurement value is used as the current measurement uncertainty evaluation value. This value is represented by a positive number and includes confidence probability information, such as the expanded uncertainty at a 95% confidence level, thus providing quantitative information about the reliability of the measurement value for subsequent decision-making.

[0024] S2: Based on the measurements of all extracted samples, determine whether the current test status meets the preset sequential decision boundary conditions. Specifically, this is implemented as follows: In sequential sampling inspection, after obtaining the measurements of all samples taken from the batch of products to be evaluated, it is necessary to determine whether the current inspection status meets the preset sequential decision boundary conditions based on the measurements of all samples taken. The core of this process is to calculate and evaluate an indicator reflecting the cumulative strength of evidence, namely the sequential probability ratio statistic. The calculation of the sequential probability ratio statistic is not arbitrary; it strictly relies on two clear and opposing statistical assumptions made about the product quality distribution during the inspection plan design stage. The first statistical assumption represents that the quality level of the product batch is in an acceptable state, that is, it meets the acceptable quality level. Acceptable quality level is a specific numerical value. For example, the acceptable quality level of the yield strength of a certain metal material is 500 MPa. It represents the lower limit of quality that the manufacturer is willing to guarantee and the user agrees to accept. The second statistical assumption represents that the quality level of the product batch is in an unacceptable state, that is, it meets the limit quality level. The limit quality level is another specific numerical value. For example, the limit quality level of the yield strength of the same metal material is 450 MPa. It represents the upper limit of quality that the user can tolerate in order to control its own risks. These two assumptions must be determined in advance based on the product technical specifications and the risk agreement between the supplier and the user to provide a benchmark for subsequent probability calculations.

[0025] The specific operation for calculating the sequential probability ratio statistic is as follows: For the i-th sample already drawn, its measured value is denoted as xi. It is necessary to calculate the probability of this measured value xi occurring under two different statistical hypotheses. First, assuming the product quality level equals the acceptable quality level, calculate the probability density or probability of the measured value xi occurring under that hypothesis, based on the probability distribution type followed by the measured value. If the measured value is a continuous variable and follows a normal distribution, use the normal distribution probability density function with the acceptable quality level as the mean and the known or estimated process standard deviation as the standard deviation. If the measured value is a discrete pass / fail judgment, following a binomial distribution, use the binomial distribution probability quality function with the acceptable quality level as the pass rate. Denote this calculation result as LAi, which represents the probability that the data supports a good batch at the i-th sample. Next, assuming the product quality level equals the limit quality level, use the same probability distribution type, but replace the distribution parameters with those corresponding to the limit quality level. Calculate the probability density or probability of the measured value xi under this hypothesis, and denote the result as LRi. It represents the probability that the data supports the batch being bad at the i-th sample. Then, calculate the individual likelihood ratio of the i-th sample. The individual likelihood ratio is equal to LRi divided by LAi, which is the ratio of the probability of supporting the bad hypothesis to the probability of supporting the good hypothesis. For the first sample that has been drawn, its individual likelihood ratio is the current sequential probability ratio statistic. For subsequent samples, for example, after drawing n samples, the current sequential probability ratio statistic is the continuous product of the individual likelihood ratios of these n samples. The calculation method is to multiply the individual likelihood ratio of the first sample by the individual likelihood ratio of the second sample, then by the individual likelihood ratio of the third sample, and so on, until multiplying by the individual likelihood ratio of the n-th sample. The result of this continuous product is the sequential probability ratio statistic up to the n-th sample. It quantifies the strength of evidence that all the observed data to date support the limit quality level hypothesis relative to the acceptable quality level hypothesis.

[0026] After obtaining the sequential probability ratio statistic, it needs to be compared with two pre-calculated, fixed numerical limits: a preset acceptance threshold and a preset rejection threshold. These thresholds are not arbitrarily chosen; their values ​​stem from strict control requirements for decision-making risk. Decision-making risk includes producer risk and consumer risk. Producer risk refers to the probability that the inspection procedure incorrectly rejects a batch when its actual quality level equals the acceptable quality level—a low-probability event that needs to be limited. Consumer risk refers to the probability that the inspection procedure incorrectly accepts a batch when its actual quality level equals the limit quality level—again, a low-probability event that needs to be limited. In the sequential probability... In the comparison test theory, the pre-set acceptance threshold can be calculated using the formula: Pre-set acceptance threshold = User risk / (1 - Producer risk), and the pre-set rejection threshold can be calculated using the formula: Pre-set rejection threshold = (1 - User risk) / Producer risk. For example, if both parties agree that the producer risk is 0.05 and the user risk is 0.10, then substituting into the formula, the acceptance threshold = 0.10 / (1 - 0.05) = 0.10 / 0.95 ≈ 0.1053, and the rejection threshold = (1 - 0.10) / 0.05 = 0.90 / 0.05 = 18.00. This yields two specific numerical limits, where the pre-set acceptance threshold is a positive number less than 1, and the pre-set rejection threshold is a positive number much greater than 1. These limits divide the data into three regions on the numerical axis.

[0027] The comparison and judgment process is carried out in a strict logical order. The first step is to compare the calculated sequential probability ratio statistic with the preset acceptance threshold and determine whether the sequential probability ratio statistic is less than or equal to the preset acceptance threshold. If the sequential probability ratio statistic is less than or equal to the preset acceptance threshold, the current test status is determined to meet the acceptance decision boundary conditions. The satisfaction of the acceptance decision boundary conditions is a clear signal that the accumulated evidence strongly tends to support that the quality of the product batch has reached an acceptable quality level, and the user risk borne by making the acceptance decision is controlled below the pre-agreed level. At this time, sampling can be stopped and a decision to accept the product batch can be made.

[0028] If the sequential probability ratio statistic is not less than or equal to the preset acceptance threshold (i.e., the sequential probability ratio statistic is greater than the preset acceptance threshold), then proceed to the second step of comparison. Compare the sequential probability ratio statistic with the preset rejection threshold to determine whether the sequential probability ratio statistic is greater than or equal to the preset rejection threshold. If the sequential probability ratio statistic is greater than or equal to the preset rejection threshold, then the current inspection status is determined to meet the rejection decision boundary conditions. The satisfaction of the rejection decision boundary conditions is another clear signal, indicating that the accumulated evidence strongly tends to support that the quality of the product batch is only at the limit quality level or even worse, and the producer risk borne by making the rejection decision is controlled below the pre-agreed level. At this time, sampling can be stopped and a decision to reject the product batch can be made.

[0029] If, after the above two comparisons, the sequential probability ratio statistic neither meets the condition of being less than or equal to the preset acceptance threshold nor the condition of being greater than or equal to the preset rejection threshold, this means that the value of the sequential probability ratio statistic strictly falls between the preset acceptance threshold and the preset rejection threshold, being greater than the preset acceptance threshold and less than the preset rejection threshold. In this case, it is determined that the test status does not meet the sequential decision boundary conditions. The sequential decision boundary conditions are a collective term for the acceptance decision boundary conditions and the rejection decision boundary conditions. When the test status does not meet the sequential decision boundary conditions, it indicates that the strength of evidence derived from the measurements of all currently drawn samples is neither sufficient to safely accept the batch while controlling the risk, nor sufficient to control the risk while controlling the risk. Under the premise of controlling risk, if a batch is explicitly rejected, the inspection status is in the gray area of ​​continuing sampling. At this time, the inspection process will not make a final judgment, but will instruct that the next sample should be drawn from the product batch according to the sequential sampling plan. After obtaining the measurement value of the new sample, it will be included in the calculation, the sequential probability ratio statistic will be updated, and the above comparison and judgment process will be repeated. This process will continue until the sequential probability ratio statistic of a certain sample point falls into the acceptance area or rejection area, thereby satisfying the acceptance decision boundary condition or the rejection decision boundary condition. This dynamic mechanism, which relies on cumulative evidence to make decisions step by step, is the mathematical basis for sequential sampling inspection to minimize the average inspection sample size while ensuring that the risk is controlled.

[0030] S3: If the test state does not meet the sequential decision boundary conditions, then perform pattern recognition analysis on the current measurement uncertainty assessment value sequence based on the statistical process control chart to determine whether the measurement process is under statistical control. Specifically, the implementation is as follows: In the sequential sampling inspection process, when the sequential probability ratio statistic calculated based on the measurements of all sampled samples is between the preset acceptance threshold and the preset rejection threshold, i.e., the inspection status does not meet the boundary conditions of the sequential decision, it indicates that the conventional sequential inspection decision logic cannot draw a clear conclusion under the current sample size. At this point, the process enters the diagnostic stage of the stability of the measurement process itself. The core of this stage is to analyze the change pattern of the measurement uncertainty assessment value accompanying each measurement in order to determine whether the measurement process that produces these measurements is in a stable statistical control state that is only affected by random factors. This is because an unstable measurement process will directly lead to the distortion of the measurement value and its uncertainty assessment value, thereby affecting the reliability of the batch quality risk assessment based on these values.

[0031] The first step in determining whether a measurement process is under statistical control is to construct a sequence of current measurement uncertainty assessment values ​​based on the current measurement uncertainty assessment values. This sequence is a data set arranged in chronological order, where each data element is the current measurement uncertainty assessment value obtained synchronously after each measurement of the latest sample during the sequential sampling inspection process. For example, after the inspection starts, the current measurement uncertainty assessment value obtained from the first sampling measurement is denoted as U1, the second as U2, the third as U3, and so on, until the current measurement uncertainty assessment value obtained from the most recent m-th measurement is denoted as Um. Arranging U1, U2, U3, and so on, in the order they were generated, constitutes a sequence of current measurement uncertainty assessment values ​​containing m data points. This sequence reflects the dynamic changes in the claimed measurement reliability index of each specific measurement activity from the start of the inspection to the current moment, and serves as the basic data source for subsequent statistical analysis.

[0032] After constructing the current measurement uncertainty assessment value sequence, two types of statistical process control charts need to be built based on this sequence for pattern recognition analysis. These two types of control charts are individual value control charts and moving range control charts. When constructing an individual value control chart, each data point in the current measurement uncertainty assessment value sequence is directly plotted on a graph in its order of occurrence. The vertical axis of the graph represents the numerical magnitude of the current measurement uncertainty assessment value, and the horizontal axis represents the sampling order or time series. Next, the arithmetic mean of all data points in the current measurement uncertainty assessment value sequence needs to be calculated. This arithmetic mean is called the center line of the individual value control chart, which represents the long-term trend of the current measurement uncertainty assessment value. The average level of fluctuation is calculated first. Then, the control limits of the single-value control chart are calculated. The upper and lower control limits are obtained by adding an offset to the center line and subtracting an offset from the center line, respectively. This offset is determined by the degree of fluctuation of the current measurement uncertainty assessment value sequence. Specifically, the moving range sequence of the current measurement uncertainty assessment value sequence needs to be calculated first. The moving range refers to the absolute value of the difference between two adjacent data points in the sequence. For example, the absolute value of the difference between U2 and U1 is the first moving range R1, the absolute value of the difference between U3 and U2 is the second moving range R2, and so on, to obtain the moving range sequence R1, R2, ..., R{m-1}. The values ​​of all moving ranges are then calculated. The arithmetic mean is used to obtain the moving range average. Finally, the upper control limit of the individual control chart is equal to the center line plus a coefficient multiplied by the moving range average, and the lower control limit is equal to the center line minus the same coefficient multiplied by the moving range average. This coefficient is a statistical constant, the value of which depends on the proportion of random fluctuations that are desired to be included in the control limits. For example, a value of 2.66 is often used. This value of 2.66 comes from the derivation of the relationship between individual observations and moving range in statistical theory and is used to convert the moving range average into an appropriate control limit width. When constructing the moving range control chart, the calculated moving range sequence R1, R2, ..., R{m-1} is directly plotted. On another graph, the vertical axis represents the magnitude of the moving range, and the horizontal axis represents the order. The center line of the moving range control chart is the average moving range. The upper control limit of the moving range control chart is obtained by multiplying the average moving range by a coefficient, for example, a coefficient often taken as 3.27. This value of 3.27 also comes from the calculation of the control limit of the range distribution in statistical theory. The lower control limit of the moving range control chart is theoretically 0, but it is usually not plotted or regarded as 0 in actual control charts. Thus, the single-value control chart is used to monitor the level and deviation of the current measurement uncertainty assessment value itself, while the moving range control chart is used to monitor the dispersion of the successive fluctuations of the current measurement uncertainty assessment value.

[0033] After the control charts are constructed, it is necessary to apply preset statistical process control outlier detection rules to analyze the data point distribution patterns in the single-value control chart and the moving range control chart. These preset rules are a set of predefined logical criteria used to identify non-random fluctuation patterns. Based on probability theory, when the process is fully controlled, the probability of data points violating these rules is very small. Therefore, once triggered, they indicate a possible abnormality in the process. Common preset statistical process control outlier detection rules include, but are not limited to, the following: The first rule is that any single data point falls above or below the upper control limit of the single-value control chart or the moving range control chart. The second rule is that in a single-value control chart, nine consecutive data points fall on the same side of the center line. The third rule is that in a single-value control chart, six consecutive data points show a continuous upward or downward trend. The fourth rule is that in a single-value control chart, fourteen consecutive data points oscillate alternately up and down. The fifth rule is that in a single-value control chart, two out of three consecutive data points fall on the same side of the center line and are more than [a certain distance] from the center line. For areas outside twice the standard deviation, the estimated standard deviation can be obtained by dividing the moving range average by a transformation constant, such as 1.128. The sixth rule is that in a single-value control chart, four out of five consecutive data points fall on the same side of the center line and are more than one standard deviation away from the center line. The seventh rule is that in a single-value control chart, all fifteen consecutive data points fall within one standard deviation on both sides of the center line. The eighth rule is that in a single-value control chart, all eight consecutive data points fall on both sides of the center line and none of them are within one standard deviation. During analysis, it is necessary to check the overall distribution from the first data point to the last data point in both the single-value control chart and the moving range control chart to see if any of the above-mentioned preset statistical process control outlier rules are triggered. This can be done automatically by a computer algorithm that compares the position of each data point with the pre-calculated boundaries of various rules. For example, the algorithm will check whether each point is out of bounds and check whether each continuous subsequence of data points of a specified length conforms to a specific trend or clustering pattern.

[0034] After checking all preset statistical process control anomaly detection rules, a final judgment can be made. If, in the single-value control chart and moving range control chart, the distribution pattern of all data points strictly adheres to the characteristics of random fluctuations and no preset statistical process control anomaly detection rule is triggered, then the measurement process is judged to be in a statistical control state. A statistical control state means that various factors affecting the current measurement uncertainty assessment value during the measurement process, such as fluctuations in environmental conditions, slight changes in equipment performance, and minor differences in operation, are random and their combined effect is stable. The measurement system itself has not undergone significant, attributable abnormal changes. At this time, the measurement process can be considered reliable, and its output measurement value and the corresponding current measurement uncertainty assessment value can be used for subsequent risk assessment. Conversely, if, in the single-value control chart or moving range control chart, the distribution pattern of all data points strictly adheres to the characteristics of random fluctuations and no preset statistical process control anomaly detection rule is triggered, then the measurement process is judged to be in a statistical control state. The distribution pattern of data points triggers at least one of the preset statistical process control anomaly detection rules. For example, if a current measurement uncertainty assessment value is found to exceed the upper control limit of the moving range control chart, or if seven consecutive points in the single-value control chart show an upward trend, it is determined that the measurement process is not under statistical control. The measurement process not being under statistical control is an early warning signal, indicating that the measurement process may be disturbed by some specific abnormal factor, such as a sudden drift of the measuring equipment, an uncontrolled step change in the ambient temperature, or a change in the operator who did not follow the procedures. Under such circumstances, the reliability of the current and recent measurement values ​​and their current measurement uncertainty assessment values ​​is questionable, and therefore they are not suitable for direct use in subsequent rigorous product batch quality risk assessments. It is necessary to first investigate and eliminate the abnormal causes of the measurement process.

[0035] S4: If the measurement process is under statistical control, then for each sample, based on its measured value and the corresponding current measurement uncertainty assessment value, calculate the membership value of the measured value to the preset qualified interval. Specifically, the implementation is as follows: In the sequential sampling inspection process, if the control chart constructed from the current measurement uncertainty assessment value sequence is analyzed and it is determined that the measurement process is under statistical control, it indicates that the measurement system itself is stable and reliable. At this time, the measured value and its corresponding current measurement uncertainty assessment value are credible and can be used for more refined product batch quality level assessment. The next step is to calculate a value between 0 and 1 for each sampled sample, based on its measured value and the corresponding current measurement uncertainty assessment value. This value represents the membership degree of the measured value to the preset qualified interval. The membership degree value is used to quantify the degree to which the measurement result of a single sample meets the quality requirements. A value of 1 represents complete compliance, a value of 0 represents complete non-compliance, and a value between 0 and 1 represents a certain degree of fuzziness or uncertainty in the degree of compliance. This conversion process is to transform the explicit measured value containing measurement error into a soft index that can reflect its qualification probability, laying the foundation for subsequent integration of multiple sample information for overall risk assessment.

[0036] The first step in calculating the membership value of a measured value to a preset acceptable range is to determine the lower and upper limits of the preset acceptable range. A preset acceptable range is a numerical range defined in advance according to product quality standards or technical specifications. It represents the range within which a single quality characteristic of a product must fall to be deemed acceptable. The lower limit of the preset acceptable range is a specific value, such as 4.82 mm for the thread pitch diameter of a certain fastener. The upper limit of the preset acceptable range is another specific value, such as 4.97 mm for the thread pitch diameter of the same type of fastener. These two values ​​are usually derived from product design drawings, technical agreements, or applicable product standards. They are fixed technical requirements determined before the inspection begins and serve as the absolute basis for judging whether each sample is acceptable throughout the inspection process. Determining the lower and upper limits of the preset acceptable range means reading these two pre-stored values ​​from storage devices or input interfaces.

[0037] After determining the lower and upper limits of the preset acceptable range, a key parameter needs to be determined based on the current measurement uncertainty assessment value: the width of the transition zone around the lower and upper limits. The transition zone width is a length quantity, and its value is directly related to the current measurement uncertainty assessment value. The reason for introducing the transition zone width is that any measurement has uncertainty. When a measured value is very close to the acceptable limit, its true value may fall on the other side of the limit due to measurement errors. Therefore, it is not appropriate to arbitrarily determine it as completely acceptable or completely unacceptable. Instead, a transition zone is needed to gently express this possibility of being in a boundary state. The method for determining the transition zone width is to multiply the current measurement uncertainty assessment value by a proportionality coefficient greater than 0. The specific value of the proportionality coefficient needs to be preset by the implementer based on the balance between the control requirements for measurement risk and the strictness of decision-making. For example, the proportionality coefficient can be set to 2, or the proportionality coefficient can be set to... Setting the proportionality coefficient to 3 means that the transition zone width is equal to twice the current measurement uncertainty assessment value, and setting it to 3 means that the transition zone width is equal to three times the current measurement uncertainty assessment value. The principle for setting the proportionality coefficient is that the larger the proportionality coefficient, the wider the transition zone, and the more moderate and ambiguous the judgment of the measured value near the qualified boundary, which can more fully accommodate the influence of measurement error. The smaller the proportionality coefficient, the narrower the transition zone, and the more stringent and clear the judgment, which is closer to the traditional hard qualified judgment. For example, if the current measurement uncertainty assessment value of a certain measured value is 0.02 mm and the preset proportionality coefficient is 2, then the calculated transition zone width is 0.02 mm × 2 = 0.04 mm. This transition zone width will be applied to both the lower limit and the upper limit of the preset qualified interval, that is, a transition area with a width of 0.04 mm will be established on the left side of the lower limit and the right side of the upper limit.

[0038] After obtaining the lower limit, upper limit, and transition zone width of the preset qualified interval, the membership value of each sample measurement value to the preset qualified interval can be calculated using a piecewise function based on the specific position of the measurement value within the segmented interval formed by the lower limit, upper limit, and transition zone width. The segmentation of the interval is determined by the following five critical points: the first critical point is the lower limit minus the transition zone width; the second critical point is the lower limit itself; the third critical point is the upper limit itself; and the fourth critical point is the upper limit plus the transition zone width. These five critical points divide the entire real number axis into five continuous regions. The calculation logic is as follows: when the measurement value is less than or equal to the first critical point (lower limit minus transition zone width), it indicates that the... Even considering its maximum possible positive measurement error, the measured value is still significantly lower than the acceptable lower limit, and therefore is judged as completely unacceptable, with its corresponding membership value assigned to 0. When the measured value is greater than the first critical lower limit minus the transition zone width, and simultaneously less than the second critical lower limit, it indicates that the measured value falls within the transition zone to the left of the lower limit value. Its true value may be acceptable or unacceptable. In this case, linear interpolation is required to continuously increase the membership value from 0 to 1. The specific formula for linear increase is: the membership value equals the measured value minus the lower limit value (in parentheses) minus the transition zone width, and then the difference is divided by the transition zone width. For example, if the lower limit value is 4.82 mm and the transition zone width is 0.04 mm... If the measured value is 4.80 mm, the calculation process is as follows: First, calculate the lower limit value minus the transition zone width: 4.82 mm - 0.04 mm = 4.78 mm. Then, calculate the difference between the measured value and this difference: 4.80 mm - 4.78 mm = 0.02 mm. Finally, divide 0.02 mm by the transition zone width of 0.04 mm to obtain a membership value of 0.5. When the measured value is greater than or equal to the second critical point lower limit value and simultaneously less than or equal to the third critical point upper limit value, it indicates that the measured value is clearly within the preset qualified range and is judged as fully qualified, and its corresponding membership value is assigned a value of 1. When the measured value is greater than the third critical point upper limit value and simultaneously less than the fourth critical point upper limit value plus the transition zone width, it indicates that the measured value is not qualified. If the measured value falls within the transition zone to the right of the upper limit, linear interpolation is required to continuously decrease the membership value from 1 to 0. The specific formula for linear decrease is as follows: the membership value equals the sum of the upper limit value (within parentheses) and the width of the transition zone, minus the measured value. Then, the difference is divided by the width of the transition zone. For example, if the upper limit value is 4.97 mm, the width of the transition zone is 0.04 mm, and the measured value is 4.98 mm, the calculation process is as follows: first, calculate the upper limit value plus the width of the transition zone: 4.97 mm + 0.04 mm = 5.01 mm; then, calculate the difference between this sum and the measured value: 5.01 mm - 4.98 mm = 0.03 mm; finally, divide 0.03 mm by the width of the transition zone (0.04 mm) to obtain a membership value of 0.75. When the measured value is greater than or equal to the upper limit of the fourth critical point plus the width of the transition zone, it indicates that even considering its maximum possible negative measurement error, the measured value is still significantly higher than the acceptable upper limit. Therefore, it is judged as completely unacceptable, and its corresponding membership value is assigned to 0. Through the above segmented calculation, the measured value of each sampled item is mapped to a membership value between 0 and 1. This value not only reflects the magnitude of the measured value itself but also incorporates uncertainty assessment information for that specific measurement. This achieves a softening of the traditional binary acceptance / unacceptance judgment, allowing measured values ​​on the borderline of acceptance to receive a more reasonable and quantitative evaluation that conforms to metrological principles.

[0039] S5: Based on the membership values ​​of all extracted samples, update the overall risk estimate of the product batch quality level. Specifically, this is implemented as follows: In the sequential sampling inspection process, after calculating the membership values ​​of each measured value with respect to the preset acceptable range for all sampled samples, it is necessary to update the overall risk estimate of the product batch quality level based on the membership values ​​of all sampled samples. The overall risk estimate is a comprehensive quantitative indicator used to characterize the probability of judging the product batch as unqualified or having quality risks under the current sample evidence. Its value range is usually between 0 and 1. The closer the value is to 1, the higher the risk and the greater the possibility that the batch will be judged as unqualified. The closer the value is to 0, the lower the risk and the greater the possibility that the batch will be judged as qualified. The purpose of updating is to integrate the soft compliance evaluations of multiple samples, which contain uncertainty information, into a single risk probability value that can be used for the final decision.

[0040] The first step in updating the overall risk estimate is to calculate the average membership values ​​of all sampled samples. The average is a core statistic characterizing the central tendency of batch quality. It is calculated by first summing the membership values ​​of all sampled samples, then dividing this sum by the total number of sampled samples. The quotient is the average membership value. For example, assuming five samples have been drawn with membership values ​​of 0.8, 1.0, 0.6, 1.0, and 0.9, the sum of membership values ​​is 0.8 + 1.0 + 0.6 + 1.0 + 0.9 = 4.3. Dividing this sum by the sample size of 5 gives an average of 0.86. This average of 0.86 reflects the average passability of the tested samples from the perspective of membership. The closer the average is to 1, the more likely the sample is to pass; the closer the average is to 0, the more likely the sample is to fail.

[0041] While calculating the mean, it is also necessary to calculate the standard deviation of the membership values ​​of all sampled items. The standard deviation is a core statistic characterizing the dispersion of batch quality; it reflects the degree to which the membership values ​​of each sample fluctuate around the mean. The method for calculating the standard deviation is as follows: First, calculate the difference between the membership value of each sample and the mean calculated in the previous step; this difference is called the deviation. Then, square each deviation value to obtain the squared deviation. Next, sum the squared deviations of all samples to obtain the sum of squares. Then, divide this sum of squares by the sample size minus 1 to obtain an intermediate quantity called the variance. Finally, take the square root of the variance; the positive square root is the standard deviation of the membership values. Continuing with the example above, the mean is 0.86, and the membership values ​​of the five samples are... The deviations for values ​​of 0.8, 1.0, 0.6, 1.0, and 0.9 are -0.06, 0.14, -0.26, 0.14, and 0.04, respectively. The squared deviations are 0.0036, 0.0196, 0.0676, 0.0196, and 0.0016, respectively. The sum of squares is 0.0036 + 0.0196 + 0.0676 + 0.0196 + 0.0016 = 0.112. The variance is 0.112 / 4 = 0.028. The standard deviation is the square root of 0.028, approximately equal to 0.1673. A larger standard deviation indicates poorer consistency in quality among samples, with some samples being very acceptable while others are unacceptable, and a higher risk of uneven batch quality. A smaller standard deviation indicates better consistency in quality among samples, and more uniform batch quality.

[0042] After obtaining the mean and standard deviation, it is necessary to select the corresponding target risk mapping mode from several preset risk mapping modes based on the combination relationship between the mean and standard deviation. These preset risk mapping modes are several sets of different logical rules or functional relationships predefined before implementing this method. Each set of rules or functional relationships corresponds to a specific batch quality status pattern. The purpose is to ensure that when converting the statistical characteristics, mean and standard deviation, into risk estimates, they better reflect the actual risk under different quality distribution patterns. The process of selecting the target risk mapping mode is a condition-based classification process. The classification condition is the combination relationship between the mean and standard deviation. For example, three risk mapping modes can be predefined. The first risk mapping mode is suitable for high-quality batches. The first risk mapping mode is suitable for uniform quality, where the average value is greater than a quality-excellent threshold and the standard deviation is less than a good uniformity threshold. The second risk mapping mode is suitable for poor quality and uniform quality, where the average value is less than a quality-inferior threshold and the standard deviation is less than a good uniformity threshold. The third risk mapping mode is suitable for unstable quality, where the standard deviation is greater than a uniformity deterioration threshold regardless of the average value. The specific values ​​of these thresholds need to be preset based on the quality requirements of the specific product and historical experience data. Setting these thresholds is a configuration activity based on a balance between historical data analysis and risk tolerance. Specifically, it involves collecting data on the product... Based on historical data of membership values ​​from a large number of inspected batches over a past period, the mean and standard deviation of the membership values ​​for each historical batch are calculated. Then, according to the final actual quality assessment results of the batch, the historical batches are categorized into high-quality batches, low-quality batches, and unstable-quality batches. Next, the distribution of the mean values ​​of high-quality batches is analyzed, and a value is selected such that the mean of the vast majority of high-quality batches is higher than this value. This value is set as the high-quality assessment threshold. Similarly, the distribution of the mean values ​​of low-quality batches is analyzed, and a value is selected such that the mean of the vast majority of low-quality batches is lower than this value. This value is set as the low-quality assessment threshold. The standard deviation of the membership values ​​for all historical batches is then analyzed. The distribution of quality in the uniformity distribution is analyzed, and combined with the standard deviation characteristics of batches with unstable quality, a value is selected such that the standard deviation of the vast majority of batches with unstable quality is higher than this value, while the standard deviation of stable batches is lower than this value. This value is set as the threshold for judging uniformity deterioration. The threshold for judging good uniformity can be set as a value lower than the threshold for judging uniformity deterioration, for example, multiplying the threshold for judging uniformity deterioration by 0.6. By setting each judgment threshold using this method based on the statistical quantiles of historical data, the selection of thresholds is supported by objective data rather than subjective assumptions. For example, by analyzing historical data, the threshold for judging excellent quality may be set at 0.9, the threshold for judging poor quality at 0.4, and the threshold for judging uniformity deterioration at 0.25. Setting the threshold for good uniformity to 0.15, the calculated mean and standard deviation are compared to these set values. If the mean is greater than 0.9 and the standard deviation is less than 0.15, the first risk mapping mode is selected as the target risk mapping mode. If the mean is less than 0.4 and the standard deviation is less than 0.15, the second risk mapping mode is selected. If the standard deviation is greater than 0.25, the third risk mapping mode is selected regardless of the mean. If the combination of mean and standard deviation does not meet any of the above conditions (e.g., the mean is between 0.4 and 0.9 and the standard deviation is between 0.15 and 0.25), a default fourth risk mapping mode can be defined as the target risk mapping mode. Through this conditional judgment, the most suitable risk transformation logic can be matched to the quality status of the current batch.

[0043] After selecting the target risk mapping model, the next step is to use it to process the mean and standard deviation, generating an updated overall risk estimate of the product batch quality level. According to a specific implementation method described in the weighted description, the core operation of the target risk mapping model is to calculate the ratio of the mean to the standard deviation. This ratio is then used as input to obtain the updated overall risk estimate by querying a pre-established risk mapping table. The ratio of the mean to the standard deviation is a dimensionless number that comprehensively reflects the relative relationship between the central position and dispersion of batch quality. A larger ratio indicates a higher average pass rate and relatively smaller fluctuations, suggesting a lower risk; a smaller ratio indicates a lower average pass rate or more volatile fluctuations, suggesting a higher risk. The pre-established risk mapping table... A risk mapping table is a static data lookup table consisting of two columns. The first column is a pre-set series of continuous or discrete ratios of the mean to the standard deviation, and the second column is the corresponding pre-set population risk estimate. The construction of the risk mapping table requires historical data simulation, expert experience, or theoretical model derivation. For example, computer simulation can be used to simulate a large number of product batches with different actual quality levels. Virtual sequential sampling is performed on each simulated batch, and the membership sequence is calculated to obtain the ratio of its mean to the standard deviation. Simultaneously, the actual quality status of the simulated batch is recorded as qualified. By statistically analyzing the proportion of unqualified simulated batches within each ratio interval, this proportion is used as the corresponding population for that ratio interval. The risk estimate is used to systematically populate the risk mapping table. During a query, the calculated ratio of the current mean to standard deviation is compared with the ratio input value in the first column of the risk mapping table. The closest ratio input value is found, or the ratio interval falls within that range. Then, the overall risk estimate value in the second column corresponding to that row or interval is read as the updated overall risk estimate value for the product batch quality level. For example, assuming the calculated ratio of the mean to standard deviation is 5.2, querying the risk mapping table reveals that the overall risk estimate value corresponding to a ratio input value of 5.0 is 0.05, and the overall risk estimate value corresponding to a ratio input value of 5.5 is 0.04. Since 5.2 falls between 5.0 and 5.5, it can be used as the overall risk estimate value for the product batch quality level. A more accurate value is obtained through linear interpolation. The linear interpolation method is as follows: First, calculate the difference between the ratio to be interpolated and the lower limit ratio, i.e., 5.2-5.0=0.2; then calculate the difference between the upper limit ratio and the lower limit ratio, i.e., 5.5-5.0=0.5; next, calculate the difference between the upper limit risk and the lower limit risk, i.e., 0.04-0.05=-0.01; finally, divide the first difference 0.2 by the second difference 0.5, and then multiply by the third difference -0.01 to obtain the specific value of the adjustment item as (0.2 / 0.5)×(-0.01); add the adjustment item to the lower limit risk 0.05, i.e., 0.05+(0.2 / 0.5)×(-0.01), and calculate the updated overall risk estimate of approximately 0.046.By combining table lookup and interpolation, the statistics are transformed into an intuitive risk probability, completing the final extraction of decision risk information from sample data. This updated overall risk estimate will be used in the subsequent final judgment steps.

[0044] S6: Compare the updated overall risk estimate with the preset risk threshold, and determine whether to continue sampling, accept the batch, or reject the batch based on the comparison result. The specific implementation is as follows: In the sequential sampling inspection process, after the overall risk estimate of the product batch quality level is updated based on the membership values ​​of all sampled samples, the process enters the final decision-making stage. The core of this stage is to compare the updated overall risk estimate with a pre-set decision criterion and make a clear judgment on one of three options: continue sampling, accept the batch, or reject the batch, based on the comparison result. The purpose of this step is to establish a clear decision rule between dynamically changing inspection evidence and fixed risk control standards, thereby terminating the inspection process or instructing it to continue.

[0045] The first step in making a judgment is to calculate the absolute difference between the updated overall risk estimate and the preset risk threshold. The preset risk threshold is a fixed probability value determined before the inspection begins. It represents the upper limit of batch quality risk that the decision-maker can tolerate. If the overall risk estimate is lower than this threshold, the batch quality risk is considered controllable, and acceptance is preferred; if it is higher than this threshold, the risk is considered unacceptable, and rejection is preferred. The preset risk threshold is usually set based on a comprehensive analysis of product quality requirements, risk-sharing agreements between suppliers and buyers, and historical quality loss data. The specific setting process is as follows: Collect a large number of historical product batch final quality arbitration results and actual economic loss data caused by accepting non-conforming batches or rejecting conforming batches. For each historical batch, its true quality status (conforming or non-conforming) and the associated financial losses can be clearly identified. For example, accepting a non-conforming batch may result in... Losses such as production line downtime and customer claims are recorded as Loss A; losses such as rework and delivery delays due to rejecting a qualified batch are recorded as Loss B. Using statistical models, such as decision analysis models, the average expected loss resulting from long-term inspection decisions under different preset risk thresholds is calculated. The risk probability value that minimizes this average expected loss is selected as the final preset risk threshold. This quantitative method based on historical data and economic analysis ensures that the setting of the preset risk threshold has an objective basis, rather than subjective guesswork. Once set, the preset risk threshold remains unchanged throughout the inspection process. When calculating the absolute difference, a numerical comparison is first performed to determine whether the updated overall risk estimate is larger or smaller than the preset risk threshold. Then, the smaller value is subtracted from the larger value to obtain a non-negative value, which is the absolute difference. This value quantifies the distance between the current risk assessment result and the decision boundary.

[0046] After obtaining the absolute difference, it needs to be compared with a dynamically adjusted decision margin. The dynamically adjusted decision margin is a value that changes as the testing progresses, specifically as the number of samples drawn increases. Its core design idea is that in the early stages of testing, when the sample size is small, the evaluation results fluctuate greatly; therefore, a wider decision margin is set to avoid premature erroneous decisions. As the sample size increases, the evaluation results tend to stabilize, and the decision margin narrows accordingly, making the decision more sensitive. The specific implementation of the dynamically adjusted decision margin requires a predefined mathematical relationship or mapping rule to map the current number of samples drawn to the decision margin value. One implementation method is to use a reciprocal function relationship, that is, the dynamically adjusted decision margin equals a preset base margin constant divided by the current number of samples drawn. Mathematically, this is expressed as: Dynamically Adjusted Decision Margin = Base Margin Constant / Current Number of Samples Drawn, where the base margin constant is based on experience or simulation analysis. The set positive number, for example, can be 0.2. When the number of samples drawn is 1, the decision margin is 0.2; when the number of samples is 5, the decision margin is 0.2 / 5 = 0.04; when the number of samples is 10, the decision margin is 0.2 / 10 = 0.02. Another implementation method is to use piecewise constant mapping, that is, to predefine a decision margin mapping table, which lists different sample size ranges and their corresponding fixed decision margin values. For example, when the number of samples is between 1 and 3, the dynamically adjusted decision margin is 0.15; when the number of samples is between 4 and 7, the dynamically adjusted decision margin is 0.10; when the number of samples is between 8 and 15, the dynamically adjusted decision margin is 0.05; when the number of samples is greater than 15, the dynamically adjusted decision margin is 0.03. By querying the range to which the current number of samples drawn belongs, the corresponding dynamically adjusted decision margin value can be obtained. This mechanism ensures that the decision criteria can adaptively become stricter as the test evidence accumulates.

[0047] The absolute difference is compared with the dynamically adjusted decision margin, and the final decision is made based on the relationship between the updated overall risk estimate and the preset risk threshold. In the first case, if the updated overall risk estimate is less than the preset risk threshold, and the calculated absolute difference is greater than or equal to the dynamically adjusted decision margin, the batch is determined to be accepted. This condition means that the risk of the batch currently assessed is not only lower than the acceptable upper limit, but also significantly lower than the confidence level required for the current sample size, which is sufficient to make the decision to accept. At this time, sampling can be stopped and the final instruction to accept the batch of products can be output.

[0048] In the second scenario, if the updated overall risk estimate is greater than the preset risk threshold, and the calculated absolute difference is greater than or equal to the dynamically adjusted decision margin, the batch is determined to be rejected. This condition means that the currently assessed batch risk is not only higher than the acceptable upper limit, but also significantly higher than the required confidence level for the current sample size, which is sufficient to make a rejection decision. At this point, sampling can be stopped, and the final instruction to reject the batch of the product can be output.

[0049] In the third scenario, if neither of the above two conditions is met—that is, when the updated overall risk estimate is less than the preset risk threshold but the absolute difference is less than the dynamically adjusted decision margin; or when the updated overall risk estimate is greater than the preset risk threshold but the absolute difference is less than the dynamically adjusted decision margin; or when the updated overall risk estimate equals the preset risk threshold, resulting in an absolute difference of 0—then sampling is uniformly determined to continue. This condition implies that although the current risk assessment result may deviate from the risk threshold, the degree of deviation has not yet reached the significance level required for the current testing phase, or the assessment result happens to fall on the risk threshold, and the strength of the evidence is insufficient to support a [missing information]. A clear final decision requires drawing another sample to obtain more information and reduce the uncertainty of the decision. Once it is determined to continue sampling, the entire process returns to the starting point of sequential sampling, waiting to draw and measure a new sample, and then re-executes the updated overall risk estimate and subsequent decision steps. This cycle continues until the acceptance or rejection criteria are met. This comparison method based on dynamic decision margin cleverly combines the absolute comparison of risk levels with the relative assessment of evidence strength, so that the decision strictly follows the risk control standards and fully considers the gradual accumulation of evidence during the testing process, thus achieving a more refined balance between risk control and testing economy.

[0050] Example 2: Figure 2 A schematic diagram of the product quality inspection risk assessment system based on measurement uncertainty of the present invention is provided. The product quality inspection risk assessment system based on measurement uncertainty includes: The data acquisition module is used to acquire the measurement value of each sample drawn in the sequential sampling inspection process of the product batch to be evaluated, as well as the current measurement uncertainty assessment value corresponding to each measurement value. The sequential boundary module is used to determine whether the current test status meets the preset sequential decision boundary conditions based on the measurement values ​​of all extracted samples. The control state identification module is used to perform pattern recognition analysis on the current measurement uncertainty assessment value sequence based on the statistical process control chart if the test state does not meet the sequential decision boundary conditions, in order to determine whether the measurement process is in a statistical control state. The membership calculation module is used to calculate the membership value of the measured value to the preset qualified interval for each sample, based on its measured value and the corresponding current measurement uncertainty assessment value, if the measurement process is under statistical control. The risk update module is used to update the overall risk estimate of the product batch quality level based on the membership values ​​of all extracted samples. The risk assessment module is used to compare the updated overall risk estimate with the preset risk threshold, and make a decision on whether to continue sampling, accept the batch, or reject the batch based on the comparison result.

[0051] All calculations involved in the embodiments are dimensionless numerical calculations, and the preset parameters and thresholds in the calculations are set by those skilled in the art according to the actual situation.

[0052] It should be noted that this invention can be deployed on the device itself to realize embedded applications, or it can run on a PC or other terminal with a user interface, thereby meeting various hardware environments and usage requirements.

[0053] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions according to the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wireless or wired transmission; wired transmission methods include optical fiber, twisted pair, coaxial cable, etc.; wireless transmission includes infrared, microwave, etc. Computer-readable storage media can be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more sets of available media. Available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. Semiconductor media can be solid-state drives.

[0054] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0055] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0056] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0057] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0058] If a function is implemented as a software module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0059] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0060] In conclusion, the above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A product quality inspection risk assessment method based on measurement uncertainty, characterized in that, include: S1: Obtain the measurement value of each sample drawn in the sequential sampling inspection process of the product batch to be evaluated, and the current measurement uncertainty assessment value corresponding to each measurement value; S2: Based on the measurements of all extracted samples, determine whether the current test status meets the preset sequential decision boundary conditions; S3: If the test state does not meet the sequential decision boundary conditions, then perform pattern recognition analysis on the current measurement uncertainty assessment value sequence based on the statistical process control chart to determine whether the measurement process is under statistical control. S4: If the measurement process is under statistical control, then for each sample, calculate the membership value of the measured value to the preset qualified interval based on its measured value and the corresponding current measurement uncertainty assessment value. S5: Update the overall risk estimate of the product batch quality level based on the membership values ​​of all extracted samples; S6: Compare the updated overall risk estimate with the preset risk threshold, and make a decision on whether to continue sampling, accept the batch, or reject the batch based on the comparison result.

2. The product quality inspection risk assessment method based on measurement uncertainty according to claim 1, characterized in that, S1 includes: Receive the execution measurement command for the latest sample, and then control the measuring equipment to detect the latest sample according to the execution measurement command to obtain the raw detection data; The raw detection data is processed to obtain the measured values; The current measurement uncertainty assessment value corresponding to the measured value is obtained simultaneously based on the measurement condition parameters during the measurement and the current calibration status information of the measuring equipment.

3. The product quality inspection risk assessment method based on measurement uncertainty according to claim 1, characterized in that, S2 include: Calculate the current sequential probability ratio statistic based on the measurements of all extracted samples; The sequential probability ratio statistic is compared with the preset acceptance and rejection thresholds; If the sequential probability ratio statistic is less than or equal to the acceptance threshold, it is determined that the acceptance decision boundary condition is met. If the sequential probability ratio statistic is greater than or equal to the rejection threshold, it is determined that the rejection decision boundary condition is met; otherwise, it is determined that the test state does not meet the sequential decision boundary condition.

4. The product quality inspection risk assessment method based on measurement uncertainty according to claim 1, characterized in that, S3 include: Construct a sequence of current measurement uncertainty assessment values ​​based on the current measurement uncertainty assessment values; Based on the current measurement uncertainty assessment value sequence, construct single-value control charts and moving range control charts, and apply the preset statistical process control discrepancy rules to analyze the data point distribution patterns in the single-value control charts and moving range control charts; If the data point distribution pattern does not trigger any outlier detection rule, the measurement process is determined to be in statistical control; otherwise, the measurement process is determined not to be in statistical control.

5. The product quality inspection risk assessment method based on measurement uncertainty according to claim 1, characterized in that, S4 includes: Determine the lower and upper limits of the preset qualified range; Determine the width of the transition zone around the lower and upper limits based on the current measurement uncertainty assessment value; Based on the position of the measured value within the segmented interval formed by the lower limit, upper limit, and transition zone width, the membership value of the measured value to the preset qualified interval is calculated.

6. The product quality inspection risk assessment method based on measurement uncertainty according to claim 5, characterized in that, Based on the position of the measured value within the segmented interval formed by the lower limit, upper limit, and transition zone width, the membership value of the measured value with respect to the preset qualified interval is calculated as follows: when the measured value is less than or equal to the lower limit minus the transition zone width, the membership value is 0; when the measured value is greater than the lower limit minus the transition zone width but less than the lower limit, the membership value increases linearly from 0 to 1; when the measured value is between the lower limit and the upper limit, the membership value is 1; when the measured value is greater than the upper limit but less than the upper limit plus the transition zone width, the membership value decreases linearly from 1 to 0; when the measured value is greater than or equal to the upper limit plus the transition zone width, the membership value is 0.

7. The product quality inspection risk assessment method based on measurement uncertainty according to claim 1, characterized in that, S5 include: Calculate the average membership values ​​of all extracted samples to characterize the central tendency of batch quality; Simultaneously, the standard deviation of the membership values ​​of all extracted samples is calculated to characterize the dispersion of batch quality; Based on the combination relationship between the mean and the standard deviation, the corresponding target risk mapping mode is selected from multiple preset risk mapping modes; By using the target risk mapping model to process the mean and standard deviation, an updated overall risk estimate of the product batch quality level is generated.

8. The product quality inspection risk assessment method based on measurement uncertainty according to claim 7, characterized in that, The process of using the target risk mapping model to process the mean and standard deviation and generate an updated overall risk estimate of the product batch quality level includes: taking the ratio of the mean to the standard deviation as input and obtaining the updated overall risk estimate by querying a pre-established risk mapping table.

9. The product quality inspection risk assessment method based on measurement uncertainty according to claim 1, characterized in that, S6 include: Calculate the absolute difference between the updated overall risk estimate and the preset risk threshold; The absolute difference is compared with a dynamically adjusted decision margin; If the updated overall risk estimate is less than the risk threshold and the absolute difference is greater than or equal to the decision margin, the batch is determined to be accepted. If the updated overall risk estimate is greater than the risk threshold and the absolute difference is greater than or equal to the decision margin, the batch is rejected; otherwise, sampling continues.

10. A product quality inspection risk assessment system based on measurement uncertainty, used to implement the product quality inspection risk assessment method based on measurement uncertainty as described in any one of claims 1-9, characterized in that, include: The data acquisition module is used to acquire the measurement value of each sample drawn in the sequential sampling inspection process of the product batch to be evaluated, as well as the current measurement uncertainty assessment value corresponding to each measurement value. The sequential boundary module is used to determine whether the current test status meets the preset sequential decision boundary conditions based on the measurement values ​​of all extracted samples. The control state identification module is used to perform pattern recognition analysis on the current measurement uncertainty assessment value sequence based on the statistical process control chart if the test state does not meet the sequential decision boundary conditions, in order to determine whether the measurement process is in a statistical control state. The membership calculation module is used to calculate the membership value of the measured value to the preset qualified interval for each sample, based on its measured value and the corresponding current measurement uncertainty assessment value, if the measurement process is under statistical control. The risk update module is used to update the overall risk estimate of the product batch quality level based on the membership values ​​of all extracted samples. The risk assessment module is used to compare the updated overall risk estimate with the preset risk threshold, and make a decision on whether to continue sampling, accept the batch, or reject the batch based on the comparison result.

Images