Automatic verification system health state assessment method based on verification error change

Abstract

An automatic verification system health state assessment method based on verification error change relates to the technical field of power equipment verification, and comprises the steps of determining a health score index, determining an error threshold value, collecting data, constructing a quantitative model, and calculating a health state score. The beneficial effects of the invention are that the method can achieve the more accurate early warning of the health state in advance through a threshold method, and improves the reliability and robustness of the operation of an assembly line. The health state of the equipment can be diagnosed in a short time, and data support can be provided for fault diagnosis and preventive maintenance.

Description

Technical Field

[0001] This invention relates to the field of power equipment calibration technology, and specifically to a method for assessing the health status of an automated calibration system based on changes in calibration error. Background Technology

[0002] In automated verification systems, due to the large scale of the production line and the numerous types of key components involved, traditional health status assessment methods often fail to comprehensively and accurately reflect the overall operating status of the equipment. Especially when using common scoring methods for health assessment, the production line comprises multiple subsystems and components, such as robotic arm joint motors, pneumatic solenoid valves, sensors, and control systems. Each component requires assessment of multiple state variables (such as vibration, temperature rise, and displacement accuracy), causing the number of scoring items to grow exponentially, ultimately reaching O(n log n). m The complexity of this scoring system is significant. This high-dimensional scoring system not only leads to a dramatic increase in the workload of assessment, but also makes it difficult for the system to calculate specific health status scores in real time and efficiently. Summary of the Invention

[0003] To overcome the shortcomings of existing technologies, this invention proposes a health scoring mechanism after comparing potential health indicators such as vibration, temperature, and energy consumption of automated inspection production lines. This mechanism assesses and scores the health status of the production line, enabling it to diagnose the health status of equipment in a short time and provide data support for fault diagnosis and preventive maintenance.

[0004] This invention provides a method for assessing the health status of an automated verification system based on changes in verification error. The method includes determining health scoring indicators, determining error thresholds, data acquisition, quantitative model construction, and health status score calculation.

[0005] The aforementioned determination of health scoring indicators involves screening core indicators that can directly reflect the stability of verification accuracy and have a strong correlation with the system's health status. For example, the energy meter verification line is anchored to the basic error of the energy meter, and the current transformer verification line is anchored to the ratio error.

[0006] The aforementioned method for determining the error threshold in an automated verification system health status assessment based on changes in verification error involves assessing the system's health status based on a threshold for the verification error, with an initial threshold ε. th The calculation is performed using the 3σ principle:

[0007] ε th =μ+3σ

[0008] Where μ and σ are derived from equipment break-in period data (first 200 cycles);

[0009] At the same time, an exponential decay factor 'a' is introduced to achieve adaptive adjustment of the threshold according to the device.

[0010] α(t)=e -0.001t

[0011] Where t represents the equipment running time; the overall threshold is:

[0012]

[0013] The data acquisition mentioned above, and the key points of electricity meter verification, are the basic error and creeping error. The basic error needs to be controlled within ε. th Internally, the key points of transformer calibration are to collect the ratio error and angle difference, both of which must fall within ε. th The data collection frequency is based on a batch size of 15 calibrated devices. The mean and standard deviation of the error are calculated for each batch. The laboratory environment is controlled at a temperature of 20±1℃ and a humidity of 50±5%. This excludes error deviations caused by environmental fluctuations or malfunctions of the standard instruments, ensuring that the data only reflects the health status of the production line itself.

[0014] The data acquisition can be directly connected to the PLC (Programmable Logic Controller) or MES (Manufacturing Execution System) of the production line to collect verification error data in real time. This eliminates the need for a separate data transmission and storage architecture, significantly reducing the integration cost and implementation difficulty of implementing the specifications.

[0015] The quantization model construction is characterized by the fact that this step of quantization model construction needs to be based on ε. th The interval optimization feature calculation and scoring logic needs to select key quantitative indicators that can reflect the characteristics of the system detection error in three aspects: center shift, dispersion change and distribution structure; specifically, these include: error mean shift, error fluctuation coefficient and error information entropy.

[0016] Error mean deviation measures the degree to which the current batch error mean deviates from the ideal target value (center value is 0), reflecting the overall measurement deviation of the system.

[0017] Error fluctuation coefficient is used to measure the ratio of the standard deviation of the current batch error to the absolute value of the mean error, reflecting the degree of consistency or variability within the system.

[0018] Error information entropy is used to measure the frequency distribution structure of errors within a certain preset interval, reflecting the trend of errors changing from concentration to dispersion and from regularity to disorder.

[0019] The formula for calculating the mean offset of the error is:

[0020]

[0021] μ ε U represents the average error of the current batch. tolThe upper limit (absolute value form) in the empirical or standard limits is verified by historical data or standard limits. The upper limit is 0.1 and the lower limit is -0.1, and the error center value is 0. If the mean error is 0.035, the offset = 0.035 / 0.1 × 100% = 35%, which intuitively reflects the distance from the upper limit.

[0022] The formula for calculating the error fluctuation coefficient is as follows:

[0023]

[0024] Where σ ε m is the standard deviation of the current batch error. min A preset minimum threshold (e.g., 0.01 or 0.001) is used to avoid instability caused by the denominator being too close to zero; for example, when the mean error is 0.03 and the standard deviation is 0.01, the fluctuation coefficient...

[0025] =0.01 / 0.03×100%≈33.3%, reflecting the consistency of error within the same batch;

[0026] The formula for calculating error information entropy is:

[0027]

[0028] ε th The system is divided into k preset intervals, where Pi is the relative frequency of error occurrence in the corresponding interval. The lower the entropy value, the better the error concentration and system state. The higher the entropy value, the more the error spreads and the state tends to deteriorate.

[0029] For example, five intervals: [-U tol ~-α],[-α~-β],[-β~β],[β~α],[α~U tol Specifically, α and β can be determined from historical experience, such as 0.06 and 0.2. The five equidistant intervals are: -0.1 to -0.06, -0.06 to -0.02, -0.02 to 0.02, 0.02 to 0.06, and 0.06 to 0.1. Calculate the probability Pi of error occurrence in each interval. The lower the entropy value (e.g., ≤1.0), the more concentrated the error is in the middle interval (-0.02 to 0.02), and the better the health status. The higher the entropy value (e.g., >2.0), the more the error spreads to the two ends of the interval, and the health deteriorates.

[0030] The key quantitative indicators mentioned above—error mean deviation, error fluctuation coefficient, and error information entropy—are each calculated into sub-scores, and five levels are set according to the principle of "step-down + lag tolerance": healthy—good—sub-healthy—deteriorating—failure. The level boundaries of each indicator can be determined based on historical operating data or statistical results of equipment failure cases.

[0031] For example, the score for the mean deviation of the error is as follows: ≤20% gets 100 points; 20%-40% gets 80 points; 40%-60% gets 60 points; 60%-80% gets 30 points; >80% gets 10 points.

[0032] For example, the error fluctuation coefficient scores are as follows: ≤30% gets 100 points; 30%-50% gets 80 points; 50%-80% gets 60 points; 80%-120% gets 30 points; >120% gets 10 points.

[0033] For example, the error information entropy sub-score is as follows: ≤1.0 gets 100 points; 1.0-1.5 gets 80 points; 1.5-1.8 gets 60 points; 1.8-2.0 gets 30 points; >2.0 gets 10 points.

[0034] Information entropy theory provides a quantitative basis for measuring health status by measuring error changes. It quantifies the uncertainty of error data by using information entropy, and transforms health status into a calculable and comparable numerical indicator.

[0035] Information entropy theory describes the uncertainty and disorder of data. When the pipeline is in a healthy state, the fluctuation range of error data is small and the distribution is concentrated, and its information entropy value is low, indicating that the uncertainty of error data is weak and the health state is stable. When the health state of the pipeline deteriorates, the fluctuation range of error data expands and the distribution dispersion increases, and the corresponding information entropy value increases significantly, reflecting the increased uncertainty of error data and the deteriorating trend of the health state.

[0036] The health status score is calculated using a machine learning model (such as random forest or support vector machine) that fits weighted coefficients based on historical data. The weight of each state variable is determined based on its contribution to the error; the mean deviation of the error has a weight of 45%, the volatility coefficient has a weight of 30%, and the information entropy has a weight of 25%. The mean deviation of the error is directly related to the distance to the pass line, the volatility coefficient reflects stability, and the information entropy reflects the distribution trend.

[0037] The overall health status score is calculated as follows: (mean deviation sub-score × 45%) + (fluctuation coefficient sub-score × 30%) + (information entropy sub-score × 25%). The corresponding relationship between score ranges is clearly defined: 90-100 points = healthy, 70-89 points = good, 50-69 points = sub-healthy, 20-49 points = deteriorating, and <20 points = malfunctioning.

[0038] For current transformer calibration lines, the overall health status score can be calculated without using a quantification model. Instead, the calibration error can be directly used as the evaluation index to assess the system's health status. The specific evaluation method is as follows: the error is stable at 0.2ε th Within this range, and with no trend deviation for three consecutive batches, and a standard deviation ≤ 0.005, the verification accuracy is fully controllable, and the status assessment is "healthy"; the error fluctuation range expands to 0.4ε.th The mean did not shift significantly (e.g., it remained consistently within 0.3ε). th Within (within), standard deviation ≤ 0.01, although accuracy has slightly decreased but there is no risk of exceeding tolerance, and is assessed as "good"; the error shows a unidirectional trend shift, that is, the mean error of multiple consecutive batches drifts to one side, or the fluctuation range expands to 0.6ε. th If the standard deviation is ≤0.015, further accuracy degradation should be noted, and the assessment is "sub-healthy"; if the error fluctuation exceeds 0.9ε... th Or a single error approaches ε th If the standard deviation is >0.015, a significant accuracy anomaly has occurred. Without intervention, this could easily lead to a malfunction, and is assessed as "deterioration." Errors exceeding ε... th The range of the machine no longer meets the requirements of the verification procedure and is assessed as a "malfunction," requiring immediate shutdown and troubleshooting.

[0039] The beneficial effects of this invention are as follows: By using the threshold method, more accurate early warning of health status can be achieved, improving the reliability and robustness of production line operation. The health assessment method based on calibration error can not only diagnose the health status of equipment in a short time, but also provide data support for fault diagnosis and preventative maintenance. Combined with big data analysis and deep learning algorithms, this method can also achieve adaptive learning, automatically adjusting the assessment criteria as equipment data accumulates and the system is continuously optimized, further improving the accuracy and flexibility of the assessment.

[0040] Error data can be collected in real time without the need for a large amount of additional hardware equipment, and the data sampling frequency and accuracy can meet the needs of health assessment.

[0041] Verification error directly reflects the core function of the production line: "stable output of qualified products." It is fundamentally related to the line's health status. When the line is in good health, error data fluctuates slightly and randomly around a set threshold, with a stable fluctuation range. When the line deteriorates, such as due to slight component wear or calibration deviations, the error initially shows a "trend change," such as a slow shift in the average dimensional error and an expansion of the positioning error fluctuation range. Before a failure occurs, the error exhibits a "sudden deviation," such as a doubling of the positioning error within a short period. This change directly relates to specific health problems, avoiding the drawbacks of indirect indicators like temperature and vibration, which cannot be directly read and may be caused by other factors. Error changes are significantly better than other indicators in terms of the "timeliness" of identifying health hazards: compared to vibration and temperature indicators, which "only change significantly after a failure occurs," error changes are apparent in the "early stages" of health deterioration.

[0042] The health status assessment method based on calibration error effectively addresses the limitations of traditional scoring methods, such as high dimensionality, coupling errors, and fixed thresholds, providing a new assessment approach for automated calibration systems. This method not only improves the accuracy and real-time performance of health status assessments but also adaptively adjusts assessment criteria during equipment degradation, offering crucial support for intelligent operation and maintenance and fault early warning of industrial equipment. Detailed Implementation

[0043] A method for assessing the health status of an automated verification system based on changes in verification error is characterized by including determining health scoring indicators, determining error thresholds, data acquisition, quantitative model construction, and health status score calculation.

[0044] The aforementioned determination of health scoring indicators involves screening core indicators that can directly reflect the stability of verification accuracy and have a strong correlation with the system's health status. For example, the energy meter verification line is anchored to the basic error of the energy meter, and the current transformer verification line is anchored to the ratio error.

[0045] The aforementioned method for determining the error threshold in an automated verification system health status assessment based on changes in verification error involves assessing the system's health status based on a threshold for the verification error, with an initial threshold ε. th The calculation is performed using the 3σ principle:

[0046] ε th =μ+3σ

[0047] Where μ and σ are derived from equipment break-in period data (first 200 cycles).

[0048] At the same time, an exponential decay factor 'a' is introduced to achieve adaptive adjustment of the threshold according to the device.

[0049] α(t)=e -0.001t

[0050] Where t represents the equipment running time; the overall threshold is:

[0051]

[0052] The data acquisition mentioned above, and the key points of electricity meter verification, are the basic error and creeping error. The basic error needs to be controlled within ε. th Internally, the key points of transformer calibration are to collect the ratio error and angle difference, both of which must fall within ε. th The data collection frequency is based on a batch size of 15 calibrated devices. The mean and standard deviation of the error are calculated for each batch. The laboratory environment is controlled at a temperature of 20±1℃ and a humidity of 50±5%. This excludes error deviations caused by environmental fluctuations or malfunctions of the standard instruments, ensuring that the data only reflects the health status of the production line itself.

[0053] The data acquisition can be directly connected to the PLC (Programmable Logic Controller) or MES (Manufacturing Execution System) of the production line to collect verification error data in real time. This eliminates the need for a separate data transmission and storage architecture, significantly reducing the integration cost and implementation difficulty of implementing the specifications.

[0054] The aforementioned quantization model construction step requires ε as a basis. th The interval optimization feature calculation and scoring logic needs to select key quantitative indicators that can reflect the characteristics of the system detection error in three aspects: center shift, dispersion change and distribution structure; specifically, these include: error mean shift, error fluctuation coefficient and error information entropy.

[0055] Error mean deviation measures the degree to which the current batch error mean deviates from the ideal target value (center value is 0), reflecting the overall measurement deviation of the system.

[0056] Error fluctuation coefficient is used to measure the ratio of the standard deviation of the current batch error to the absolute value of the mean error, reflecting the degree of consistency or variability within the system.

[0057] Error information entropy is used to measure the frequency distribution structure of errors within a certain preset interval, reflecting the trend of errors changing from concentration to dispersion and from regularity to disorder.

[0058] The formula for calculating the mean offset of the error is:

[0059]

[0060] μ ε U represents the average error of the current batch. tol The upper limit (absolute value form) in the empirical or standard limits is verified by historical data or standard limits. The upper limit is 0.1 and the lower limit is -0.1, and the error center value is 0. If the mean error is 0.035, the offset = 0.035 / 0.1 × 100% = 35%, which intuitively reflects the distance from the upper limit.

[0061] The formula for calculating the error fluctuation coefficient is as follows:

[0062]

[0063] Where σ ε m is the standard deviation of the current batch error. min A preset minimum threshold (e.g., 0.01 or 0.001) is used to avoid instability caused by the denominator being too close to zero; for example, when the mean error is 0.03 and the standard deviation is 0.01, the fluctuation coefficient...

[0064] =0.01 / 0.03×100%≈33.3%, reflecting the consistency of error within the same batch;

[0065] The formula for calculating error information entropy is:

[0066]

[0067] ε th The system is divided into k preset intervals, where Pi represents the relative frequency of error occurrence within the corresponding interval. A lower entropy value indicates concentrated error and a better system state; a higher entropy value indicates error propagation and a tendency for the system state to deteriorate. For example, five intervals: [-U tol ~-α],[-α~-β],[-β~β],[β~α],[α~U tol Specifically, α and β can be determined from historical experience, such as 0.06 and 0.2. The five equidistant intervals are: -0.1 to -0.06, -0.06 to -0.02, -0.02 to 0.02, 0.02 to 0.06, and 0.06 to 0.1. Calculate the probability Pi of error occurrence in each interval. The lower the entropy value (e.g., ≤1.0), the more concentrated the error is in the middle interval (-0.02 to 0.02), and the better the health status. The higher the entropy value (e.g., >2.0), the more the error spreads to the two ends of the interval, and the health deteriorates.

[0068] The key quantitative indicators mentioned above—error mean deviation, error fluctuation coefficient, and error information entropy—are each calculated into sub-scores, and five levels are set according to the principle of "step-down + lag tolerance": healthy—good—sub-healthy—deteriorating—failure. The level boundaries of each indicator can be determined based on historical operating data or statistical results of equipment failure cases.

[0069] For example, the score for the mean deviation of the error is as follows: ≤20% gets 100 points; 20%-40% gets 80 points; 40%-60% gets 60 points; 60%-80% gets 30 points; >80% gets 10 points.

[0070] For example, the error fluctuation coefficient scores are as follows: ≤30% gets 100 points; 30%-50% gets 80 points; 50%-80% gets 60 points; 80%-120% gets 30 points; >120% gets 10 points.

[0071] For example, the error information entropy sub-score is as follows: ≤1.0 gets 100 points; 1.0-1.5 gets 80 points; 1.5-1.8 gets 60 points; 1.8-2.0 gets 30 points; >2.0 gets 10 points.

[0072] Information entropy theory provides a quantitative basis for measuring health status by measuring error changes. It quantifies the uncertainty of error data by using information entropy, and transforms health status into a calculable and comparable numerical indicator.

[0073] Information entropy theory describes the uncertainty and disorder of data. When the pipeline is in a healthy state, the fluctuation range of error data is small and the distribution is concentrated, and its information entropy value is low, indicating that the uncertainty of error data is weak and the health state is stable. When the health state of the pipeline deteriorates, the fluctuation range of error data expands and the distribution dispersion increases, and the corresponding information entropy value increases significantly, reflecting the increased uncertainty of error data and the deteriorating trend of the health state.

[0074] The health status score is calculated using machine learning models, such as random forests and support vector machines, by fitting weighted coefficients based on historical data. The weight of each state variable is determined based on its contribution to the error; the mean deviation of the error has a weight of 45%, the volatility coefficient has a weight of 30%, and the information entropy has a weight of 25%. The mean deviation of the error is directly related to the distance to the pass line, the volatility coefficient reflects stability, and the information entropy reflects the distribution trend.

[0075] The overall health status score is calculated as follows: (mean deviation sub-score × 45%) + (fluctuation coefficient sub-score × 30%) + (information entropy sub-score × 25%). The corresponding relationship between score ranges is clearly defined: 90-100 points = healthy, 70-89 points = good, 50-69 points = sub-healthy, 20-49 points = deteriorating, and <20 points = malfunctioning.

[0076] For current transformer calibration lines, the overall health status score can be calculated without using a quantification model. Instead, the calibration error can be directly used as the evaluation index to assess the system's health status. The specific evaluation method is as follows: the error is stable at 0.2ε th Within this range, and with no trend deviation for three consecutive batches, and a standard deviation ≤ 0.005, the verification accuracy is fully controllable, and the status assessment is "healthy"; the error fluctuation range expands to 0.4ε. th The mean did not shift significantly (e.g., it remained consistently within 0.3ε). th Within (within), standard deviation ≤ 0.01, although accuracy has slightly decreased but there is no risk of exceeding tolerance, and is assessed as "good"; the error shows a unidirectional trend shift, that is, the mean error of multiple consecutive batches drifts to one side, or the fluctuation range expands to 0.6ε. th If the standard deviation is ≤0.015, further accuracy degradation should be noted, and the assessment is "sub-healthy"; if the error fluctuation exceeds 0.9ε... th Or a single error approaches ε th If the standard deviation is >0.015, a significant accuracy anomaly has occurred. Without intervention, this could easily lead to a malfunction, and is assessed as "deterioration." Errors exceeding ε... th The range of the machine no longer meets the requirements of the verification procedure and is assessed as a "malfunction," requiring immediate shutdown and troubleshooting.

[0077] Taking the current transformer calibration line as an example, its rated range for ratio error is -0.1 to 0.1. Data from a certain period shows that the mean ratio error of two consecutive batches (15 units per batch) is 0.03 and 0.038 respectively (offset = 0.038 / 0.1 × 100% = 38%), the standard deviation is 0.009 (fluctuation coefficient = 0.009 / 0.038 × 100% ≈ 23.7%), and the information entropy is 1.3. According to the model calculation: mean offset sub-score = 80 points (20%-40% range), fluctuation coefficient sub-score = 90 points (<30%), information entropy sub-score = 80 points (1.0-1.5 range), total health score = 80 × 45% + 90 × 30% + 80 × 25% = 36 + 27 + 20 = 83 points, corresponding to the "good" stage. In the subsequent batch, the average value rose to 0.045 (45% deviation), and the total health score dropped to 68 points, entering the "sub-healthy" stage. The maintenance personnel promptly checked the wiring terminals of the calibration module and found that a slight looseness caused the signal transmission deviation. After tightening, the average error dropped back to 0.025, and the score rose back to 88 points, effectively preventing further deterioration of accuracy.

[0078] Ultimately, the standards for the current transformer calibration line were established as follows:

[0079] In the threshold-based health status assessment method based on verification error, the initial threshold εth is calculated using the 3σ principle:

[0080] ε th =μ+3σ

[0081] Where μ and σ are derived from equipment break-in period data (first 200 cycles).

[0082] At the same time, an exponential decay factor 'a' is introduced to achieve adaptive adjustment of the threshold according to the device.

[0083] α(t)=e -0.001t

[0084] Where t represents the equipment running time; the overall threshold is:

[0085]

[0086] The final evaluation indicators are shown in Table 1:

[0087] Table 1 Health Status Assessment Table

[0088]

[0089] Where M is the detection error of the prediction algorithm after 96 time steps.

Claims

1. A method for assessing the health of an automated assay system based on changes in assay error, comprising: The method comprises determining a health score index, determining an error threshold, data collection, quantitative model construction, and health state score calculation.

2. The method of claim 1, wherein, The health score index is determined by screening core indexes that can directly reflect the stability of the accuracy of the test and have a strong correlation with the system health state, anchoring the basic error of the electric energy meter test line, and anchoring the ratio error of the mutual inductor test line.

3. The method of claim 2, wherein the method further comprises: The error threshold is determined based on the evaluation method of the health state of the automated testing system based on the change of the testing error, and the initial threshold value ε of the system health state evaluation based on the threshold value of the testing error is th The 3σ principle is used for calculation: e th = μ + 3σ Wherein μ, σ come from the data of the equipment running-in period; An exponential decay factor a is introduced to realize the adaptive adjustment of the threshold value with the equipment; a(t) = e -0.001t Wherein t represents the running time of the equipment; and the overall threshold value is:

4. The method of claim 3, wherein the method further comprises: The data collection focuses on collecting the basic error and the potential error of the electric energy meter test, and collecting the ratio error and the angle error of the mutual inductor test; the collection frequency is 1 batch for every 15 equipment tests, the error mean and the standard deviation are calculated for each batch, the temperature of the laboratory environment is controlled at 20±1℃, and the humidity is controlled at 50±5%.

5. The method of claim 4, wherein the method further comprises: The data collection can directly access the PLC or MES of the flow line to collect the test error data in real time.

6. The method of claim 5, wherein the method further comprises: The quantification model is constructed based on ε th Interval optimization feature calculation and scoring logic, key quantitative indexes that can reflect the characteristics of the system detection error in the center deviation, dispersion change and distribution structure are selected; specifically including: error mean deviation degree, error fluctuation coefficient, error information entropy.

7. The method of claim 6, wherein the method further comprises: The error mean deviation degree calculation formula is: μ ε is the mean of the current batch error, U tol is the upper limit in the empirical or normative limits; The error fluctuation coefficient calculation formula is: where σ ε is the current batch error standard deviation, m min is a preset minimum threshold value; The error information entropy calculation formula is: The ε th is divided into k preset intervals, and Pi is the relative frequency of error occurrence in the corresponding interval. The lower the entropy value, the better the system state. The higher the entropy value, the worse the system state.

8. The method of claim 7, wherein the method further comprises: The key quantitative indexes, the error mean deviation degree, the error fluctuation coefficient, and the error information entropy, are used to calculate sub-scores, and five levels are set according to the "ladder decrease + lag tolerance" principle: health, good, sub-health, deterioration, and failure; the level boundaries of each index can be determined based on historical operation data or equipment failure case statistics.

9. The method of claim 8, wherein the method further comprises: The health state score calculation is performed by a machine learning model according to historical data fitting weighting coefficients; the influence weight of each state variable is determined according to its contribution to the error; the error mean deviation degree weight is 45%, the fluctuation coefficient weight is 30%, and the information entropy weight is 25%. The total health state score = (mean deviation sub-score × 45%) + (fluctuation coefficient sub-score × 30%) + (information entropy sub-score × 25%), and the score interval corresponding relationship is clear: 90-100 points = health, 70-89 points = good, 50-69 points = sub-health, 20-49 points = deterioration, and <20 points = failure.

10. The method of claim 9, wherein, For the transformer calibration line, the calibration error can also be directly used as an evaluation index to evaluate the health status of the system. The specific evaluation method is: the error is stable at 0.2ε th , and there is no trend deviation in the last three batches, and the standard deviation is less than or equal to 0.

005. At this time, the calibration accuracy is completely controllable, and the state evaluation is "healthy"; the error fluctuation range is expanded to 0.4ε th , and the average has no obvious deviation (such as always within 0.3εth), and the standard deviation is less than or equal to 0.

01. Although the accuracy decreases slightly, there is no risk of exceeding the error, and the evaluation is "good"; The error appears one-way trend deviation, that is, the mean of the error of multiple batches in succession drifts to one side, or the fluctuation range expands to 0.6ε th , the standard deviation is less than or equal to 0.015, and the precision is further attenuated, which needs to be vigilant, and is evaluated as "sub-health"; Error fluctuation exceeds 0.9ε th , or single error reaches close to ε th , standard deviation > 0.015, at which point obvious accuracy abnormality has occurred, and if not intervened, it is easy to enter failure, and is evaluated as "deterioration"; error exceeds ε th range, which has failed to meet the requirements of the verification procedure, and is evaluated as "failure", and needs to be immediately stopped for troubleshooting.

Images