Abnormal result tracing method and system for steel material inspection and detection process

Abstract

The application discloses a steel material inspection and detection process-oriented abnormal result tracing method and system, and relates to the technical field of industrial data processing quality tracing.The steel material inspection and detection process-oriented abnormal result tracing method and system comprises the following steps: S1, preprocessing inspection event result reproduction data and link process evidence data; S2, performing link confirmation and link isolation review backfilling according to link deviation discriminant analysis results; S3, performing reproduction evidence solidification and evidence reset backfilling according to coupling inhibition reproduction discriminant results; and S4, outputting abnormal accountability discriminant conclusions and completing tracing version freezing archiving and disposal shunting according to accountability convergence discriminant analysis results.The method solves the problem that the existing steel material inspection and detection abnormal tracing method is difficult to distinguish between real material abnormalities caused by performance fluctuations of the steel material and false abnormalities introduced by detection link deviation, and thus the abnormal accountability cannot be determined clearly, and quality misjudgment or risk omission is caused.

Description

Technical Field

[0001] This invention relates to the field of industrial data processing technology, specifically to a method and system for tracing the source of abnormal results in the steel inspection and testing process. Background Technology

[0002] As a crucial basic industrial material, the quality stability of steel directly impacts the safety and reliability of engineering structures and equipment. During the production, distribution, and use of steel, various testing and inspection methods are typically employed to test its mechanical properties, chemical composition, and related indicators, and the results are interpreted according to relevant standards and specifications. Currently, steel testing and inspection are generally conducted using laboratory testing equipment. The testing process includes sample preparation, test execution, result recording, and necessary retesting and calibration operations. Test results are widely used in quality assessment, delivery acceptance, and quality management. In practical engineering applications, steel test results may be affected by various factors, such as fluctuations in the steel's own properties, the condition of the testing equipment, and environmental conditions. With the continuous increase in the scale and scope of steel testing, steel testing and inspection are gradually developing towards informatization and digitalization. Test results and related data are used for industrial data acquisition, processing, and management, placing higher demands on the standardization and consistency of the testing process and result management.

[0003] For example, invention patent CN118520994B discloses an intelligent defect detection method and device for cold-rolled steel sheets. This method statistically analyzes the production processes of cold-rolled steel sheets to obtain defect threshold data for the first rolling stage, the second annealing stage, and the third pickling stage. The method then performs detection on each of these production stages to obtain corresponding defect detection results. These results include defect characterization values ​​for the first rolling stage, the second annealing stage, and the third pickling stage. The first rolling stage obtains defect characterization results through image comparison and thickness deviation statistics; the second annealing stage obtains results through annealing temperature monitoring and cooling rate analysis; and the third pickling stage obtains results by monitoring the surface roughness and defect density of the steel sheet. By comparing the defect characterization values ​​of each production stage with the corresponding defect thresholds, the method achieves screening and alerting for defects in the cold-rolled steel sheet production process, thereby enabling the identification and detection of defective stages during cold-rolled steel sheet production.

[0004] For example, invention patent CN118071193A discloses a method for delivering electronic certificates of conformity for steel structure components. This method includes steps such as digital design and manufacturing, production process monitoring, electronic certificate generation, blockchain technology application, data sharing and transmission, and on-site identification and scanning. During production, the appearance, dimensions, and processing status of the steel structure components are monitored using CCD vision inspection devices and sensors. The monitoring data is compared with design specifications. If the component dimensions meet the requirements, an electronic certificate of conformity is automatically generated. Blockchain technology is used to store and encrypt the electronic certificate, ensuring its security and traceability. The generated electronic certificate and related component information can be uploaded to a cloud platform and identified on-site via QR code or RFID technology. Simultaneously, abnormal states occurring during production and delivery are recorded and managed, achieving digital delivery and management of steel structure component quality information.

[0005] In existing steel and steel structure inspection and quality management technologies, the methods mainly focus on identifying defects in the production process, determining the conformity of test results, and digitally recording and delivering quality information. The analysis of abnormal results is usually based on a single test result or a single step, focusing more on whether the abnormality exists and where it occurs. However, in actual engineering scenarios, abnormal test results are often affected by multiple factors, such as fluctuations in the steel's own performance, differences in the sample preparation process, changes in the status of testing equipment, and disturbances in environmental conditions. Existing technologies generally lack a systematic source analysis mechanism for abnormal events as a whole, making it difficult to effectively distinguish and quantify material-side factors and testing link-side factors within a unified framework. This leads to the judgment of the cause of abnormalities relying on human experience, resulting in insufficient consistency and auditability of the attribution conclusions, which in turn easily leads to quality misjudgments or the omission of potential risks.

[0006] Therefore, in response to the above problems, there is an urgent need for methods and systems for tracing the source of abnormal results in the steel inspection and testing process. Summary of the Invention

[0007] Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a method and system for tracing the source of abnormal results in the steel inspection and testing process. It solves the problem that existing methods for tracing the source of abnormalities in steel inspection and testing are unable to distinguish between genuine material abnormalities caused by fluctuations in the steel's own performance and false abnormalities introduced by deviations in the testing process, which leads to unclear determination of the attribution of abnormalities and is prone to quality misjudgment or risk omission.

[0008] Technical solution To achieve the above objectives, the present invention provides the following technical solution: a method for tracing the source of abnormal results in the steel inspection and testing process, comprising: S1, collecting data on the reproduction of inspection event results and evidence data of the link process, and preprocessing the data on the reproduction of inspection event results and evidence data of the link process; S2, performing link deviation discrimination analysis on the data on the reproduction of inspection event results and evidence data of the link process, and performing link confirmation and link isolation verification and backfilling based on the results of the link deviation discrimination analysis; S3, performing coupling suppression reproduction discrimination on the data on the reproduction of inspection event results and evidence data of the link process, and performing reproduction evidence solidification and evidence reset backfilling based on the results of the coupling suppression reproduction discrimination; S4, performing attribution convergence discrimination analysis based on the results of the link deviation discrimination analysis and the coupling suppression reproduction discrimination, outputting the attributability of abnormality discrimination conclusion based on the results of the attribution convergence discrimination analysis, and completing the freezing, archiving, and disposal of the traceability version.

[0009] Furthermore, the specific process for collecting test event result reproduction data and link process evidence data is as follows: Collect test event result reproduction data, which includes: abnormal event identification data, abnormal result value data, retest result value data of the same sample and the same equipment, and retest result value data of the same sample and cross-equipment; Collect link process evidence data, which includes: material control measurement result data, material certificate value data, equipment self-inspection deviation data, detection area temperature sequence data, detection area vibration acceleration sequence data, sample preparation equipment current sequence data, and sample preparation start and end timestamp data.

[0010] Furthermore, the specific preprocessing process for the test event result reproduction data and link process evidence data is as follows: Using a unified event window truncation and timestamp alignment algorithm, abnormal result value data, same sample same equipment retest result value data, same sample cross-equipment retest result value data, material control measurement result data, equipment self-inspection deviation data, detection area temperature sequence data, detection area vibration acceleration sequence data, and sample preparation equipment current sequence data are aligned to the same abnormal event window; using the Hampel anomaly identification algorithm, abnormal result value data, same sample same equipment retest result value data, same sample cross-equipment retest result value data, and... Outlier removal was performed on the material control measurement results data; noise smoothing was applied to the temperature sequence data, vibration acceleration sequence data, and sample preparation equipment current sequence data of the detection area using the Savitzky-Golay algorithm; short-term missing segments of the temperature sequence data, vibration acceleration sequence data, and sample preparation equipment current sequence data of the detection area were completed using the monotonic cubic interpolation completion algorithm; and dimensionless processing was performed on the equipment self-test deviation data, temperature sequence data, vibration acceleration sequence data, and sample preparation equipment current sequence data using distribution standardization and linear normalization algorithms.

[0011] Furthermore, the specific process of performing link deviation discriminant analysis on the test event result reproduction data and link process evidence data is as follows: The material control measurement result data and material certificate value data are compared using a relative deviation calculation algorithm to obtain the material control deviation amount; equipment self-test deviation data is obtained, and the zero-point verification deviation result and range verification deviation result are compared using a maximum value mapping algorithm and normalized to obtain the equipment self-test deviation amount; based on the temperature sequence data and vibration acceleration sequence data of the detection area, a sliding window peak-valley amplitude extraction is performed on the abnormal event window sequence. The algorithm is normalized to obtain the environmental disturbance quantity; the ratio of the device self-test deviation quantity to the value after subtracting the device self-test deviation quantity and adding the zero-division protection term is calculated and added to the constant to obtain the device self-test amplification modulation term; the ratio of the environmental disturbance quantity to the value after subtracting the environmental disturbance quantity and adding the zero-division protection constant is calculated and added to the constant to obtain the environmental disturbance amplification modulation term; the product of the material control deviation quantity, the device self-test amplification modulation term, and the environmental disturbance amplification modulation term is calculated, and the opposite number is taken before exponential operation is performed. Then, the result of the exponential operation is subtracted from the value to obtain the link deviation evidence discrimination value.

[0012] Furthermore, the specific process of performing link confirmation and link isolation verification and backfilling based on the link deviation discrimination analysis results is as follows: Real-time comparison of the link deviation evidence discrimination value and the link deviation evidence discrimination threshold: When the link deviation evidence discrimination value is less than the link deviation evidence discrimination threshold, the link confirmation process is executed: Under the condition of not changing the sample, the equipment self-inspection process is repeated once and the equipment self-inspection deviation data is retained. At the same time, the material control measurement is repeated once and the material control measurement result data is retained. After updating the data, an abnormal event data packet is generated and input into the coupling suppression reproduction discrimination; When the link deviation evidence discrimination value is greater than or equal to the link deviation evidence discrimination threshold, the link isolation, verification and backfilling process is executed: The output of the detection method identification data corresponding to the current equipment identification data is suspended, and the equipment self-inspection process and material control measurement are repeated; The vibration acceleration sequence data of the detection area is used to trigger the station vibration reduction verification, and the temperature sequence data of the detection area is used to trigger the constant temperature control verification; The equipment self-inspection deviation data and material control measurement result data generated by the verification are backfilled into the abnormal event data packet, and the backfilled abnormal event data packet is input into the abnormal attribution discrimination and traceability freeze output process.

[0013] Furthermore, the specific process for coupling suppression and reproducibility discrimination of test event result reproduction data and link process evidence data is as follows: For abnormal result value data and retest result data of the same sample and the same equipment, the difference between the retest results of the same sample and the same equipment is obtained through a relative difference calculation algorithm; for abnormal result value data and retest result data of the same sample across different equipment, the difference between the retest results of the same sample across different equipment is obtained through a relative difference calculation algorithm; for the current sequence data of the sample preparation equipment and the sample preparation start and end timestamp data, the current sequence within the abnormal event window is used to perform current square integral energy calculation and normalization to obtain the sample preparation disturbance; the product of the natural logarithm of the retest difference of the same sample and the same equipment plus one and the natural logarithm of the retest difference of the same sample across different equipment plus one is calculated to obtain the retest difference logarithmic coupling term; the sample preparation disturbance is calculated, added to the divide-by-zero protection constant, and then added to the constant one to obtain the sample preparation disturbance modulation term; the product of the retest difference logarithmic coupling term and the sample preparation disturbance modulation term is calculated, and an exponential operation is performed, then the constant one is added and the reciprocal is taken to obtain the material reproduction consistency support discrimination value.

[0014] Furthermore, the specific process of solidifying and resetting evidence based on the coupling suppression reproduction discrimination result is as follows: Real-time comparison of the reproduction consistency support discrimination value and the reproduction consistency support discrimination threshold: When the reproduction consistency support discrimination value is less than the reproduction consistency support threshold, the sample preparation and retest evidence reset process is executed: The sample preparation equipment is re-performed with no-load self-test and tool status verification, and the sample is re-prepared once at the same sampling location, and the current sequence data of the sample preparation equipment is re-acquired; the same sample and same equipment retest and the same sample and cross-equipment retest are performed n times each on the newly prepared sample; the new round of retest result data and the sample preparation equipment current sequence data are backfilled into the abnormal event data packet and... The process involves inputting anomaly attributability judgment and source tracing freeze output flow. When the reproducibility consistency support judgment value is greater than or equal to the reproducibility consistency support judgment threshold, the material evidence solidification process is executed: the abnormal result value data, the same sample same equipment retest result value data, and the same sample cross equipment retest result value data are solidified into a reproducibility evidence group triplet under the same abnormal event identification data. The reproducibility evidence group triplet includes unified abnormal event identification data, corresponding evidence data payload, and evidence generation timestamp. An immutable data fingerprint is calculated for the triplet to uniquely identify the evidence status. The solidified abnormal event data package is then input into the anomaly attributability judgment and source tracing freeze output flow.

[0015] Furthermore, based on the results of the link deviation discrimination analysis and the coupling suppression reproduction discrimination analysis, the specific process of the attribution convergence discrimination analysis is as follows: For the data of the same sample and same equipment retest results, the data of the same sample and cross-equipment retest results, the data of the material control measurement results, and the material certificate value data corresponding to the material control measurement results, evidence integrity checks are performed respectively. Evidence integrity checks include existence checks, identifier consistency checks, and anomaly event window consistency checks. When any type of data passes the integrity check, the evidence is deemed complete; when at least one type of data fails the integrity check, the evidence is deemed incomplete. The number of passed sub-items is counted to obtain the number of passed evidence items. The ratio of the number of passed evidence items to the total number of verified sub-items is used as the evidence coverage rate. An exponential mapping is performed on the evidence coverage rate to obtain the real-time evidence completeness. The ratio of the material reproduction consistency support discrimination value to the sum of the link deviation evidence discrimination value and the zero-division protection constant is calculated and multiplied by the real-time evidence completeness to obtain the attribution driving quantity. The attribution driving quantity is inversely inverted and then exponentially calculated. Finally, the result of the exponential calculation is subtracted from the constant to obtain the anomaly attributability discrimination value.

[0016] Furthermore, the specific process of outputting the attribution of responsibility judgment conclusion based on the attribution convergence judgment results and completing the freezing, archiving, and disposal of the traceability version is as follows: Real-time comparison of the attribution of responsibility judgment value and the attribution of responsibility judgment threshold. The attribution of responsibility judgment threshold includes a primary attribution judgment threshold and a secondary attribution judgment threshold. When evidence is incomplete, the time window and input evidence set of the current abnormal event data package are frozen. Evidence is supplemented sequentially in the order of material control measurement, cross-equipment retesting of the same sample, and retesting of the same sample and equipment. After supplementation, the attribution of responsibility judgment value is recalculated, and the supplemented evidence is created and written into the abnormal traceability evidence database. When the evidence is complete and the attribution of responsibility judgment value is greater than or equal to the primary attribution judgment threshold, a material-side disposal closed loop is executed: an extended retesting plan is issued for the same batch of samples corresponding to the batch identification data; the batch test results are issued and enter the suspension process and isolation release is executed; the frozen version of the abnormal event data package is pushed to the quality traceability interface of the smelting and rolling processes to trigger process-side traceability; when the evidence is complete... When the evidence is complete and the anomaly attributability threshold is less than the secondary attribution threshold, a closed-loop handling procedure is executed on the link side: the issuance of the detection method identification data corresponding to the equipment identification data is suspended; calibration verification and material comparison are re-executed for the equipment; and backtracking and verification of results already issued within the same time window for the same equipment and detection method are re-issued. The material comparison measurement results data and equipment self-inspection deviation data generated by the correction verification are written into the anomaly tracing evidence database and bound to the equipment identification data. When the evidence is complete and the anomaly attributability threshold is greater than or equal to the secondary attribution threshold and less than the primary attribution threshold, a convergence judgment closed-loop is executed: the batch issuance suspension process remains unchanged; a dominant judgment is performed on the current link deviation evidence judgment value and the material reproduction consistency support judgment value to determine the dominant support direction of the current anomaly in the material-side evidence and link-side evidence; based on the dominant support direction, the anomaly data packet is classified into one of the material-side handling closed loop and the link-side handling closed loop, and the attribution result and corresponding evidence version are written into the anomaly tracing evidence database.

[0017] The second aspect of this invention provides an anomaly traceability system for steel inspection and testing processes, comprising: a data acquisition and preprocessing module for acquiring reproducible data of inspection events and evidence data of the chain process, and preprocessing the reproducible data of inspection events and evidence data of the chain process; a detection chain deviation evidence discrimination module for performing chain deviation discrimination analysis on the reproducible data of inspection events and evidence data of the chain process, and performing chain confirmation and chain isolation verification backfilling based on the chain deviation discrimination analysis results; a material reproduction consistency support discrimination module for performing coupling suppression reproduction discrimination on the reproducible data of inspection events and evidence data of the chain process, and performing reproduction evidence solidification and evidence reset backfilling based on the coupling suppression reproduction discrimination results; and an anomaly attributability discrimination and traceability freeze output module for performing attribution convergence discrimination analysis based on the chain deviation discrimination analysis results and the coupling suppression reproduction discrimination results, outputting anomaly attributability discrimination conclusions based on the attribution convergence discrimination analysis results, and completing traceability version freeze archiving and disposal distribution.

[0018] Beneficial effects The present invention has the following beneficial effects: (1) This invention constructs a detection link deviation evidence discrimination mechanism to express the link-side influences such as control measurement, equipment self-inspection and environmental disturbance in a quantitative and evidentiary way, thereby achieving the effect of continuous identification and interpretable output of unstable factors in the detection link, effectively solving the problem that link disturbances are difficult to effectively characterize and are easily misinterpreted as material anomalies in the prior art.

[0019] (2) This invention constructs a material reproducibility consistency support discrimination mechanism, which combines the consistency of retesting of the same sample, the consistency of retesting of cross-equipment and the stability of the sample preparation process, thereby realizing the continuous judgment effect of the reproducibility support strength of abnormal results, effectively solving the problem of the fragmented use of retest evidence and the difficulty in forming stable material-side support conclusions in the prior art.

[0020] (3) This invention nonlinearly couples material-side support evidence with link-side deviation evidence and introduces a suppression constraint relationship to form an attribution judgment output, thereby achieving a quantitative distinction between real material anomalies and detected link false anomalies, effectively solving the problem of unclear attribution direction caused by the mixture of material factors and link factors in the prior art.

[0021] (4) This invention introduces a source version freezing and traceable archiving mechanism to form a reproducible record chain of evidence collection, judgment process and handling results of abnormal events, thereby achieving the effect of reproducibility, auditability and traceability of abnormal judgment process, effectively solving the problem of lack of closed-loop traceability in abnormal handling and difficulty in supporting responsibility definition and review in the prior art.

[0022] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0023] Figure 1 This is a flowchart of the abnormal result tracing method for the steel inspection and testing process according to the present invention; Figure 2 This is a structural diagram of the abnormal result tracing system for the steel inspection and testing process of the present invention; Figure 3 This is a graph showing the comparison between the link deviation driving quantity and the link deviation evidence discriminant value of the present invention. Figure 4 This is a flowchart of the abnormal attributability judgment and handling decision-making process of the present invention. Detailed Implementation

[0024] 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.

[0025] Please see Figures 1-4 This invention provides a technical solution: a method for tracing the source of abnormal results in the steel inspection and testing process, comprising the following steps: S1, collecting data on the reproduction of inspection event results and evidence data of the link process, and preprocessing the data on the reproduction of inspection event results and evidence data of the link process; S2, performing link deviation discrimination analysis on the data on the reproduction of inspection event results and evidence data of the link process, and performing link confirmation and link isolation verification and backfilling based on the results of the link deviation discrimination analysis; S3, performing coupling suppression reproduction discrimination on the data on the reproduction of inspection event results and evidence data of the link process, and performing reproduction evidence solidification and evidence reset backfilling based on the results of the coupling suppression reproduction discrimination; S4, performing attribution convergence discrimination analysis based on the results of the link deviation discrimination analysis and the coupling suppression reproduction discrimination, outputting the attributability of abnormality discrimination conclusion based on the results of the attribution convergence discrimination analysis, and completing the freezing, archiving, and disposal of the traceability version.

[0026] Specifically, the process of collecting reproducible data of inspection event results and evidence data of the chain process is as follows: Collecting reproducible data of inspection event results includes: abnormal event identification data, which is jointly generated from furnace identification data, batch identification data, sample identification data, test item identification data, test method identification data, equipment identification data, and test timestamp data, used to bind all evidence of the same abnormality as the same event; and abnormal result value data, which refers to the test result value that triggers the abnormal event identification in the initial test result according to the test item judgment rules during the steel inspection and testing process, representing the original abnormal manifestation of the abnormal event. The data is used for subsequent link deviation evidence discrimination and coupling suppression reproduction discrimination; the retest result data of the same sample and the same equipment is used for coupling suppression reproduction discrimination; the retest result data of the same sample and cross-equipment is used for coupling suppression reproduction discrimination. Among them, the retest timestamp data corresponding to the retest result data of the same sample and cross-equipment is constrained by the abnormal event window of the detection timestamp data in the abnormal event identifier data, and its detection item identifier data, detection method identifier data and abnormal result value data are consistent. The consistency check is performed based on the range information and detection resolution corresponding to the equipment identifier data to ensure that the retest result data of cross-equipment is comparable under the same detection condition semantics.

[0027] The data collected includes: material control measurement results and material certificate values, used for identifying evidence of link deviation; equipment self-inspection deviation data, including zero-point verification deviation results and range verification deviation results directly output by the built-in self-inspection process of the testing equipment, used for identifying evidence of link deviation; temperature sequence data and vibration acceleration sequence data of the testing area, used for identifying evidence of link deviation; and current sequence data and start and end timestamp data of the sample preparation equipment, used for identifying coupling suppression recurrence.

[0028] In this implementation plan, by uniformly collecting and aggregating the data on the reproduction of test results and the evidence data of the link process, a complete evidentiary foundation centered on the anomaly event is formed. This enables the anomaly detection results to obtain multi-source information support within the same time scale and the same event range, thereby achieving a holistic characterization and stable reproduction of the anomaly event. This step provides continuous, alignable, and comparable data input conditions for subsequent link deviation evidence discrimination and material reproduction consistency discrimination, avoiding the anomaly analysis from relying solely on a single detection result or isolated retest data. It improves the completeness and consistency of anomaly events in the analysis of the detection link side and the material side, establishing a reliable data foundation and engineering prerequisite for determining the attributability of anomalies.

[0029] Specifically, the preprocessing process for the test event result reproduction data and the link process evidence data is as follows: Using a unified event window truncation and timestamp alignment algorithm, abnormal result value data, same sample same equipment retest result value data, same sample cross-equipment retest result value data, material control measurement result data, equipment self-inspection deviation data, detection area temperature sequence data, detection area vibration acceleration sequence data, and sample preparation equipment current sequence data are aligned to the same abnormal event window. The detection timestamp data corresponding to the abnormal result value data is used as the center time, and the abnormal event is determined based on the duration of the detection process and the environmental disturbance response delay. The alignment window is defined as follows: the pre-event duration is the larger of the single detection duration corresponding to the current detection item and the minimum response delay of the environmental disturbance; the post-event duration is the larger of the single detection duration and the maximum response delay of the environmental disturbance. This ensures that all evidence points to the same time period and satisfies comparability. The allowable time offset during alignment is jointly limited by the pre-event and post-event durations of the anomaly event window. All data participating in the alignment only participate in subsequent calculations within the coverage area of ​​this event window to avoid cross-event temporal aliasing affecting the judgment results. The Hampel anomaly identification algorithm is used to identify anomaly result values ​​and related data. Outlier removal is performed on the test results data from the same equipment, the test results data from cross-equipment testing of the same sample, and the measurement results data from the material control to avoid single-point mutations that could distort subsequent evidence. The outlier identification process is adaptively completed based on the local statistical characteristics of the data within the outlier event window, acting only on a subset of data within the current outlier event range without affecting the independence between outliers. Noise smoothing is performed on the temperature sequence data, vibration acceleration sequence data, and current sequence data of the sample preparation equipment in the detection area using the Savitzky-Golay algorithm. – The Golay smoothing filter algorithm improves the stability of perturbation evidence calculation. The smoothing process is only used to weaken the impact of high-frequency sampling jitter on the perturbation trend expression. Its output still maintains the overall change pattern of the original sequence to ensure the sensitivity of subsequent perturbation calculation to abnormal changes. The monotonic cubic interpolation completion algorithm completes the short-term missing segments of the temperature sequence data, vibration acceleration sequence data, and current sequence data of the sample preparation equipment in the detection area, ensuring that the subsequent discrimination values ​​can be continuously calculated and visualized. The interpolation process is limited to the abnormal event window to maintain the continuity of the time series and does not introduce information outside the abnormal event window.By employing distribution standardization and linear normalization algorithms, the self-inspection deviation data, temperature sequence data, vibration acceleration sequence data, and current sequence data of the sample preparation equipment are processed to unify their dimensions. This maps continuous numerical data collected under different physical dimensions to a consistent numerical range, providing a comparable data foundation for subsequent discrimination processes based on nonlinear calculations. Dimensional unification is performed independently at the anomaly event granularity level, ensuring data processing is isolated between different anomalies and preventing cross-event statistical characteristics from interfering with the discrimination results.

[0030] In this implementation plan, by applying unified event window constraints and time benchmark alignment to anomaly-related multi-source data, various detection results and process evidence form a consistent temporal sequence within the same anomaly occurrence stage, thereby ensuring the comparability and consistency of data from different sources in anomaly analysis. Simultaneously, by performing outlier suppression, noise smoothing, and continuous completion processing on result-type and process-type data respectively, the interference of transient mutations, sampling jitter, and data gaps on the discrimination results is effectively reduced, making the expression of anomaly features more stable and reliable. Furthermore, by unifying the units of measurement, data with different physical properties are mapped to a consistent numerical space, eliminating the bias effect caused by unitary differences. This provides a continuous, stable, and comparable data foundation for subsequent link deviation evidence discrimination and material reproduction consistency discrimination, improving the robustness and engineering interpretability of the overall anomaly analysis process.

[0031] Specifically, the process of performing link deviation discrimination analysis on the test event result reproduction data and link process evidence data is as follows: The material control measurement result data and material certificate value data are compared using a relative deviation calculation algorithm to obtain the material control deviation amount. A dimensional consistency check is introduced into the relative deviation calculation process using the same test item identification data constraint to avoid misinterpretation of deviations caused by cross-item comparisons. Equipment self-test deviation data is acquired, and the zero-point verification deviation result and range verification deviation result are mapped using a maximum value mapping algorithm and normalized to obtain the equipment self-test deviation amount. The maximum value mapping ensures that the self-test deviation amount maintains monotonic sensitivity to the most unfavorable self-test link and suppresses the mutual cancellation effect between the zero-point verification deviation result and the range verification deviation result. Based on the temperature sequence data and vibration acceleration sequence data of the test area, a sliding window peak-valley amplitude extraction algorithm is performed on the sequence within the abnormal event window and normalized to obtain the environmental disturbance amount. The sliding window peak-valley amplitude extraction algorithm captures the upper bound of the transient disturbance amplitude within the abnormal event window, and boundary smoothing is performed on the overlapping window interval to reduce amplitude jitter caused by sampling phase difference.

[0032] The ratio of the calculated equipment self-test deviation to the value minus the equipment self-test deviation plus a zero-division protection term is added to a constant to obtain the equipment self-test amplification modulation term. A fractional saturation structure is used to generate a rapid enhanced response when the equipment self-test deviation approaches the constant to highlight potential link instability trends. The equipment self-test deviation is normalized to a range of zero to one before entering the amplification modulation calculation. Its upper bound is naturally limited by the maximum observable deviation state of the equipment self-test process within the current abnormal event window, characterizing the relative degree of evolution from stable to unstable equipment self-test status. The ratio of the calculated environmental disturbance to the value minus the environmental disturbance plus a zero-division protection constant is added to a constant to obtain the environmental disturbance amplification modulation term. An isomorphic modulation structure is used to nonlinearly amplify the contribution of the environmental disturbance to the link deviation evidence discriminant value when it approaches the constant, enhancing the identification capability of environmentally sensitive detection stations. The environmental disturbance is normalized by the amplitude of temperature and vibration field disturbances within the abnormal event window. The value range is limited by the actual disturbance level of the workstation corresponding to the current abnormal event, thus ensuring that the environmental disturbance amplification modulation term only amplifies when the abnormal disturbance is significantly enhanced. The product of the material control deviation, the equipment self-test amplification modulation term, and the environmental disturbance amplification modulation term is calculated, and after taking the opposite number, an exponential operation is performed. Then, the result of the exponential operation is subtracted from one to obtain the link deviation evidence discrimination value. The link deviation driving quantity is compressed to the range of zero to one through exponential saturation mapping, and a visually friendly monotonic response curve is obtained to facilitate online discrimination of stability control. The fractional amplification modulation form and the exponential saturation mapping together constitute a monotonic and limited nonlinear response structure, so that the link deviation evidence discrimination value remains flat under low deviation conditions and rapidly amplifies when multiple types of link instability factors are superimposed and approach extreme states. At the same time, it avoids unbounded amplification or numerical oscillation in the output result, thus ensuring that the discrimination result has a consistent explanatory scale among different abnormal events. The specific calculation formula is as follows: ; In the formula, This represents the link deviation evidence discrimination value, which is used to continuously quantify the strength of evidence on the link side and serve as an input constraint for subsequent attribution convergence calculation; It represents the deviation of a substance from its control, characterizes the degree of deviation of the measurement results of a substance control from the substance certificate value, and forms a direct amount of evidence in the control chain; It represents the deviation of the equipment self-test, which is used to characterize the potential deviation risk of the internal self-test status of the equipment on the test output and form the amount of evidence on the equipment side. It represents the amount of environmental disturbance, characterizes the intensity of the influence of temperature and vibration field disturbances within the abnormal event window on the detection stability, and forms the amount of evidence at the work station; This represents the division-to-zero protection constant, which is calculated from the minimum resolvable quantity corresponding to the resolution of the detection equipment, and its value ranges from 0.0005 to 0.001.

[0033] In this embodiment, Table 1 presents example data on the changes in link deviation driving force and link deviation evidence discriminant value. It details the correspondence between the link deviation driving force and the link deviation evidence discriminant value, formed by the combined effects of material control deviation, equipment self-test deviation, and environmental disturbance under different typical detection scenarios. The link deviation driving force represents the initial deviation driving force formed by the superposition of multiple link factors, used to characterize the changing characteristics of the detection link deviation response under multi-source disturbance conditions. In the baseline scenario, the link deviation driving force is 0.0617, and the corresponding link deviation evidence discriminant value is 0.0598, which is at a relatively low level. When only the material control deviation increases, the link deviation driving force increases to 0.1851, and the link deviation evidence discriminant value correspondingly rises to 0.1690. In scenarios where only equipment self-test deviation increases or only environmental disturbance increases, the link deviation driving force is approximately 0.0925, and the corresponding link deviation evidence discriminant value is approximately 0.0884, showing a consistent amplification trend. When all three factors are at a moderately high level, the link deviation driving force further increases to 0.4161, and the link deviation evidence discrimination value rises to 0.3404. When the device self-test deviation or environmental disturbance approaches the boundary conditions, the link deviation driving force reaches 0.5505, and the link deviation evidence discrimination value increases to 0.4234. When the device self-test deviation and environmental disturbance approach the boundary at the same time, the link deviation driving force is significantly amplified to 4.9113, and the link deviation evidence discrimination value approaches saturation, reaching 0.9926.

[0034] Table 1. Example data on changes in link deviation driving forces and link deviation evidence discriminant values.

[0035] like Figure 3 The graph shows the changes in the link deviation driving force and the link deviation evidence discriminant value. Combined with Table 1, it can be seen that there is a clear non-linear mapping relationship between the link deviation driving force and the link deviation evidence discriminant value. When the link deviation driving force increases from 0.0617 to 0.1851, the link deviation evidence discriminant value increases from 0.0598 to 0.1690, with a relatively slow growth trend. When the link deviation driving force further increases to 0.4161 and 0.5505, the link deviation evidence discriminant value accelerates to 0.3404 and 0.4234, respectively. When the link deviation driving force jumps to 4.9113, the link deviation evidence discriminant value rapidly approaches 1, entering the saturation range. Overall, this curve intuitively reflects the transformation of the link deviation evidence discriminant value from slow growth to rapid amplification under the superposition of multi-source disturbances, providing an intuitive and visual basis for link stability assessment.

[0036] In this implementation plan, by uniformly quantifying the effects of material control deviation, equipment self-inspection status, and environmental disturbances, the multi-source unstable factors in the detection link from the control system, the equipment itself, and the workstation environment are transformed into a comparable and accumulative form of evidence, enabling the link-side deviation risk to be characterized in a continuous numerical manner. Utilizing nonlinear amplification and saturation mapping mechanisms, potential unstable trends in the link are effectively highlighted when approaching a critical state, while maintaining a stable response under low-disturbance conditions. This achieves stable quantitative output of link deviation evidence across the entire range, providing a link-side evidentiary basis with engineering interpretability and consistent judgment for subsequent anomaly attribution determination.

[0037] Specifically, the process of performing link confirmation and link isolation verification and backfilling based on the link deviation discrimination analysis results is as follows: Real-time comparison of link deviation evidence discrimination values ​​and link deviation evidence discrimination thresholds: When the link deviation evidence discrimination value is less than the link deviation evidence discrimination threshold, the link confirmation process is executed: without changing the sample, the equipment self-test process is repeated once and the equipment self-test deviation data is retained. At the same time, the material control measurement is repeated once and the material control measurement result data is retained. After updating the data, an abnormal event data package is generated and input into the coupling suppression reproduction discrimination. The abnormal event data package refers to the data set that uses the abnormal event identification data as a unique index, assembles the event result reproduction data and link process evidence data within the same abnormal event window into one, and backfills the latest verification evidence in this process. Specifically, it includes: abnormal event identification data, abnormal result value data, same sample same equipment retest result value data, same sample cross equipment retest result value data, material control measurement result data, material certificate value data, equipment self-test deviation data, detection area temperature sequence data, detection area vibration acceleration sequence data, sample preparation equipment current sequence data, and sample preparation start and end timestamp data, which are used to lock the link baseline before material reproduction consistency discrimination.

[0038] When the link deviation evidence discrimination value is greater than or equal to the link deviation evidence discrimination threshold, the link isolation, verification, and backfilling process is executed: The issuance of results for the detection method identification data corresponding to the current device identification data is suspended. This suspension is achieved by outputting a link isolation control signal to the detection result publishing module. The control signal is generated based on the device identification data and the detection method identification data, and is used to set an isolation status flag for the corresponding detection result instance in the system result publishing queue. This prevents the instance from entering the result issuance and downstream transmission process before the abnormal event data packet completes the attribution judgment and source tracing freeze, while not affecting the normal flow of results corresponding to other device identification data or detection method identification data. The device self-test is then re-executed. The process involves comparing and measuring the process with the materials; triggering vibration reduction verification at the workstation based on the vibration acceleration sequence data of the detection area, and triggering constant temperature control verification based on the temperature sequence data of the detection area; backfilling the equipment self-inspection deviation data and material comparison measurement results generated during the review into the abnormal event data package, and binding the result instances of the original isolated state to the version based on the abnormal event identification data, so that the link evidence generated before and after the review forms a comparable evidence sequence within the same abnormal event data package; inputting the backfilled abnormal event data package into the abnormal attribution judgment and source tracing freeze output process to form verifiable closed-loop evidence of false anomalies in the link, while avoiding misleading judgments on the consistency of material reproduction when the link has not converged.

[0039] In this implementation plan, real-time diversion control is implemented on the results of link deviation evidence discrimination, allowing the detected link to enter differentiated processing paths under different stable states, thereby achieving effective management of link risks in the early stages of anomaly analysis. When the link is in a verifiable state, the link baseline is solidified by reviewing key comparisons and self-inspection information, ensuring that subsequent material reproduction consistency discrimination is based on stable and reliable testing premises. When the link deviation risk reaches the warning level, the output of results is paused, link isolation and environmental verification are performed, and verification evidence is backfilled, prompting the link state to converge first and form verifiable closed-loop evidence. This mechanism ensures that unstable factors on the link side are fully identified and dealt with before attribution discrimination, preventing link disturbances from being mistakenly transmitted to the material side, and improving the reliability of the anomaly judgment process and engineering safety.

[0040] Specifically, the process for coupling suppression and reproducibility discrimination of test event result reproduction data and link process evidence data is as follows: For abnormal result value data and retest result data of the same sample and equipment, a relative difference calculation algorithm is used to obtain the retest difference of the same sample and equipment. The consistency check of the effective bits of the result is introduced into the relative difference calculation process to suppress the artificial increase in difference caused by resolution differences. For abnormal result value data and retest result data of the same sample from multiple equipment, a relative difference calculation algorithm is used to obtain the retest difference of the same sample from multiple equipment. The consistency check of the equipment identification data is performed on the retest difference of the same sample from multiple equipment to ensure that the retest result data of the multiple equipment and the abnormal result value data are under comparable test method identification data conditions. For the current sequence data of the sample preparation equipment and the sample preparation start and end timestamp data, the current sequence within the abnormal event window is calculated using the square integral energy and normalized to obtain the sample preparation disturbance. The square integral energy calculation characterizes the energy occupancy level of the sample preparation load and remains sensitive to short-term peak currents to characterize the potential sample preparation process disturbance intensity.

[0041] The product of the natural logarithm of the difference between the same sample and the same equipment, plus one, and the natural logarithm of the difference between the same sample and the cross-equipment retesting, plus one, is calculated to obtain the retest difference logarithmic coupling term. Logarithmic transformation is used to compress the dynamic range of the difference, and the product coupling structure is used to enhance the joint penalty effect when two types of retest differences coexist. By adding one to the difference before the logarithmic operation, the logarithmic term is continuously zero under zero difference conditions, thus ensuring numerical stability in the discrimination process when the retest results are highly consistent and avoiding logarithmic singularities. The sample preparation disturbance is calculated, added to the division-by-zero protection constant, and then added to the constant one to obtain the sample preparation disturbance modulation term. This term is used to introduce the instability of the sample preparation process into the consistency discrimination in a monotonic manner and to prevent the sample preparation disturbance from being in a state of... The numerical values ​​at the boundary are unstable. The division-to-zero protection constant is consistent with the division-to-zero protection constant used in the link deviation evidence discrimination process to ensure the continuity of the modulation term calculation process when the sample preparation disturbance is close to zero. The product of the logarithmic coupling term of the retest difference and the modulation term of the sample preparation disturbance is calculated and exponentially operated. Then, a constant is added and the reciprocal is taken to obtain the material reproduction consistency support discrimination value. The discrimination value is constrained to the range of zero to one through a logistic compression mapping to obtain a continuous visual curve for easy judgment and stability control. The exponential compression structure ensures that the discrimination value maintains high consistency output under the condition that the retest difference and sample preparation disturbance are both small. It smoothly converges when multiple inconsistency factors are superimposed, avoiding the excessive amplification effect of a single small fluctuation on the discrimination result. The specific calculation formula is as follows: ; In the formula, This represents the material reproducibility consistency support discrimination value, which is used to continuously quantify the support strength of abnormal results under the constraints of retest consistency and sample preparation process disturbance; It indicates the difference in retesting of the same sample and the same equipment, characterizes the degree of deviation of abnormal results in the same detection execution link, and reflects the stability of the output; It indicates the difference in the retest results of the same sample using different equipment, characterizes the degree of deviation of abnormal results under equipment switching conditions, and reflects the influence of equipment correlation. It represents the sample preparation disturbance, characterizes the potential impact of load fluctuations during the sample preparation process on the consistency of the sample state, and forms the sample preparation-side constraint factor; This represents the division-to-zero protection constant, which is calculated from the minimum resolvable quantity corresponding to the resolution of the detection equipment, and its value ranges from 0.0005 to 0.001.

[0042] In this implementation scheme, by jointly modeling the differences in retesting of abnormal results under the same detection execution link and cross-device conditions, and introducing sample preparation process disturbances as a constraint factor, a comprehensive quantitative characterization of the reproducibility stability of abnormal results is achieved. This allows the consistency support of material-side anomalies to be expressed in continuous numerical form. This process compresses the dynamic range of retesting differences while strengthening the penalty effect when multiple types of retesting inconsistencies occur simultaneously, and maintains monotonic sensitivity to sample preparation process instability. This ensures that the material reproducibility consistency support judgment results maintain smooth, stable, and interpretable response characteristics even under complex operating conditions, providing a reliable material-side evidence basis for subsequent anomaly attributability judgment and avoiding judgment bias caused by occasional retesting fluctuations or sample preparation disturbances.

[0043] Specifically, the process of solidifying and resetting evidence based on the coupling suppression reproduction discrimination result is as follows: real-time comparison of the reproduction consistency support discrimination value and the reproduction consistency support discrimination threshold: When the reproducibility consistency support threshold is less than the reproducibility consistency support threshold, the sample preparation retest evidence reset process is executed: First, the sample preparation equipment is re-performed with no-load self-test and tool status verification, and the sample is re-prepared once at the same sampling location, while the current sequence data of the sample preparation equipment is re-acquired; then, the same sample and the same equipment retest and the same sample and cross-equipment retest are performed n times each on the newly prepared sample, where n is jointly determined by the available retest period in the abnormal event window corresponding to the abnormal event identification data and the single test duration corresponding to the test item identification data, to ensure that multiple retests are completed within the same abnormal event window and meet the comparability of results, with a value range of 1 to 5; the new round of retest result data and the sample preparation equipment current sequence data are backfilled into the abnormal event data package and input into the abnormal attributability judgment and traceability freeze output process, to separate the sample preparation-induced abnormality and material non-reproducibility at the evidence level.

[0044] When the reproducibility consistency support threshold is greater than or equal to the reproducibility consistency support threshold, the material evidence solidification process is executed: abnormal result data, retest result data of the same sample and the same equipment, and retest result data of the same sample and cross-equipment are solidified into a reproducibility evidence triplet under the same abnormal event identification data. The reproducibility evidence triplet includes unified abnormal event identification data, corresponding evidence data payload, and evidence generation timestamp. An immutable data fingerprint is calculated for the triplet to uniquely identify the evidence status. The solidified abnormal event data package is input into the abnormal attributability judgment and source tracing freeze output process to provide a reproducible and consistent direct evidence chain during attribution judgment.

[0045] In this implementation plan, by implementing diversion control on the results of material reproducibility consistency support judgment, abnormal results can be promptly distinguished and handled based on whether they have a stable reproducibility basis on the material side. When the reproducibility consistency support is insufficient, the sample preparation and retesting evidence reset process is triggered, prompting the sample preparation process and retesting results to be reacquired under controlled conditions, thus separating the sample preparation inducing factors from the material's own non-reproducible factors at the evidence level. When the reproducibility consistency support meets the requirements, the retesting results are solidified to form a stable and reproducible material-side evidence chain, providing direct support for subsequent liability determination. This mechanism ensures that material-side evidence has sufficient stability and verifiability before entering the liability determination stage, avoiding interference from sample preparation or retesting uncertainties on the liability determination conclusion, and improving the accuracy and engineering credibility of the anomaly analysis process.

[0046] Specifically, based on the results of link deviation discrimination analysis and coupling suppression recurrence discrimination analysis, the specific process of attribution convergence discrimination analysis is as follows: For the retest result data of the same sample and equipment, the retest result data of the same sample across different equipment, the measurement result data of the substance control, and the substance certificate value data corresponding to the substance control measurement result data, evidence integrity checks are performed respectively. Evidence integrity checks include existence checks, identifier consistency checks, and anomaly window consistency checks. Specifically, for the same anomaly identifier data, the retest result data of the same sample and equipment, the retest result data of the same sample across different equipment, the measurement result data of the substance control, and the corresponding substance certificate value data are sequentially subjected to existence checks and identifier consistency checks. Consistency verification is used to confirm that all types of evidence are complete and correspond to each other in the abnormal event data packet. Under the premise of passing the existence verification and the identifier consistency verification, the collection timestamp of each type of data is subjected to the abnormal event window consistency verification to confirm that all types of evidence point to the same abnormal occurrence stage and to determine whether each type of data exists under the current abnormal event identifier data and whether the time window is consistent. When each type of data passes the integrity verification, the evidence is deemed complete. When at least one type of data fails the integrity verification, the evidence is deemed incomplete. The number of evidence items that pass the verification is counted to obtain the number of evidence items that pass the verification. The ratio of the number of evidence items that pass the verification to the total number of verification items is used as the evidence coverage rate. The evidence coverage rate is subjected to exponential mapping to obtain the real-time evidence completeness.

[0047] The ratio of the material reproduction consistency support discriminant value to the link deviation evidence discriminant value and the sum of the zero-based protection constant is calculated and multiplied by the real-time evidence completeness to obtain the attribution driving quantity. The material reproduction consistency support discriminant value is then used to apply a ratio constraint to the link deviation evidence discriminant value, ensuring that the attribution driving quantity converges when the link deviation evidence discriminant value increases, thus preventing link-side instability from amplifying the attribution conclusion. The attribution driving quantity is then inversely inverted and exponentially calculated. The result of the exponential calculation is then subtracted from the constant to obtain the anomaly attributability discriminant value. This anomaly attributability discriminant value is stably constrained within the range of zero to one through exponential saturation mapping, obtaining a monotonically continuous visual response curve to support online discriminant consistency. The specific calculation formula is as follows: ; In the formula, This represents the attribution value for anomalies, used to continuously quantify the attribution strength of anomalies under the joint constraints of supporting evidence on the material side and deviation evidence on the link side. This represents the material reproducibility consistency support discriminant value, used to provide evidence strength input for the stability of reproducibility of abnormal results and form a positive support term for the attribution driving force; This represents the link deviation evidence discrimination value, which is used to provide the evidence strength input for detecting link deviation risk and to form a suppression constraint term for the attribution driver; It indicates the real-time completeness of evidence and imposes gating constraints on the extent of evidence coverage to avoid outputting unauditable attribution conclusions when evidence is missing; This represents the division-to-zero protection constant, which is calculated from the minimum resolvable quantity corresponding to the resolution of the detection equipment, and its value ranges from 0.0005 to 0.001.

[0048] This implementation plan applies unified integrity checks and time consistency constraints to key retest evidence and control evidence, ensuring that anomalies have a clear, complete, and mutually verifiable evidentiary basis before entering the attribution analysis. This guarantees that attribution judgments are based on a genuine and valid dataset. Furthermore, by gating and coupling material reproduction consistency support, link deviation evidence strength, and real-time evidence completeness, a constrained attribution-driven mechanism is formed. This mechanism allows material-side evidence to automatically converge under conditions of link instability or evidence deficiency, preventing link risks or insufficient evidence from amplifying the attribution conclusion. This process achieves stable output and continuous quantitative expression of anomaly attributability judgment results under different evidentiary states, ensuring the consistency, auditability, and engineering credibility of the attribution conclusions, providing a reliable basis for subsequent handling decisions and source tracing freezes.

[0049] Specifically, the process of outputting the attribution attribution determination conclusion based on the attribution convergence discriminant analysis results and completing the source tracing, version freezing, archiving, and disposal triage is as follows: Figure 4 The diagram shows the flowchart for anomaly attributability judgment and handling decision-making. It compares the anomaly attributability judgment value and the anomaly attributability judgment threshold in real time. The anomaly attributability judgment threshold includes a primary attribution threshold and a secondary attribution threshold. When the evidence is incomplete, freeze the time window of the current abnormal event data packet and the set of input evidence, and supplement the evidence in the order of material control measurement, cross-equipment retesting of the same sample and retesting of the same sample and the same equipment. After the evidence is supplemented, recalculate the attribution value of the abnormality, and create and write it into the abnormality tracing evidence database.

[0050] When the evidence is complete and the attributability threshold for anomalies is greater than or equal to the first-level attribution threshold, a closed-loop material-side handling procedure is executed: an extended retest plan is issued for samples in the same batch corresponding to the batch identification data; the batch test results are issued and put into a pause process and isolated for release. The pause is achieved by setting the batch identification data to a freeze status in the result release module, so that the corresponding test results do not enter the result release and downstream calling channels before the freeze is lifted; the isolation is achieved by batch-level result status marking, which is used to distinguish between frozen abnormal batches and normal batches for parallel processing; the frozen version of the abnormal event data packet is pushed to the quality traceability interface of the smelting and rolling processes to trigger process-side traceability.

[0051] When the evidence is complete and the attributability of the anomaly is less than the secondary attribution threshold, a closed-loop handling procedure is executed on the link side: the issuance of the detection method identification data corresponding to the equipment identification data is suspended. The suspension of issuance is achieved by locking the status of the release control identifier generated by the combination of equipment identification data and detection method identification data, so that newly generated test results under this combination are temporarily stored in the internal queue of the system and not released externally. The calibration verification and material control re-execution are performed on the equipment, and the results already issued within the same time window for the same equipment, the same detection method, are retrospectively recalculated and re-issued. The retrospective recalculation is based on the original test data and the frozen test parameter version to recalculate the result value and generate a new result version to replace the original release version. The material control measurement result data and equipment self-inspection deviation data generated by the correction verification are written into the anomaly tracing evidence database and bound to the equipment identification data to realize the industrial data storage and management of abnormal process data.

[0052] When the evidence is complete and the attributability judgment value of the anomaly is greater than or equal to the secondary attribution judgment threshold and less than the primary attribution judgment threshold, a convergence judgment closed loop is executed: while keeping the batch issuance suspension process unchanged, a dominant judgment is performed on the current link deviation evidence judgment value and the material reproduction consistency support judgment value to determine the dominant support direction of the current anomaly event in the material-side evidence and link-side evidence; based on the dominant support direction, the anomaly event data packet is classified into one of the material-side handling closed loop and the link-side handling closed loop, and the attribution result and the corresponding evidence version are written into the anomaly tracing evidence database. At the same time, the handling status identifier of the anomaly event data packet is updated to constrain the subsequent process to continue only along the determined handling closed loop.

[0053] This implementation plan employs tiered control over the handling of anomaly attributability determination results, ensuring that anomalies enter corresponding standardized processing paths at different stages of evidence status and attribution intensity. This achieves orderly closed-loop management of anomaly analysis, handling, and tracing. When evidence is insufficient, the time window and evidence set of the anomaly are frozen, and evidence is supplemented in a predetermined order to ensure that the attribution conclusion is based on complete and verifiable data. When the attribution intensity clearly points to the material side or the link side, the material side handling closed loop or the link side handling closed loop is triggered respectively, ensuring that the handling actions are consistent with the attribution conclusion and are simultaneously entered into the tracing system. When the attribution intensity is in the intermediate range, the direction of attribution is converged through dominant discrimination, avoiding wavering conclusions or duplicate handling. This mechanism achieves closed-loop control of the entire process from anomaly attribution results to determination, handling, and archiving, ensuring clear anomaly handling paths, traceable evidence versions, and stable and consistent attribution conclusions, thereby improving the overall engineering controllability and audit credibility of anomaly management.

[0054] like Figure 2As shown, the second aspect of the present invention provides an anomaly result tracing system for the steel inspection and testing process, comprising: a data acquisition and preprocessing module for acquiring inspection event result reproduction data and link process evidence data, and preprocessing the inspection event result reproduction data and link process evidence data; a detection link deviation evidence discrimination module for performing link deviation discrimination analysis on the inspection event result reproduction data and link process evidence data, and performing link confirmation and link isolation verification backfilling based on the link deviation discrimination analysis results; a material reproduction consistency support discrimination module for performing coupling suppression reproduction discrimination on the inspection event result reproduction data and link process evidence data, and performing reproduction evidence solidification and evidence reset backfilling based on the coupling suppression reproduction discrimination results; and an anomaly attributability discrimination and tracing freeze output module for performing attribution convergence discrimination analysis based on the link deviation discrimination analysis results and the coupling suppression reproduction discrimination results, outputting anomaly attributability discrimination conclusions based on the attribution convergence discrimination analysis results, and completing the tracing version freeze archiving and disposal distribution.

[0055] This implementation plan modularizes the abnormal inspection and testing handling process, forming a seamless chain of judgment and handling. This creates a complete closed loop from data collection and evidence analysis to accountability determination and source tracing output. The data collection and preprocessing module provides a unified and stable data foundation for subsequent analysis. The detection chain deviation evidence judgment module and the material reproduction consistency support judgment module independently characterize and constrain the causes of abnormalities from the chain side and the material side, respectively. The abnormality attributability judgment and source tracing freeze output module, supported by the combined results of multiple sources, achieves convergence judgment and standardized output of the attribution direction. This overall architecture makes the abnormal analysis process clear in hierarchy, consistent in judgment logic, and controllable in handling path, enabling accurate attribution and traceability management of abnormal results, and improving the stability, reliability, and auditability of steel inspection and testing abnormal handling.

[0056] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0057] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for tracing the source of abnormal results in the steel inspection and testing process, characterized in that, Includes the following steps: S1, Collect and verify event result reproduction data and link process evidence data, and preprocess the event result reproduction data and link process evidence data; S2, perform link deviation discrimination analysis on the data of the test event result reproduction and the evidence data of the link process, and perform link confirmation and link isolation review and backfilling based on the link deviation discrimination analysis results; S3, perform coupling suppression and reproduction discrimination on the event result reproduction data and link process evidence data, and perform reproduction evidence solidification and evidence reset backfilling based on the coupling suppression and reproduction discrimination results; S4. Based on the link deviation discrimination analysis results and coupling suppression reproduction discrimination results, perform attribution convergence discrimination analysis, output the attribution of fault judgment conclusion based on the attribution convergence discrimination analysis results, and complete the source tracing version freezing, archiving, and disposal.

2. The method for tracing the source of abnormal results in the steel inspection and testing process according to claim 1, characterized in that: The specific process for collecting and verifying event result reproduction data and link process evidence data is as follows: Collect test event result reproduction data, which includes: abnormal event identification data, abnormal result value data, retest result value data of the same sample and the same equipment, and retest result value data of the same sample and cross-equipment. Collect evidence data of the link process, which includes: material control measurement results data, material certificate value data, equipment self-inspection deviation data, temperature sequence data of the detection area, vibration acceleration sequence data of the detection area, current sequence data of the sample preparation equipment, and sample preparation start and end timestamp data.

3. The method for tracing the source of abnormal results in the steel inspection and testing process according to claim 1, characterized in that: The specific process for preprocessing the test event result reproduction data and the link process evidence data is as follows: A unified event window truncation and timestamp alignment algorithm is used to align abnormal result values, retest results of the same sample and equipment, retest results of the same sample across different equipment, material control measurement results, equipment self-inspection deviation data, detection area temperature sequence data, detection area vibration acceleration sequence data, and sample preparation equipment current sequence data with the same abnormal event window. The Hampel outlier identification algorithm is used to remove outliers from these data. The Savitzky-Golay algorithm is used to smooth noise in the detection area temperature sequence data, detection area vibration acceleration sequence data, and sample preparation equipment current sequence data. A monotonic cubic interpolation completion algorithm is used to complete short-term missing segments in the detection area temperature sequence data, detection area vibration acceleration sequence data, and sample preparation equipment current sequence data. Finally, a distribution standardization and linear normalization algorithm is used to unify the dimensions of the equipment self-inspection deviation data, detection area temperature sequence data, detection area vibration acceleration sequence data, and sample preparation equipment current sequence data.

4. The method for tracing the source of abnormal results in the steel inspection and testing process according to claim 1, characterized in that: The specific process of performing link bias discriminant analysis on the event result reproduction data and link process evidence data is as follows: The deviation of the material reference measurement results and the material certificate value data are obtained by using a relative deviation calculation algorithm; the deviation data of the equipment self-inspection is obtained, and the deviation results of the zero point verification and the deviation results of the range verification are processed by the maximum value mapping algorithm and normalized to obtain the deviation of the equipment self-inspection; based on the temperature sequence data and vibration acceleration sequence data of the detection area, the abnormal event window sequence is processed by the sliding window peak and valley amplitude extraction algorithm and normalized to obtain the environmental disturbance amount; The ratio of the calculated equipment self-test deviation to the value of one minus the equipment self-test deviation plus the zero-division protection term is added to the constant to obtain the equipment self-test amplification modulation term. Calculate the ratio of the environmental disturbance quantity to the value of one minus the environmental disturbance quantity plus the protection constant divided by zero, and add it to the constant to obtain the environmental disturbance amplification modulation term; calculate the product of the material control deviation quantity, the equipment self-test amplification modulation term and the environmental disturbance amplification modulation term, take the opposite number and perform exponential operation, and then subtract the result of the exponential operation from one to obtain the link deviation evidence discrimination value.

5. The method for tracing the source of abnormal results in the steel inspection and testing process according to claim 1, characterized in that: The specific process of performing link confirmation and link isolation verification and backfilling based on the link deviation discrimination analysis results is as follows: Real-time comparison of link deviation evidence discriminant value and link deviation evidence discriminant threshold: When the link deviation evidence discrimination value is less than the link deviation evidence discrimination threshold, the link confirmation process is executed: without changing the sample, the equipment self-test process is repeated once and the equipment self-test deviation data is retained. At the same time, the material control measurement is repeated once and the material control measurement result data is retained. After updating the data, an abnormal event data packet is generated and input into the coupling suppression reproduction discrimination. When the link deviation evidence discrimination value is greater than or equal to the link deviation evidence discrimination threshold, the link isolation, verification, and backfilling process is executed: the output of the detection method identification data corresponding to the current equipment identification data is suspended, and the equipment self-inspection process and material comparison measurement are re-executed; the vibration acceleration sequence data of the detection area triggers the workstation vibration reduction verification, and the temperature sequence data of the detection area triggers the constant temperature control verification; the equipment self-inspection deviation data and material comparison measurement result data generated by the verification are backfilled into the abnormal event data packet, and the backfilled abnormal event data packet is input into the abnormal attributability judgment and traceability freeze output process.

6. The method for tracing the source of abnormal results in the steel inspection and testing process according to claim 1, characterized in that: The specific process for coupling and suppressing the recurrence of test event result reproduction data and link process evidence data is as follows: The difference between abnormal result data and retest result data of the same sample and the same equipment is obtained by using a relative difference calculation algorithm; the difference between abnormal result data and retest result data of the same sample and cross-equipment is obtained by using a relative difference calculation algorithm; the sample preparation equipment current sequence data and sample preparation start and end timestamp data are used to calculate the current square integral energy within the abnormal event window and normalize it to obtain the sample preparation disturbance. The product of the natural logarithm of the difference between the same sample and the same equipment plus one and the natural logarithm of the difference between the same sample and the cross-equipment is calculated to obtain the coupling term of the logarithm of the difference; the sample preparation disturbance is calculated, added to the protection constant divided by zero, and then added to the constant one to obtain the sample preparation disturbance modulation term. The product of the logarithmic coupling term of the retest difference and the modulation term of the sample preparation disturbance is calculated and exponentially operated on. Then, a constant is added and the reciprocal is taken to obtain the material reproduction consistency support discrimination value.

7. The method for tracing the source of abnormal results in the steel inspection and testing process according to claim 1, characterized in that: The specific process of performing evidence solidification and evidence reset backfilling based on the coupling suppression reproduction discrimination result is as follows: Real-time comparison of the consistency support discriminant value and the consistency support threshold for reproducibility: When the consistency support threshold for reproducibility is less than the consistency support threshold for reproducibility, the sample preparation and retest evidence reset process is executed: the sample preparation equipment is re-performed with no-load self-test and tool status verification, and the sample is re-prepared once at the same sampling location, and the current sequence data of the sample preparation equipment is re-acquired; the same sample and same equipment retest and the same sample and cross-equipment retest are performed n times each on the newly prepared sample; the new round of retest result data and the current sequence data of the sample preparation equipment are backfilled into the abnormal event data package and input into the abnormal attribution judgment and source tracing freeze output process; When the reproducibility consistency support discrimination value is greater than or equal to the reproducibility consistency support discrimination threshold, the material evidence solidification process is executed: the abnormal result value data, the same sample same equipment retest result value data, and the same sample cross equipment retest result value data are solidified into a reproducibility evidence group triple under the same abnormal event identification data. The reproducibility evidence group triple includes unified abnormal event identification data, corresponding evidence data payload, and evidence generation timestamp. An immutable data fingerprint is calculated for the triple to uniquely identify the evidence status. The solidified abnormal event data package is input into the abnormal attributability discrimination and source tracing freeze output process.

8. The method for tracing the source of abnormal results in the steel inspection and testing process according to claim 1, characterized in that: The specific process of performing the attribution convergence discriminant analysis based on the link deviation discriminant analysis results and the coupling suppression recurrence discriminant analysis results is as follows: For the data of the same sample and the same equipment retest results, the data of the same sample and cross-equipment retest results, the data of the material control measurement results, and the material certificate value data corresponding to the material control measurement results, the evidence integrity verification shall be performed respectively. The evidence integrity verification includes existence verification, identification consistency verification and abnormal event window consistency verification. When all types of data pass the integrity verification, the evidence is deemed complete; when at least one type of data fails the integrity verification, the evidence is deemed incomplete. The number of evidence items that pass the verification is counted, and the ratio of the number of evidence items that pass the verification to the total number of verification items is used as the evidence coverage rate. The evidence coverage rate is then subjected to exponential mapping to obtain the real-time evidence completeness. The ratio of the material reproduction consistency support discrimination value to the link deviation evidence discrimination value and the sum of the protection constant divided by zero is calculated and multiplied by the real-time evidence completeness to obtain the attribution driving quantity. The attribution driving quantity is then inversely inverted and exponentially calculated. Finally, the result of the exponential calculation is subtracted from the constant to obtain the anomaly attributability discrimination value.

9. The method for tracing the source of abnormal results in the steel inspection and testing process according to claim 1, characterized in that: The specific process of outputting anomaly attributability judgment conclusions based on the attribution convergence discriminant analysis results and completing the source tracing, version freezing, archiving, and disposal is as follows: Real-time comparison of anomaly attributability criteria values ​​and anomaly attributability criteria thresholds. The anomaly attributability criteria thresholds include primary attribution thresholds and secondary attribution thresholds. When the evidence is incomplete, freeze the time window of the current abnormal event data packet and the set of input evidence, and supplement the evidence in the order of material control measurement, cross-equipment retesting of the same sample and retesting of the same sample and the same equipment. After the evidence is supplemented, recalculate the attribution value of the abnormality, and create and write the supplemented evidence into the abnormality tracing evidence database. When the evidence is complete and the attributability of the anomaly is greater than or equal to the first-level attribution threshold, the material-side handling closed loop is executed: an extended retesting plan is issued for the same batch of samples corresponding to the batch identification data, the batch test results are issued and put into a pause process and isolated release is executed; the frozen version of the abnormal event data packet is pushed to the quality traceability interface of the smelting process and the rolling process to trigger process-side traceability. When the evidence is complete and the attributability of the anomaly is less than the secondary attribution threshold, the following closed-loop handling is implemented on the link side: the issuance of the detection method identification data corresponding to the equipment identification data is suspended; the equipment undergoes calibration verification and material comparison re-execution; and the results already issued within the same time window for the same equipment and detection method are retrospectively recalculated and re-issued. The material comparison measurement results data and equipment self-inspection deviation data generated by the correction verification are written into the anomaly tracing evidence database and bound to the equipment identification data. When the evidence is complete and the attributability of the anomaly is greater than or equal to the secondary attribution threshold and less than the primary attribution threshold, a convergence judgment closed loop is executed: the batch issuance suspension process remains unchanged, and a dominant judgment is performed on the current link deviation evidence judgment value and the material reproduction consistency support judgment value to determine the dominant support direction of the current anomaly in the material-side evidence and the link-side evidence. Based on the dominant support direction, abnormal event data packets are classified into one of the material-side handling closed loop and the link-side handling closed loop, and the accountability results and corresponding evidence versions are written into the abnormal source tracing evidence database.

10. An anomaly traceability system for steel inspection and testing processes, employing the anomaly traceability method for steel inspection and testing processes as described in any one of claims 1-9, comprising: The data acquisition and preprocessing module is used to acquire data on the reproducibility of test event results and evidence data on the link process, and to preprocess the data on the reproducibility of test event results and evidence data on the link process. The link deviation evidence discrimination module is used to perform link deviation discrimination analysis on the test event result reproduction data and link process evidence data, and to perform link confirmation and link isolation review and backfilling based on the link deviation discrimination analysis results; The material reproduction consistency support and discrimination module is used to perform coupling suppression reproduction discrimination on the test event result reproduction data and the link process evidence data, and to perform reproduction evidence solidification and evidence reset backfilling based on the coupling suppression reproduction discrimination result; The anomaly attributability judgment and source tracing freeze output module is used to perform attribution convergence judgment analysis based on the link deviation judgment analysis results and coupling suppression reproduction judgment results. Based on the attribution convergence judgment analysis results, it outputs the anomaly attributability judgment conclusion and completes the source tracing version freeze archiving and disposal distribution.

Images