CN-122022202-A - Continuous casting quality root cause self-healing method, system and program product based on causal map

CN122022202ACN 122022202 ACN122022202 ACN 122022202ACN-122022202-A

Abstract

The invention provides a causal map-based continuous casting quality root cause self-healing method, a causal map-based continuous casting quality root cause self-healing system and a causal map-based continuous casting quality cause self-healing program product, wherein the method comprises the steps of reading an exception handling instruction and outputting a task scheduling result; collecting vibration original data, defect image original data and process parameter original data, cleaning and extracting features, outputting a multi-mode feature sequence, executing Lagrangian time-space alignment on the multi-mode feature sequence, outputting Lagrangian feature tensor, constructing a target dynamic time-space causal graph based on the Lagrangian feature tensor, screening candidate root cause sets, executing inverse fact verification and causal entropy sequencing on the candidate root cause sets based on the target dynamic time-space causal graph, outputting true root cause judgment results, generating a candidate self-healing strategy based on the true root cause judgment results, executing digital twin verification on the candidate self-healing strategy, outputting the target self-healing strategy, executing, collecting and executing feedback data sets and writing back.

Inventors

YU JIONG
YANG JINGJING
ZHANG ZHE
XU QING
CAI HUA

Assignees

华院计算技术(上海)股份有限公司

Dates

Publication Date: 20260512
Application Date: 20260413

Claims (10)

1. The continuous casting quality root cause self-healing method based on the causal map is characterized by comprising the following steps of: s1, reading an exception handling instruction, generating a task request record, performing task decomposition and resource arbitration on the task request record, and outputting a task scheduling result; S2, acquiring vibration original data, defect image original data and process parameter original data based on a task scheduling result, cleaning and extracting features, and outputting a multi-mode feature sequence; S3, carrying out Lagrange space-time alignment on the multi-mode feature sequence, backtracking the defect generation time and the motion section, and outputting Lagrange feature tensor; s4, constructing a target dynamic space-time causal graph based on Lagrange characteristic tensor, and screening a candidate root cause set; and S5, executing anti-fact verification and causal entropy sequencing on the candidate root cause set based on the target dynamic space-time causal graph, outputting a true root cause judgment result, and generating a candidate self-healing strategy based on the true root cause judgment result.
2. The causal map-based continuous casting quality root cause self-healing method according to claim 1, further comprising S6, performing digital twin verification on the candidate self-healing strategy, outputting the target self-healing strategy and executing, collecting and executing a feedback data set and writing back.
3. The causal map-based continuous casting quality root cause self-healing method according to claim 1, wherein S1 specifically comprises: Reading an abnormal handling instruction, a stream number identifier, a steel grade identifier, a defect type identifier, a target quality index and a production plan identifier, generating a task request record, deleting the task request record with missing or repeated fields, and outputting an effective task request set; Inputting the effective task request set into a meta-cognition coordination agent, carrying out semantic analysis, object extraction and task decomposition by combining a task analysis knowledge graph, generating a task arrangement record, deleting the task arrangement record depending on chain breakage, and outputting the effective task arrangement set; And generating a candidate scheduling scheme set based on the effective task arrangement set, the current calculation power, the yield plan value and the energy consumption budget value, calculating the total utility value of the system, deleting the candidate scheduling scheme record lower than the preset lower limit value, and outputting a task scheduling result.
4. The causal map-based continuous casting quality root cause self-healing method according to claim 1, wherein S2 specifically comprises: Reading vibration original data, defect image original data and process parameter original data corresponding to a target heat, a target flow number and a target time window based on a task scheduling result, performing time reference verification and basic cleaning, deleting records with incomplete time coverage, continuous missing points reaching a preset threshold value and exceeding equipment range or image size abnormality, and outputting a preprocessing observation data set; Respectively extracting the health state characteristics, the defect image characteristics and the process state characteristics of the equipment from the preprocessed observation data set, and outputting a multi-mode characteristic sequence; And executing Lagrange time-space alignment based on the multi-mode feature sequence, calculating defect generation time, reconstructing a material slice motion track section, aligning various features according to the section and time steps, and outputting Lagrange feature tensor.
5. The causal map-based continuous casting quality root cause self-healing method according to claim 1, wherein S3 specifically comprises: Reading Lagrangian feature tensors, performing variable mapping according to a preset variable dictionary, generating a system state vector sequence, an observation state slice set and a causal modeling sample set, and deleting records with inconsistent variable dimensions or missing current defect risk value fields; inputting the effective causal modeling sample set into a space-time variation causal discovery engine, constructing a space-time structural equation model, and outputting an initial dynamic space-time causal graph and an effective initial side weight set; and performing continuous constraint optimization on the initial dynamic space-time causal graph and the effective initial side weight set to obtain a target dynamic space-time causal graph, screening ancestor nodes with path communication relation with a target result variable, and outputting a candidate root cause set.
6. The causal map-based continuous casting quality root cause self-healing method according to claim 1, wherein S4 specifically comprises: Reading a target dynamic space-time causal graph, a candidate root cause set and an effective observation state slice set, constructing a fact world network and a counter fact world network, deducing the posterior distribution of exogenous noise, deleting records with noise variance exceeding a preset upper limit value, and outputting an effective exogenous noise deducing set; Executing anti-fact intervention on the candidate variables based on the effective exogenous noise inference set, calculating pure causal effects, deleting the candidate variables with the pure causal effects smaller than a preset threshold, and outputting a true root cause candidate set; And merging the pure causal effect, the defect classification confidence, the equipment anomaly degree and the mechanism consistency identification, calculating the holographic causal entropy and arranging the holographic causal entropy, and outputting a true root cause judgment result.
7. The causal map-based continuous casting quality root cause self-healing method according to claim 1, wherein S5 specifically comprises: Reading a true root cause judgment result, a target dynamic space-time causal graph and an effective observation state slice set, generating a reinforcement learning state set and a constrained action set, constructing an action mask according to an ancestor node set of a target result variable, and outputting an effective action set; executing causal constraint reinforcement learning search based on the reinforcement learning state set and the effective action set, calculating instant reward values of all action combinations, deleting the action combinations exceeding a preset action boundary, and outputting a strategy scoring result; And reading a strategy scoring result, selecting a candidate self-healing strategy, deleting a candidate self-healing strategy record with repeated action combinations, and outputting a candidate self-healing strategy set.
8. The causal map-based continuous casting quality root cause self-healing method according to claim 1, wherein S6 specifically comprises: Reading candidate self-healing strategy records in the candidate self-healing strategy set, synchronously reading steel grade process library records corresponding to steel grade fields, executing virtual trial and error on the candidate self-healing strategy by utilizing a digital twin verification agent, and outputting a digital twin prediction result record; Performing physical safety boundary verification on the digital twin prediction result record, generating a overrule feedback record for candidate self-healing strategies with the safety boundary judging field not passing, returning the overrule feedback record, generating a target self-healing strategy record for candidate self-healing strategies with the safety boundary judging field passing, and outputting a target self-healing strategy set; Reading a target self-healing strategy set, transmitting the target self-healing strategy set to a programmable logic controller for execution, generating a strategy execution record, collecting execution feedback data to form an execution feedback data set, and writing the complete execution feedback data set back to a historical time sequence database.
9. A causal graph-based continuous casting quality root self-healing system, employing the causal graph-based continuous casting quality root self-healing method of any one of claims 1 to 8, comprising: The task scheduling module is used for reading the abnormal handling instruction, generating a task request record, performing task decomposition and resource arbitration on the task request record and outputting a task scheduling result; The data perception module is used for acquiring vibration original data, defect image original data and process parameter original data based on a task scheduling result, cleaning and extracting features of the original data and outputting a multi-mode feature sequence; The space-time alignment module is used for executing Lagrange space-time alignment on the multi-mode feature sequence, backtracking the defect generation time and the material slice motion section, and outputting Lagrange feature tensor; The causal modeling module is used for constructing a target dynamic space-time causal graph based on Lagrange characteristic tensor and screening candidate root cause sets; The root cause judging module is used for executing inverse fact verification and causal entropy sequencing on the candidate root cause set based on the target dynamic space-time causal graph and outputting a true root cause judging result; The strategy generation module is used for generating candidate self-healing strategies based on the true root cause judgment result; And the verification execution module is used for carrying out digital twin verification on the candidate self-healing strategy, outputting the target self-healing strategy and executing the target self-healing strategy, collecting and executing the feedback data set and writing back the feedback data set.
10. A computer program product, characterized in that the computer program product comprises computer program code which, when run on a computer, causes the computer to implement the causal pattern based continuous casting quality root self-healing method according to any one of claims 1 to 8.

Description

Continuous casting quality root cause self-healing method, system and program product based on causal map Technical Field The invention relates to the technical field of intelligent control of quality in a continuous casting process, in particular to a causal graph-based self-healing method, a causal graph-based self-healing system and a causal graph-based self-healing program product for the quality root cause of continuous casting. Background In the continuous casting production process, quality problems such as longitudinal crack, slag rolling, segregation and the like on the surface of a casting blank are generally related to various factors such as drawing speed, liquid level fluctuation, temperature change, cooling water distribution, roll gap state, equipment vibration state, casting powder working condition and the like. The existing field treatment mode is mostly dependent on manual experience, single parameter alarming or post-investigation, and generally judges the abnormal reasons according to the defect detection result and the on-duty process record, and then adjusts the pulling speed, the cooling water quantity or the covering slag scheme. Since the defect forming time is usually earlier than the outlet detecting time, if the current process state corresponding to the detecting time is directly adopted for analysis, the current environment without direct relation is easily misjudged as the causative environment, and the causative positioning is inaccurate. In addition, vibration data, defect image data, and process parameter data of the continuous casting site differ in sampling frequency, temporal granularity, and spatial position. The existing analysis method generally performs simple splicing or rough alignment on multi-source data, and then uses an empirical rule, statistical analysis or conventional model to judge the cause of the abnormality. Although the method can find partial correlation, the method is difficult to accurately reflect the actual historical process environment in the defect forming process and is also difficult to distinguish the actual root cause from the associated correlation factors. Thus, the first and second substrates are bonded together, the invention provides a causal map-based continuous casting quality root cause self-healing method, a causal map-based continuous casting quality root cause self-healing system and a program product. The above information disclosed in the background section is only for enhancement of understanding of the background of the disclosure and therefore it may contain prior art information that is not otherwise known to a person of ordinary skill in the art. Disclosure of Invention The invention aims at overcoming the defects of the prior art and providing a causal map-based continuous casting quality root cause self-healing method, a causal map-based continuous casting quality root cause self-healing system and a causal map-based continuous casting quality root cause self-healing program product, so as to solve the technical problems in the background art. In order to achieve the above purpose, the present invention provides the following technical solutions: The first part of the invention provides a causal map-based continuous casting quality root cause self-healing method, which comprises the following steps: s1, reading an exception handling instruction, generating a task request record, performing task decomposition and resource arbitration on the task request record, and outputting a task scheduling result; S2, acquiring vibration original data, defect image original data and process parameter original data based on a task scheduling result, cleaning and extracting features, and outputting a multi-mode feature sequence; S3, carrying out Lagrange space-time alignment on the multi-mode feature sequence, backtracking the defect generation time and the motion section, and outputting Lagrange feature tensor; s4, constructing a target dynamic space-time causal graph based on Lagrange characteristic tensor, and screening a candidate root cause set; S5, executing inverse fact verification and causal entropy sequencing on the candidate root cause set based on the target dynamic space-time causal graph, outputting a true root cause judgment result, and generating a candidate self-healing strategy based on the true root cause judgment result; S6, carrying out digital twin verification on the candidate self-healing strategy, outputting the target self-healing strategy, executing the target self-healing strategy, collecting and executing a feedback data set, and writing back. The method specifically comprises the steps of reading an abnormal handling instruction, a stream number identifier, a steel grade identifier, a defect type identifier, a target quality index and a production plan identifier, generating a task request record, deleting task request records with missing or repeated fields, outputting an effective task request set, input