Intelligent Optimization Method for Precision Hydraulic Valve Manufacturing Process Based on Causal Inference

Abstract

This application discloses an intelligent optimization method for precision hydraulic valve manufacturing processes based on causal inference. Based on knowledge graphs and full-process data, causal discovery technology is used to identify the causal relationship structure between key process parameters and key quality indicators in the full-process manufacturing data. This causal relationship structure guides feature engineering and model selection, training at least one quality prediction model. After defining constraints and optimization objectives, the quality prediction model, combined with the causal relationship structure, is used to estimate the intervention effect and / or counterfactual effect of process parameter adjustments. This guides the search process of a multi-objective optimization algorithm, obtaining a Pareto optimal set of process parameter combinations that balances the optimization objectives. The intelligent optimization method for precision hydraulic valve manufacturing processes based on causal inference provided by this invention, by mining causal relationships, can go beyond surface correlation and gain a deeper understanding of the true impact mechanism of process parameters on quality.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing, and in particular to an intelligent optimization method for the manufacturing process of precision hydraulic valves based on causal inference, as well as the corresponding device, computer and storage medium. Background Technology

[0002] Hydraulic valves, as core components of high-end equipment, require precise manufacturing and consistent performance. Their manufacturing process is complex, involving the coupled influence of multiple processes and parameters. Optimization methods for hydraulic valve manufacturing processes include trial and error, experimental design, data mining-based correlation analysis, and traditional simulation. While these methods have improved process performance to some extent, each has significant limitations. Lack of deep mechanistic understanding: Most methods struggle to distinguish between the true causal relationship and superficial correlation between process parameters and quality indicators. Correlation-based optimization may adjust parameters that are not the root cause, leading to unstable optimization results or unexpected side effects. Data dependence and generalization issues: Purely data-driven models (especially black-box models) may experience decreased predictive accuracy and reliability of optimization guidance when faced with new operating conditions, material batches, or process drift, as they fail to capture the underlying causal mechanisms. Limited optimization efficiency and reliability: Without causal guidance, optimization searches may be inefficient across a vast parameter space, and the found "optimal solution" may perform poorly or be unstable in actual production due to unconsidered causal chain effects. Insufficient knowledge accumulation and utilization: Implicit process knowledge and patterns discovered through data analysis are difficult to accumulate and utilize in a structured and explicit manner, especially making it difficult to form a systematic understanding of the causal relationships in the process and to continuously iterate.

[0003] Therefore, there is an urgent need for a new intelligent optimization method for precision hydraulic valve manufacturing processes that can go beyond simple data fitting and deeply explore and utilize causal relationships in the manufacturing process, so as to achieve more fundamental, robust and intelligent control of the hydraulic valve manufacturing process. Summary of the Invention

[0004] The main objective of this invention is to provide an intelligent optimization method for the manufacturing process of precision hydraulic valves based on causal inference, aiming to solve the problems of lack of deep understanding of the underlying mechanisms, data dependence, and limited optimization efficiency and reliability in the process of hydraulic valve optimization.

[0005] To achieve the above objectives, this invention provides an intelligent optimization method for the manufacturing process of precision hydraulic valves based on causal inference, comprising: S1. Receive the knowledge graph and full-process data of the manufacturing process, wherein the knowledge graph includes prior knowledge and relationships, and the full-process data is collected during the execution of the manufacturing process and includes process parameters, process status, intervention information, potential confounding factors and quality inspection results. S2. Based on the knowledge graph and the full-process data, causal discovery technology is used to identify the causal relationship structure between key process parameters and key quality indicators in the manufacturing full-process data, wherein the key process parameters are selected from the process parameters and the key quality indicators are selected from the quality inspection results. S3. Based on the causal relationship structure, feature engineering and model selection are guided to train at least one quality prediction model, wherein the quality prediction model predicts the corresponding quality index according to the combination of input process parameters and reflects the causal effect between variables; the causal relationship structure guides feature engineering including: prioritizing the selection of variables with direct or indirect causal paths to the target quality index as features and identifying and handling the influence of confounding factors on the quality prediction model. S4. After defining the constraints and optimization objectives, the intervention effect and / or counterfactual effect of process parameter adjustment are estimated using the quality prediction model and the causal relationship structure. This guides the search process of the multi-objective optimization algorithm to obtain a Pareto optimal process parameter combination solution set that can balance the optimization objective. The Pareto optimal process parameter combination solution set includes multiple Pareto optimal solutions. In the process of using the estimated intervention effect and / or counterfactual effect to guide the search of the multi-objective optimization algorithm, process parameters with strong causal effects on the optimization objective are adjusted first. When evaluating candidate solutions, their stability and potential side effects under the causal model are considered.

[0006] Furthermore, step S4 is followed by: S5. Select a Pareto optimal solution to execute, collect the full process data, update the causal relationship structure, and update the quality prediction model and the knowledge graph with the updated causal relationship structure.

[0007] Furthermore, in step S5, the update strategy in the causal relationship structure is based on at least one selected from periodic time, cumulative data volume threshold, completion of intervention experiment, and degree of performance degradation of quality prediction model.

[0008] Furthermore, the method for updating the causal relationship structure in step S5 is as follows: Compare the actual results of the intervention experiment with the intervention effect predicted by the quality prediction model; Alternatively, it can be tested whether the actual results satisfy the conditional independence relation implied by the updated causal structure.

[0009] Furthermore, the method for estimating the intervention effect and / or counterfactual effect of process parameter adjustment in step S4 includes at least one selected from do-calculus, backdoor adjustment criterion, frontdoor adjustment criterion, instrumental variable method, and model-based counterfactual inference.

[0010] The present invention also provides an apparatus for implementing the above-described intelligent optimization method for precision hydraulic valve manufacturing process based on causal inference, comprising: The acquisition and integration module is used to receive knowledge graphs and full-process data of the manufacturing process; The causal discovery module is used to identify the causal relationship structure between key process parameters and key quality indicators in the manufacturing process data based on the knowledge graph and the full-process data, using causal discovery technology. The prediction model module is used to guide feature engineering and model selection based on the causal relationship structure, and train at least one quality prediction model. The optimization module is used to estimate the intervention effect and / or counterfactual effect of process parameter adjustment by using the quality prediction model and combining the causal relationship structure after defining constraints and optimization objectives. This guides the search process of the multi-objective optimization algorithm to obtain a Pareto optimal combination of process parameters that can balance the optimization objectives.

[0011] The present invention also provides a computer, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described intelligent optimization method for precision hydraulic valve manufacturing process based on causal inference.

[0012] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described intelligent optimization method for precision hydraulic valve manufacturing process based on causal inference.

[0013] This invention provides an intelligent optimization method for precision hydraulic valve manufacturing processes based on causal inference, along with corresponding devices, computers, and storage media. By mining causal relationships, it can go beyond surface correlations and gain a deeper understanding of the true impact mechanism of process parameters on quality, providing a more scientific basis for root cause analysis. Optimization based on causal relationships can focus more on the key parameters that truly play a role, avoiding ineffective adjustments to parameters with only spurious correlations, thus improving optimization efficiency and the reliability of the results. Predictive models incorporating causal relationships or physical mechanisms typically have better extrapolation capabilities and robustness to changes in operating conditions than purely data-driven black-box models. By making the mined causal relationships explicit (e.g., storing them in a knowledge graph) and continuously verifying and correcting them through closed-loop feedback, it achieves the structured accumulation and dynamic evolution of process knowledge. By predicting intervention effects and considering causal side effects, it helps to select more robust and lower-risk optimization schemes. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the process optimization method based on causal inference in the first embodiment of the present invention; Figure 2This is a data distribution diagram of simulated data of sample points in the process optimization method based on causal inference in the second embodiment of the present invention; Figure 3 This is a correlation heatmap of simulated data of sample points in the process optimization method based on causal inference in the second embodiment of the present invention; Figure 4 This is an example of a comparison chart of predicted and actual values ​​of the variability of the coordination gap in the process optimization method based on causal inference in the second embodiment of the present invention; Figure 5 This is an example of a comparison chart between the predicted and actual static leakage values ​​in the process optimization method based on causal inference in the second embodiment of the present invention. Figure 6 This is an example of a Pareto front plot in the process optimization method based on causal inference in the second embodiment of the present invention; Figure 7 This is an example of the distribution diagram of the average intervention effect estimate in the process optimization method based on causal inference in the second embodiment of the present invention; Figure 8 This is a conceptual schematic diagram of an apparatus for running a process optimization method based on causal inference, according to the third embodiment of the present invention.

[0015] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0016] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0017] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” “the,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of the stated features, integers, steps, operations, elements, units, modules, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, units, modules, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connection or wireless coupling. The term “and / or” as used herein includes all or any of the units and all combinations of one or more associated listed items.

[0018] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0019] Reference Figure 1 In one embodiment of the present invention, a method for intelligent optimization of precision hydraulic valve manufacturing process based on causal inference includes: S1. Receive the knowledge graph and full-process data of the manufacturing process, wherein the knowledge graph includes prior knowledge and relationships, and the full-process data is collected during the execution of the manufacturing process and includes process parameters, process status, intervention information, potential confounding factors and quality inspection results. S2. Based on the knowledge graph and the full-process data, causal discovery technology is used to identify the causal relationship structure between key process parameters and key quality indicators in the manufacturing full-process data, wherein the key process parameters are selected from the process parameters and the key quality indicators are selected from the quality inspection results. S3. Based on the causal relationship structure, feature engineering and model selection are guided to train at least one quality prediction model, wherein the quality prediction model predicts the corresponding quality index according to the combination of input process parameters and reflects the causal effect between variables. S4. After defining the constraints and optimization objectives, the intervention effect and / or counterfactual effect of process parameter adjustment are estimated by using the quality prediction model and combining it with the causal relationship structure. This guides the search process of the multi-objective optimization algorithm to obtain a Pareto optimal process parameter combination solution set that can balance the optimization objectives. The Pareto optimal process parameter combination solution set includes multiple Pareto optimal solutions.

[0020] The intelligent optimization method for precision hydraulic valve manufacturing process based on causal inference provided by this invention includes: Step S1: Construct a manufacturing process knowledge system, for example, the manufacturing process knowledge system could be about precision hydraulic valves.

[0021] This step aims to establish a high-quality data foundation and integrate prior knowledge, laying a solid foundation for accurate causal discovery.

[0022] 1. Data Acquisition: Systematically and accurately collect data from the entire manufacturing process to ensure coverage: (a) Process parameters, set values ​​and actual execution values; (b) Process status, including intermediate quality inspection data; (c) Final product quality inspection results; (d) Related information (equipment status, material batch, environment); (e) Clearly and completely record information on interventions to process parameters, including intervention parameters, time points, and adjustment values; (f) Collect all known or domain-knowledge-based data on potential confounding factors.

[0023] 2. Database Construction: Use appropriate database technology to store data throughout the entire process, ensuring data consistency, time-series accuracy, and ease of data association and extraction for causal analysis.

[0024] A process database is a digital system specifically designed for storing, managing, and retrieving various process data involved in manufacturing production. Its core function is to transform scattered process knowledge into structured data, supporting goals such as production process optimization, process standardization, and quality control.

[0025] The core components of the process database include: process parameters, including quantitative indicators that directly affect product quality such as temperature, pressure, and time; process flow, covering the entire process operation steps and sequence design from raw materials to finished products; material properties, storing the physicochemical properties, applicable scenarios, and performance data of materials; quality control specifications, integrating quality assurance rules such as testing standards and process deviation thresholds; and equipment parameters, recording the configuration requirements, operating procedures, and maintenance records of production equipment.

[0026] The main functions of the process database include: data standardization, which solves the problem of fragmented traditional process experience through structured storage and improves data sharing efficiency; process optimization, which identifies production bottlenecks by analyzing historical data and improves process routes and parameter combinations; quality control, which compares production data with preset standards in real time to achieve anomaly warning and closed-loop management; knowledge transfer, which transforms craftsmanship experience into reusable digital assets and reduces the risk of technology loss; and system integration, which serves as the data hub for systems such as MES and PDM to support intelligent manufacturing decision-making.

[0027] 3. Knowledge Graph Construction and Prior Input: The initial knowledge graph is constructed. The key lies in integrating prior causal knowledge with high confidence, confirmed by domain experts (such as deterministic causal relationships caused by physical laws and basic technological principles). This knowledge serves as a mandatory constraint on subsequent causal discovery algorithms to avoid results that clearly violate domain common sense and to improve algorithm efficiency. A knowledge graph uses a graph model (nodes and edges) to describe entities, concepts, and their relationships in the real world. Its core objective is to transform fragmented information into a machine-understandable, interconnected knowledge base to support intelligent reasoning and decision-making.

[0028] The following are some examples of simulation data collected during the manufacturing of precision hydraulic valves: Step S2: Based on the knowledge graph and the entire process data, causal discovery techniques are used to identify the causal relationship structure between key process parameters and key quality indicators in the manufacturing process data. Key process parameters are selected from process parameters, and key quality indicators are selected from quality inspection results. Causal discovery techniques aim to identify causal relationships between variables from observational data and are widely used in fields such as medicine, economics, and artificial intelligence. The main methods include constraint-based, rating-based, function-based causal model-based, and hybrid methods combining domain knowledge.

[0029] The causal discovery technique is selected from at least one of the following: constraint-based methods, scoring-based methods, function-based causal model methods, and hybrid causal discovery methods that combine domain knowledge. Assume the following directed acyclic graph is obtained: H influences P; H influences V (operational habits); G influences sigma_gap; P influences sigma_gap; V influences sigma_gap; F influences Q_leak; sigma_gap influences Q_leak. GBDT models P(sigma_gap|P, V, G, H) and P(Q_leak|sigma_gap, F) are trained to ensure that the model's response to changes in P, V, and G conforms to the causal direction and relative strength indicated by the directed acyclic graph.

[0030] Step S3: Based on the causal relationship structure, feature engineering and model selection are guided to train a quality prediction model, wherein the quality prediction model predicts the corresponding quality indicators according to the combination of input process parameters and reflects the causal effect between variables.

[0031] Feature engineering is a crucial concept in machine learning, referring to the process of extracting, transforming, and selecting features from raw data. These features are the variables input into machine learning models, and their quality directly impacts model performance. The goal of feature engineering is to transform data into a form more suitable for machine learning algorithms, thereby improving the model's predictive ability. Effective feature engineering can significantly improve model performance, and in some cases, it's even more important than choosing a more complex model.

[0032] Model selection is another core concept in machine learning, referring to choosing a suitable machine learning model based on the specific task requirements and optimizing it to achieve optimal performance. Feature engineering and model selection are central parts of the machine learning workflow. Feature engineering enhances the model's learning ability by optimizing the way data is represented, while model selection uses scientific methods to find the most suitable model for the problem. The two complement each other and jointly determine the final performance of the machine learning system.

[0033] Step S4: Perform causal-guided multi-objective process parameter optimization. This step utilizes reliable causal knowledge and quality prediction models for more targeted and reliable optimization.

[0034] 1. Optimization problem definition: Clearly define the optimization objective, decision variables and their range, as well as the constraints that must be satisfied.

[0035] 2. Causal-based optimization strategy: Utilizing multi-objective optimization algorithms such as NSGA-II. The core principle is to leverage causal information to guide the optimization process. The intervention effect refers to the observed change in a system (such as a manufacturing process or production flow) after some intervention (e.g., adjusting parameters) is performed in actual operation. The counterfactual effect refers to the comparative analysis of the results that the system might have produced without intervention and the results after actual intervention.

[0036] Intervention effect estimation: For each candidate solution (combination of process parameters), using the S3 quality prediction model and causal structure, the average intervention effect of adjusting the decision variables on each optimization objective is quantitatively estimated through backdoor adjustment (an effective method for estimating the average intervention effect when the causal structure is clear and confounding factors are measurable) or other applicable methods. This step aims to remove confounding effects, obtain the net effect of parameter adjustment, and provide a causal-based decision-making basis for optimization.

[0037] Guided Search / Evaluation: Integrate the estimated average intervention effect into the optimization algorithm. For example: (a) In the fitness function, in addition to the target predicted value, add a term related to the average intervention effect so that the optimization favors parameter adjustments with significant positive causal effects; (b) In the population evolution or solution selection phase, prioritize retaining or selecting solutions whose key parameters are located in their strong causal effect range.

[0038] 3. Output the optimal solution: Output the Pareto optimal solution set, along with an analysis report on the expected causal effects of key process parameter adjustments, to help engineers understand the causal logic behind the optimization results and make the final decision.

[0039] In summary, the beneficial effects of the present invention are as follows: In-depth process understanding and root cause analysis: By exploring causal relationships, we can go beyond surface correlations and gain a deeper understanding of the true impact mechanism of process parameters on quality, providing a more scientific basis for root cause analysis.

[0040] More guiding optimization: causal optimization can focus more on the key parameters that actually play a role, avoid making ineffective adjustments to parameters with only spurious correlations, and improve optimization efficiency and the reliability of the results.

[0041] Improving model robustness and generalization ability: Predictive models that incorporate causal relationships or physical mechanisms generally have better extrapolation ability and robustness to changes in operating conditions than purely data-driven black-box models.

[0042] Explicit knowledge and continuous accumulation: The discovered causal relationships are made explicit (e.g., stored in a knowledge graph) and continuously verified and corrected through closed-loop feedback, thus realizing the structured accumulation and dynamic evolution of process knowledge.

[0043] Reduce optimization risks: By estimating the intervention effect and considering causal side effects, it is helpful to select more robust and lower-risk optimization solutions.

[0044] In one embodiment, step S4 is followed by: S5. Select a Pareto optimal solution to execute, collect the full process data, update the causal relationship structure, and update the quality prediction model and the knowledge graph with the updated causal relationship structure.

[0045] Implementing process control and causal verification feedback involves actively testing causal hypotheses to enable continuous learning and evolution of the model. Considering that the Pareto optimal solution generally doesn't change much from the initial manufacturing process (especially after setting constraints), the causal relationship structure can be updated using the collected full-process data. This updated causal relationship structure, in turn, updates the quality prediction model and knowledge graph.

[0046] 1. Program Implementation and Intervention Experiment Design: Implement the optimal program. Simultaneously, carefully design and execute small-scale, targeted intervention experiments. Experiments should focus on validating the key (or uncertain) causal relationships identified in S2. For example, only change the strong causal variable in the hypothesis, keeping other variables (especially their parent nodes and confounding factors) as constant as possible, and establish a control group.

[0047] 2. Causal relationship verification: using intervention experiment data: Effect Quantification Validation: Accurately calculate the differences in quality test results between the intervention group and the control group to obtain the experimentally measured average intervention effect. Statistically compare this effect with the average intervention effect predicted by the quality prediction model in S3 (e.g., t-test). Significant differences indicate problems with the causal hypothesis or the model.

[0048] Structural consistency test: This test examines whether the data after intervention still satisfy the independence relationships derived from the causal relationship structure.

[0049] 3. Model and Knowledge Iteration: If the verification fails, you must return to step S2 and correct the causal relationship structure based on the verification result (e.g., add / delete edges, change direction).

[0050] Based on the revised causal structure and the latest dataset including intervention data, the quality prediction model for step S3 is retrained or fine-tuned.

[0051] This closed loop of verification-correction-update is the core mechanism by which this method achieves adaptation and continuous optimization.

[0052] In one embodiment, step S5 is followed by: If key process parameters are duplicated among multiple quality prediction models, the Pareto optimal solution that adjusts the duplicated key process parameters is selected for execution. After collecting the full process data, the causal relationship structure is updated, and the quality prediction model and the knowledge graph are updated with the updated causal relationship structure.

[0053] In this embodiment, considering that the quality prediction model in this invention is formed by multiple models based on a causal relationship structure, there is a high probability that key process parameters will overlap. The Pareto optimal solution related to the overlapping key process parameters is selected for execution, in which case the update of the causal relationship structure can correct the parts involving overlapping key process parameters to the greatest extent.

[0054] In one embodiment, the update strategy in the causal relationship structure in step S5 is based on at least one selected from periodic time, cumulative data volume threshold, completion of intervention experiment, and degree of performance degradation of quality prediction model.

[0055] This embodiment provides multiple monitoring mechanisms to optimize the timing of updates under various specific circumstances. These include: periodic time condition (updating at a fixed time interval); cumulative data threshold condition (updating when a certain judgment parameter reaches a set threshold); intervention experiment completion condition (updating upon completion of each intervention experiment); and quality prediction model performance degradation condition (updating when the quality prediction model's performance degrades to a certain level).

[0056] In one embodiment, the method for updating the causal relationship structure in step S5 is as follows: Compare the actual results of the intervention experiment with the intervention effect predicted by the quality prediction model; Alternatively, it can be tested whether the actual results satisfy the conditional independence relation implied by the updated causal structure.

[0057] In this embodiment, the actual results of the intervention experiment are compared with the intervention effect predicted by the quality prediction model to determine whether the causal relationship structure is correct. Alternatively, it can be used to examine whether the actual results of the intervention experiment still satisfy other conditional independence relationships derived from the causal diagram.

[0058] In one embodiment, step S3, the causal relationship structure-guided feature engineering includes: prioritizing the selection of variables that have direct or indirect causal paths to the target quality index as features, and identifying and addressing the impact of confounding factors on the quality prediction model.

[0059] In this embodiment, feature engineering, an important concept in machine learning, refers to the process of extracting, transforming, and selecting features from raw data. Guiding feature engineering with causal relationship structures ensures that the data extracted from the entire process is more appropriate, accurate, and efficient, resulting in a higher quality prediction model. Furthermore, the entire process data includes confounding factors, and the causal relationship structure also includes relevant causal relationships involving these confounding factors. Therefore, guiding feature engineering with causal relationship structures can include identifying and addressing the impact of confounding factors on the quality prediction model.

[0060] In one embodiment, during the search process of the multi-objective optimization algorithm in step S4, the estimated intervention effect and / or counterfactual effect are used to guide the algorithm. Process parameters with strong causal effects on the optimization objective are adjusted first, while the stability and potential side effects of candidate solutions under the causal model are considered when evaluating candidate solutions.

[0061] In this embodiment, combining the intervention effect and / or counterfactual effect of process parameter adjustments already estimated through the causal relationship structure, the process parameters with a strong causal effect on the optimization objective are prioritized for adjustment during the search process of the multi-objective optimization algorithm. This utilizes the causal relationship structure to adjust the most critical parameters, avoiding over-adjustment of other parameters. Furthermore, considering the stability and potential side effects of candidate solutions under the causal model when evaluating them also improves the robustness of the Pareto optimal process parameter combination solution set.

[0062] In one embodiment, the method for estimating the intervention effect and / or counterfactual effect of process parameter adjustment in step S4 includes at least one selected from do-calculus, backdoor adjustment criterion, frontdoor adjustment criterion, instrumental variable method, and model-based counterfactual inference.

[0063] In this embodiment, several methods are provided for estimating the intervention effect and / or counterfactual effect of process parameter adjustments.

[0064] In one embodiment, step S2 is followed by representing the causal relationship structure as a directed acyclic graph and / or a partial ancestor graph, and storing it in the knowledge graph.

[0065] In this embodiment, the causal relationship structure is updated in the knowledge graph, thereby updating the knowledge. Specifically, it is recorded in the form of a directed acyclic graph and / or a partial ancestor graph, which makes it more readable.

[0066] In one embodiment, in step S2, the causal discovery technique is selected from at least one of constraint-based methods, scoring-based methods, function-based causal model methods, and hybrid causal discovery methods that combine domain knowledge.

[0067] This embodiment provides several common causal discovery techniques. Constraint-based methods construct the causal relationship framework through statistical tests (such as conditional independence tests), and then orient edges based on rules (such as V-structures) to infer causal relationships. Scoring-based methods evaluate the quality of candidate causal relationship structures using scoring functions (such as Bayesian information criterion) and perform optimization searches to select the best structure. Functional causal model-based methods assume that causal relationships are modeled by functional relationships (such as linear or nonlinear functions) and determine the causal direction through noise distribution characteristics (such as independence). Hybrid causal discovery methods that combine domain knowledge integrate constraints, scoring, or functional models with domain knowledge (such as expert experience or prior structures) to overcome data limitations and improve accuracy.

[0068] In one embodiment, in step S2, the key process parameters include data selected from at least one of the following processes: valve body machining, valve core machining, heat treatment, surface treatment, precision assembly, and performance testing. The key quality indicators in step S2 and the optimization objectives in step S4 include at least one selected from dimensional tolerance compliance rate, fit clearance accuracy, surface roughness, seal leakage rate, dynamic response time, pressure-flow characteristics, fatigue life prediction value, pass rate, manufacturing cost, and processing time.

[0069] This embodiment presents the conventional selection methods for key process parameters, key quality indicators, and optimization objectives in the field of valves.

[0070] In one embodiment, in step S3, the quality prediction model is selected from at least one of structural equation modeling and machine learning model incorporating regularization terms or constraints based on causal relationship structures.

[0071] In this embodiment, restrictions are placed on the selection of the quality prediction model, thereby improving the quality of work.

[0072] The present invention also provides an apparatus for running the above-described process optimization method based on causal inference, comprising: The acquisition and integration module is used to receive knowledge graphs and full-process data of the manufacturing process; The causal discovery module is used to identify the causal relationship structure between key process parameters and key quality indicators in the manufacturing process data based on the knowledge graph and the full-process data, using causal discovery technology. The prediction model module is used to guide feature engineering and model selection based on the causal relationship structure, and train at least one quality prediction model. The optimization module is used to estimate the intervention effect and / or counterfactual effect of process parameter adjustment by using the quality prediction model and combining the causal relationship structure after defining constraints and optimization objectives. This guides the search process of the multi-objective optimization algorithm to obtain a Pareto optimal combination of process parameters that can balance the optimization objectives.

[0073] In this embodiment, the operation of the device is the same as that in the aforementioned method embodiment, and will not be repeated here.

[0074] The present invention also provides a computer, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described process optimization method based on causal inference.

[0075] The computer device includes a processor, storage, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The storage includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a file storage method. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad located on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0076] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described process optimization method based on causal inference.

[0077] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media provided in this application and used in the embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory, programmable ROM, electrically programmable ROM, electrically erasable programmable ROM, or flash memory. Volatile memory can include random access memory or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM, dynamic RAM, synchronous DRAM, dual-rate SDRAM, enhanced SDRAM, synchronous link DRAM, memory bus direct RAM, direct memory bus dynamic RAM, and memory bus dynamic RAM, etc. Example

[0078] This embodiment uses a process optimization method based on causal inference to optimize the honing process of the valve core of a precision servo valve. The goal is to simultaneously minimize the dimensional variability of the mating clearance (sigma_gap) and the static leakage (Q_leak).

[0079] S1: Accurately collect the set and actual values ​​of honing pressure (P), reciprocating speed (V), abrasive grit size (G), coolant flow rate (F), process dimensions, material hardness (H), and final sigma_gap and Q_leak. Record detailed manual adjustments (interventions) to P and V. Prior knowledge: H affects P (hardness affects required pressure), sigma_gap affects Q_leak (gap affects leakage).

[0080] S2: Using historical data and prior knowledge, run the PC algorithm to discover causal relationships, assuming the following directed acyclic graph is obtained: H affects P; H affects V (operational habits); G affects sigma_gap; P affects sigma_gap; V affects sigma_gap; F affects Q_leak; sigma_gap affects Q_leak.

[0081] Train GBDT models P(sigma_gap|P,V,G,H) and P(Q_leak|sigma_gap,F) to ensure that the model's response to changes in P, V, and G conforms to the causal direction and relative strength indicated by the directed acyclic graph.

[0082] S3: Optimize the objective min(sigma_gap, Q_leak), with decision variables P, V, and F. Use NSGA-II. When evaluating candidate solutions (P', V', F'), estimate the average intervention effect (P' affects sigma_gap), average intervention effect (V' affects sigma_gap), and average intervention effect (F' affects Q_leak) through backdoor adjustments (using H to control for confounding P and V). Increase the preference for adjustments with significant negative average intervention effects (i.e., reducing sigma_gap or Q_leak) in the fitness assessment.

[0083] S4: Select the optimal solution (P*, V*, F*). Design the intervention experiment: (a) Control group (P_base, V_base, F*); (b) Intervention group P (P*, V_base, F*); (c) Intervention group V (P_base, V*, F*). Compare the differences in the mean sigma_gap measured by each group to verify the accuracy of the estimated average intervention effect (P affects sigma_gap) and average intervention effect (V affects sigma_gap). If the average intervention effect (P affects sigma_gap) is found to be significantly smaller than expected, return to S2. It may be necessary to correct the causal relationship between P and sigma_gap or retrain the model to incorporate more interaction features.

[0084] The following is an overview of a typical simulation result, achieved by simulating the execution of the above steps using a computer program: Step K1: Simulation Data Generation: Refer to Figures 2 to 3 The program generated a simulated dataset containing 1000 sample points, covering material hardness, abrasive grit size, coolant flow rate, honing pressure, reciprocating speed, and target quality indicators (along with gap variability and static leakage). The distribution of the generated data and the correlation between variables were visualized and analyzed using data distribution plots and correlation heatmaps to verify the rationality of the data generation.

[0085] Step K2 Model Training and Evaluation: Based on the assumed causal relationship structure (e.g., material hardness affects honing pressure and reciprocating speed; abrasive grit size, honing pressure, and reciprocating speed jointly affect clearance variability; coolant flow rate and clearance variability jointly affect static leakage), the program trained two gradient boosting regression tree (an ensemble learning algorithm) models to predict clearance variability and static leakage, respectively. Feature selection followed causal guidance principles; for example, the clearance variability prediction model used its direct causes (honing pressure, reciprocating speed, abrasive grit size) and common cause (material hardness) as input features. (Refer to...) Figures 4 to 5After training, the models were evaluated on a reserved test set. The mean squared error (MSE) of the gap variability model (a measure of prediction accuracy) was approximately 0.2336, and the MSE of the static leakage model was approximately 0.8220. The predictive performance of the models was visualized using a comparison graph of predicted and actual values.

[0086] In addition, a feature importance analysis plot was used to show the contribution of each input feature to the model's prediction. The plot used horizontal bars and labeled with specific values ​​for easy understanding.

[0087] Step K3: Multi-objective optimization and average intervention effect estimation: (Refer to...) Figure 6 A non-dominated sorting genetic algorithm II (NSGA-II) was employed to perform multi-objective optimization, aiming to simultaneously minimize the predicted fit clearance variability and static leakage. The optimization process searched for combinations of honing pressure, reciprocating speed, and coolant flow rate, ultimately identifying 100 Pareto optimal solutions. These solutions represent different trade-offs between two conflicting objectives. The resulting Pareto front (representing the distribution of the optimal solution set in the objective space) was visualized using a Pareto front plot.

[0088] For each Pareto optimal solution obtained, the program further utilizes the model and causal structure from step S2 to estimate the average intervention effect of key process parameters (honing pressure, reciprocating speed, and coolant flow rate) relative to baseline values ​​using a simplified backdoor adjustment method (a commonly used causal inference technique). For example, the estimated average intervention effect of adjusting the honing pressure on the variability of the fit clearance is approximately -0.18, the average intervention effect of adjusting the reciprocating speed on the variability of the fit clearance is approximately -0.18, and the average intervention effect of adjusting the coolant flow rate on the static leakage is approximately -1.76. These negative values ​​indicate that, according to the model's predictions and causal assumptions, increasing the corresponding parameters tends to reduce the corresponding quality indicators (these are simulation data results; the actual physical meaning needs to be interpreted in conjunction with domain knowledge). (Refer to...) Figure 8 The distribution of the estimated average intervention effect was visualized using a corresponding distribution plot.

[0089] Step K4 Proof of Concept: As mentioned above, the next step will select the optimal solution for practical verification, and correct the causal relationship structure and model through feedback from intervention experiment data to form a closed-loop optimization.

[0090] This embodiment demonstrates how the method of the present invention combines simulation data generation, causal-guided modeling, multi-objective optimization, and average intervention effect estimation to obtain a series of optimized process parameter combinations, providing quantitative basis and decision support for subsequent practical verification and continuous improvement. This embodiment achieves honing parameter optimization based on a causal verification closed loop, ensuring the scientific validity and reliability of the optimization decisions.

[0091] In summary, the intelligent optimization method for precision hydraulic valve manufacturing processes based on causal inference provided by this invention, along with the corresponding device, computer, and storage medium, can go beyond surface correlations by mining causal relationships, gaining a deeper understanding of the true impact mechanism of process parameters on quality, and providing a more scientific basis for root cause analysis. Optimization based on causal relationships can focus more on the key parameters that truly play a role, avoiding ineffective adjustments to parameters with only spurious correlations, thus improving optimization efficiency and the reliability of the results. Predictive models incorporating causal relationships or physical mechanisms typically have better extrapolation capabilities and robustness to changes in operating conditions than purely data-driven black-box models. Explicitizing the mined causal relationships (e.g., storing them in a knowledge graph) and continuously verifying and correcting them through closed-loop feedback enables the structured accumulation and dynamic evolution of process knowledge. By estimating intervention effects and considering causal side effects, it helps to select more robust and lower-risk optimization schemes.

[0092] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for intelligent optimization of precision hydraulic valve manufacturing process based on causal inference, characterized in that, include: S1. Receive the knowledge graph and full-process data of the manufacturing process, wherein the knowledge graph includes prior knowledge and relationships, and the full-process data is collected during the execution of the manufacturing process and includes process parameters, process status, intervention information, potential confounding factors and quality inspection results. S2. Based on the knowledge graph and the full-process data, causal discovery technology is used to identify the causal relationship structure between key process parameters and key quality indicators in the manufacturing full-process data, wherein the key process parameters are selected from the process parameters and the key quality indicators are selected from the quality inspection results. S3. Based on the causal relationship structure, feature engineering and model selection are guided to train at least one quality prediction model, wherein the quality prediction model predicts the corresponding quality index according to the combination of input process parameters and reflects the causal effect between variables; the causal relationship structure guides feature engineering including: prioritizing the selection of variables with direct or indirect causal paths to the target quality index as features and identifying and handling the influence of confounding factors on the quality prediction model. S4. After defining the constraints and optimization objectives, the intervention effect and / or counterfactual effect of process parameter adjustment are estimated using the quality prediction model and the causal relationship structure. This guides the search process of the multi-objective optimization algorithm to obtain a Pareto optimal process parameter combination solution set that can balance the optimization objective. The Pareto optimal process parameter combination solution set includes multiple Pareto optimal solutions. In the process of using the estimated intervention effect and / or counterfactual effect to guide the search of the multi-objective optimization algorithm, process parameters with strong causal effects on the optimization objective are adjusted first. When evaluating candidate solutions, their stability and potential side effects under the causal model are considered.

2. The optimization method according to claim 1, characterized in that, The step S4 is followed by: S5. Select a Pareto optimal solution to execute, collect the full process data, update the causal relationship structure, and update the quality prediction model and the knowledge graph with the updated causal relationship structure.

3. The optimization method according to claim 2, characterized in that, The update strategy in the causal relationship structure in step S5 is based on at least one selected from periodic time, cumulative data volume threshold, completion of intervention experiment, and degree of performance degradation of quality prediction model.

4. The optimization method according to claim 2, characterized in that, The method for updating the causal relationship structure in step S5 is as follows: Compare the actual results of the intervention experiment with the intervention effect predicted by the quality prediction model; Alternatively, it can be tested whether the actual results satisfy the conditional independence relation implied by the updated causal structure.

5. The optimization method according to claim 1, characterized in that, The method for estimating the intervention effect and / or counterfactual effect of process parameter adjustment in step S4 includes at least one selected from do-calculus, backdoor adjustment criterion, frontdoor adjustment criterion, instrumental variable method, and model-based counterfactual inference.

6. An apparatus for performing the optimization method according to any one of claims 1 to 5, characterized in that, include: The acquisition and integration module is used to receive knowledge graphs and full-process data of the manufacturing process; The causal discovery module is used to identify the causal relationship structure between key process parameters and key quality indicators in the manufacturing process data based on the knowledge graph and the full-process data, using causal discovery technology. The prediction model module is used to guide feature engineering and model selection based on the causal relationship structure, and train at least one quality prediction model. The optimization module is used to estimate the intervention effect and / or counterfactual effect of process parameter adjustment by using the quality prediction model and combining the causal relationship structure after defining constraints and optimization objectives. This guides the search process of the multi-objective optimization algorithm to obtain a Pareto optimal combination of process parameters that can balance the optimization objectives.

7. A computer, comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the optimization method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the optimization method according to any one of claims 1 to 5.

Images