Detection method for evaluating sintering quality of artificial heat-conducting film

Abstract

The invention provides a detection method for evaluating the sintering quality of an artificial heat-conducting film. The method comprises the following steps: constructing a personalized database containing electrical conductivity data corresponding to the artificial heat-conducting film under different materials and different process conditions; identifying performance fluctuation parameters of different sintering production devices and the same sintering production device in different production periods by analyzing the conductivity data; the method comprises the following steps: arranging multiple types of sensors on artificial heat-conducting film sintering production equipment, and collecting key parameters in the sintering production equipment; fusing the multiple types of key parameters through a data fusion algorithm to obtain a fused data set; the sintering production equipment is adjusted according to the performance fluctuation parameters; and inputting the key parameters in the fusion data set into the sintering quality evaluation model, and comparing the key parameters with the standard parameters of the personalized database to obtain a comparison result. Through the quality evaluation model and the automatic feedback mechanism, intelligent adjustment of the sintering process is achieved, and the product quality and the production efficiency are improved.

Description

Technical Field

[0001] The invention relates to the technical field of artificial graphite heat-conducting film preparation, and in particular to a detection method for evaluating the sintering quality of an artificial heat-conducting film. Background Art

[0002] The main raw material used for artificial graphite thermal conductive film is PI, and the main preparation process is PI slitting, PI rewinding, carbonization, graphitization and calendering. The most critical processes are carbonization and graphitization. Carbonization and graphitization are heat treatment processes, in which the graphitization process needs to experience a high temperature of 3000℃. The temperature measurement method uses infrared non-contact temperature measurement, which is limited by the furnace environment and the product environment, resulting in slightly poor reliability and a certain deviation from the set temperature curve. (For example, the poor exhaust of individual equipment causes unequal concentrations of smoke in the furnace, and the different pollution levels of the light-transmitting window cause measurement errors); the deviations between the equipment and the deviations of a single equipment in different time periods cause poor product consistency, and the qualified rate of some high-end sensitive products is seriously affected and cannot meet the requirements. The key measurement method for the quality of artificial graphite film is thermal conductivity. This test requires professional testers, professional equipment and professional environment, and the test efficiency is extremely low; artificial graphite thermal conductive film has excellent electrical conductivity and thermal conductivity, and its thermal and electrical conductivity are directly related to the quality of graphitization. At present, the measuring tools for testing the conductivity are reliable and efficient, and the requirements for measuring personnel are not high. By establishing a database for comparing the thermal and electrical conductivity of products of different quality and formula, a product conductivity judgment standard is established to guide the quality level of graphitized products. This method can achieve the effect of fast and efficient judgment. Prepare N copies of the same material, the same batch, and the same pre-treatment product as product monitoring (which can eliminate material fluctuations and monitoring equipment fluctuations). For products sintered with the same material, one copy is followed each time sintering. By measuring the monitored products, SPC analysis is established to quickly identify equipment abnormalities by comparing their conductivity, or quickly identify whether the equipment status has a deterioration trend, and the engineering staff will make timely corrections. However, there are certain defects in the current equipment capacity judgment of this process. Products with defects occur occasionally under the premise that all indicators of the equipment are qualified. In addition, the product differences between the equipment are large, and the quality risks brought by individual equipment cannot be identified for highly sensitive products.

[0003] Prior art 1, application number: CN202411558285.0 discloses a graphene thermal conductive film material quality detection and analysis method, by detecting the appearance performance, thermal conductivity, physical properties, chemical properties, mechanical properties and electrical properties of the graphene thermal conductive film, and analyzing the detected data, respectively constructing the evaluation index of the corresponding performance, further comprehensively evaluating the material quality of the graphene thermal conductive film, and comprehensively detecting the material quality of the graphene thermal conductive film from multiple dimensions, thereby improving the quality evaluation system of the graphene thermal conductive film. Although it is conducive to better guaranteeing the material quality and quality of the graphene thermal conductive film; and generating a material quality detection report for the graphene thermal conductive film, so that enterprises can find problems existing in the production process of the graphene thermal conductive film, thereby providing a basis for enterprises to improve production processes and optimize product design; however, by constructing an evaluation index for data analysis, on the one hand, the evaluation index lacks the ability to update itself, resulting in certain errors in the data analysis results; on the other hand, its intelligent analysis ability is poor, and it cannot adapt to the current intelligent production of graphene thermal conductive films.

[0004] Prior art 2, application number: CN 202411222530.0 discloses a method for measuring thermal conductivity of thermal conductive films based on an improved steady-state electrothermal method. The improved steady-state electrothermal method is used to obtain the in-plane thermal conductivity of the thermal conductive film based on the temperature distribution on the surface of the thermal conductive film at different powers. The measurement error is less than 3.0%, the uncertainty is reduced to 0.5%, and the response time is at the millisecond level. Although it provides useful guidance for accurately evaluating the thermal conductivity of materials, it also provides technical support for the thermal management engineering application of graphite film materials; however, its calculation process is complicated and requires many parameters, which puts forward high requirements on professional computing capabilities.

[0005] Prior art three, application number: CN 202411480722.1 discloses a graphene thermal conductive film material inferior screening system based on image recognition. Based on the image of the graphene thermal conductive film, the graphene thermal conductive film surface thickness unevenness coefficient and component distribution unevenness coefficient are obtained to analyze the surface unevenness of the graphene thermal conductive film; and the information of holes, cracks, wrinkles, ripples, and edge defects on the surface of the graphene thermal conductive film is obtained to analyze the surface defects of the graphene thermal conductive film; and the information of foreign matter, oxidation and corrosion on the surface of the graphene thermal conductive film is obtained to analyze the surface impurities of the graphene thermal conductive film; the surface quality of the graphene thermal conductive film is further comprehensively evaluated to determine whether the surface quality of the graphene thermal conductive film is qualified, inferior products are screened, and a quality inspection report is generated. Although the appearance of the graphene thermal conductive film is tested and evaluated from multiple dimensions to ensure the surface quality of the graphene thermal conductive film material; however, screening through image recognition puts forward higher requirements on the image shooting quality, which increases the cost of image recognition to a certain extent.

[0006] At present, the existing technologies 1, 2 and 3 have the problem that the detection means are relatively simple and the operation is cumbersome, so that the accuracy of the detection results needs to be further improved. Therefore, the present invention provides a detection method for evaluating the sintering quality of artificial thermal conductive films. Summary of the invention

[0007] In order to solve the above technical problems, the present invention provides a detection method for evaluating the sintering quality of an artificial thermal conductive film, which is characterized by comprising the following steps:

[0008] Build a personalized database containing the corresponding conductivity performance data of artificial thermal conductive films under different materials and different process conditions; through the analysis of conductivity performance data, identify the performance fluctuation parameters of different sintering production equipment and the same sintering production equipment in different production periods;

[0009] Multiple types of sensors are deployed on the artificial thermal conductive film sintering production equipment to collect key parameters of temperature, pressure and time in the sintering production equipment; multiple types of key parameters are fused through data fusion algorithm to obtain a fused data set; the sintering production equipment is adjusted according to the performance fluctuation parameters;

[0010] The key parameters in the fused data set are input into the sintering quality assessment model and compared with the standard parameters in the personalized database to obtain the comparison results. Among them, the comparison results include which key parameters deviate from the standard range, and the comparison results are automatically fed back to the sintering production equipment for corresponding adjustments.

[0011] Optionally, the process of analyzing the conductivity performance data includes the following steps:

[0012] Build a hybrid storage architecture for personalized databases, where the basic data layer stores static data on material properties and sintering production equipment parameters; the process data layer stores dynamic data during the sintering process; and the performance data layer stores conductive performance test results and related indicators;

[0013] Clean the material data, process data, equipment data and conductive performance data, and establish corresponding classification index, attribute index, numerical index and time index according to key fields;

[0014] According to the index, key patterns related to conductive properties are extracted from the personalized database, and data points that deviate from normal patterns are identified; the performance fluctuation parameters corresponding to the data points are decomposed into influencing factors at different levels, and the changes in conductive properties are decomposed into microstructural changes at the material level.

[0015] Optionally, the process of obtaining a fused data set includes the following steps:

[0016] Normalize key parameters such as temperature, pressure and time collected by different sensors, and evaluate the influence of each key parameter on sintering quality by calculating the covariance matrix; dynamically adjust the weight distribution using adaptive weight adjustment factors according to the real-time status of the sintering production equipment;

[0017] The normalized key parameters are weighted fused according to dynamic weights to generate a preliminary fused data set, outliers in the fused data set are removed, and the fused data set without outliers is smoothed;

[0018] The optimized fused data set is standardized to meet the input requirements of the sintering quality assessment model, and the standardized fused data set is output to the sintering quality assessment model for comparison with the standard parameters of the personalized database.

[0019] Optionally, the process of building a sintering quality assessment model includes the following steps:

[0020] The multi-type sensor data of sintering production equipment is mapped into a multi-dimensional state space, and dynamic weights are assigned to each key parameter based on the historical data of different materials and process conditions in the personalized database; the action space adopts a composite structure, including two levels of macro adjustment and micro optimization, and the reward function adopts a hierarchical design;

[0021] The adaptive exploration strategy dynamically adjusts the exploration rate according to the uncertainty of the current state; at the same time, the strategy gradient optimizes multiple goals such as sintering quality, equipment energy consumption and production efficiency, realizing the combination of action value evaluation and multi-objective optimization of strategy gradient; receives sensor data in real time and updates the adaptive exploration strategy;

[0022] If the adjusted sintering quality still does not meet expectations, the sintering quality evaluation model automatically triggers the self-correction mechanism, re-evaluates the definition of the state space and action space, and optimizes the reward function.

[0023] Optionally, the reward function adopts a hierarchical design process, including the following steps:

[0024] Design the base layer rewards, select the key parameters for sintering quality, calculate the deviation between the actual value and the standard value of each key parameter; penalize the parameters with large deviations to ensure that the model tends to control the parameters within the standard range; through historical performance trend analysis, give additional incentives for long-term quality improvement to design optimization layer rewards;

[0025] The base layer reward and the optimization layer reward are combined to obtain a comprehensive reward function. To adapt to different materials and process conditions, the dynamic weight is adjusted according to historical data and current status.

[0026] When the sintering quality assessment model triggers the self-correction mechanism, the reward function is re-evaluated and optimized, and the definition of key parameters and action space is adjusted according to changes in the current production environment; the dynamic weights are recalculated based on the latest data; and the weight coefficients of the base layer reward and the optimization layer reward are adjusted based on the optimization results.

[0027] Optionally, the process of combining action value evaluation with policy gradient multi-objective optimization includes the following steps:

[0028] Sintering quality, equipment energy consumption and production efficiency are determined as core optimization targets, and the multi-objective optimization problem is transformed into a unified evaluation system. Different targets are integrated into a comprehensive optimization function through weighting. Action value evaluation is used to quantify the contribution of each action to multi-objective optimization, and the action value evaluation at the macro and micro levels is combined to form a composite evaluation system.

[0029] Decompose the multi-objective optimization problem into multiple sub-objectives, each of which corresponds to an independent policy gradient optimization path; by weighting, merge the policy gradients of each sub-objective into a comprehensive gradient to guide the update direction of the strategy; dynamically adjust the weight of each sub-objective according to the current state and goal priority;

[0030] Dynamically adjust the exploration rate based on the uncertainty of the current state and the optimization progress; receive sensor data in real time and update the exploration strategy.

[0031] Optionally, the process of dynamically adjusting the exploration rate and updating the exploration strategy includes the following steps:

[0032] The current state is evaluated through multi-dimensional uncertainty quantification; the target conflict index is used to measure the mutual influence between the optimization targets, and the historical data deviation reflects the degree of deviation between the current observation value and the historical trend; through comprehensive calculation, a dynamic uncertainty score is generated;

[0033] Nonlinear optimization progress evaluation is used to deeply analyze the optimization progress; sub-goal convergence speed is used to evaluate the optimization efficiency of each sub-goal, while the strategy stability index measures the robustness of the current strategy under different states, dynamically identifies bottlenecks in the optimization process, and adjusts the exploration strategy;

[0034] Based on the uncertainty score and optimization progress evaluation results, the exploration rate is adjusted according to the target priority weight and the environment complexity factor.

[0035] The optional process of re-evaluating the definitions of the state space and action space consists of the following steps:

[0036] Based on the current quality deviation and optimization goals, re-evaluate the contribution of each parameter in the multi-dimensional state space and identify the key parameters that have a significant impact on quality; assign new dynamic weights to key parameters based on real-time sensor data and historical performance trends; incorporate external environmental variables and equipment status into the state space;

[0037] Redefine the magnitude and priority of macro-adjustment actions according to the severity of the current quality deviation; fine-tune local parameters, local temperature, and control actions for pressure gradient adjustment;

[0038] According to the distribution characteristics of the current quality deviation, the weights of the deviations between key parameters and the standard range are readjusted; a dynamic threshold mechanism is introduced to dynamically adjust the standard range according to historical performance trends; and additional incentives are given for long-term quality improvement through historical performance trend analysis.

[0039] Optionally, the process of dynamically adjusting the standard range based on historical performance trends includes the following steps:

[0040] Based on historical performance data, identify the regular characteristics of its changes over time and extract the key features in the historical performance trends; use trend models to predict the possible range of changes in key parameters in the future;

[0041] Compare the actual values ​​of current key parameters with historical performance trends to assess their degree of deviation;

[0042] Based on the real-time performance evaluation results and combined with trend forecast information, the threshold range is dynamically adjusted; if the current parameter value is higher than the historical trend forecast value, the upper threshold limit is increased; if it is lower than the forecast value, the lower threshold limit is lowered.

[0043] Optionally, the process of comparing the current actual values ​​of key parameters with historical performance trends includes the following steps:

[0044] Time-align the actual values ​​of current key parameters with historical performance trends; standardize the actual values ​​of current key parameters and historical performance trends to eliminate dimension and unit differences; extract key features from historical performance trends as a benchmark for comparison;

[0045] Match the actual value of the current key parameter with the key features of the historical performance trend to identify their performance differences in different dimensions; calculate the absolute difference between the actual value of the current key parameter and the predicted value of the historical performance trend to reflect the magnitude of the deviation; compare the absolute deviation with the fluctuation range of the historical performance trend to calculate the relative deviation rate to reflect the severity of the deviation;

[0046] Analyze the consistency between the actual value change trend of the current key parameters and the historical performance trend to determine whether they deviate from the expected direction; comprehensively score the degree of deviation of the actual value of the current key parameters based on multi-dimensional indicators such as absolute deviation and relative deviation rate; and divide the degree of deviation into different levels based on the comprehensive scoring results.

[0047] The present invention constructs a personalized database, collects the conductive performance data under different materials and process conditions, and forms a standardized database; through data analysis, the performance fluctuation parameters of different sintering production equipment and the same equipment in different production periods are identified; the law of performance fluctuation of sintering production equipment is identified, and a reference basis is provided for quality evaluation and adjustment of sintering production equipment. Multi-type sensor data acquisition and fusion, multiple types of sensors are arranged on the sintering production equipment to collect key parameters such as temperature, pressure and time in real time; through the data fusion algorithm, multiple types of key parameters are integrated into a unified fusion data set to improve the integrity and accuracy of the data; the sintering production equipment is adjusted according to the performance fluctuation parameters to ensure that the sintering production equipment operates in the best state. Sintering quality evaluation and feedback adjustment, the key parameters of the fusion data set are input into the sintering quality evaluation model, and compared with the standard parameters of the personalized database; the key parameters that deviate from the standard range are identified to provide a quantitative basis for quality evaluation; according to the comparison results, the sintering production equipment is automatically fed back to make corresponding adjustments to ensure stable sintering quality.

[0048] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description and the accompanying drawings.

[0049] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0051] Figure 1 Flow chart of the detection method for evaluating the sintering quality of the artificial thermal conductive film in Example 1 of the present invention;

[0052] Figure 2 is a process diagram for analyzing the conductive performance data in Example 2 of the present invention;

[0053] Figure 3 A process diagram of obtaining a fused data set in Example 3 of the present invention;

[0054] Figure 4This is a process diagram of constructing a sintering quality assessment model in Example 4 of the present invention;

[0055] Figure 5 This is a process diagram of a hierarchical design of a reward function in Embodiment 5 of the present invention;

[0056] Figure 6 A process diagram for realizing the combination of action value evaluation and policy gradient multi-objective optimization in Embodiment 6 of the present invention;

[0057] Figure 7 This is a process diagram of dynamically adjusting the exploration rate and updating the exploration strategy in Example 7 of the present invention;

[0058] Figure 8 A process diagram for re-evaluating the definition of the state space and the action space in Embodiment 8 of the present invention;

[0059] Fig. 9 A process diagram for dynamically adjusting the standard range according to the historical performance trend in Example 9 of the present invention;

[0060] Fig.10 This is a process diagram for comparing the actual value of the current key parameter with the historical performance trend in Example 10 of the present invention. DETAILED DESCRIPTION

[0061] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0062] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the embodiments of the present application. The singular forms of "a", "said" and "the" used in the embodiments of the present application are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used in this article refers to and includes any or all possible combinations of one or more associated listed items.

[0063] When the following description relates to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. On the contrary, they are only examples of devices and methods consistent with some aspects of the present application. In the description of the present application, it should be understood that the terms "first", "second", "third", etc. are only used to distinguish similar objects, and do not have to be used to describe a specific order or sequence, nor can they be understood as indicating or implying relative importance. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to the specific circumstances.

[0064] Example 1: Figure 1 As shown, an embodiment of the present invention provides a detection method for evaluating the sintering quality of an artificial thermal conductive film, comprising the following steps:

[0065] S100: Build a personalized database containing the corresponding conductivity performance data of artificial thermal conductive films under different materials and different process conditions; through the analysis of conductivity performance data, identify the performance fluctuation parameters of different sintering production equipment and the same sintering production equipment in different production periods;

[0066] S200: Multiple types of sensors are deployed on the artificial thermal conductive film sintering production equipment to collect key parameters such as temperature, pressure and time in the sintering production equipment; multiple types of key parameters are fused through data fusion algorithm to obtain a fused data set; the sintering production equipment is adjusted according to the performance fluctuation parameters;

[0067] S300: Input the key parameters in the fused data set into the sintering quality assessment model and compare them with the standard parameters in the personalized database to obtain the comparison results; the comparison results include which key parameters deviate from the standard range, and automatically feed back to the sintering production equipment based on the content of the comparison results for corresponding adjustments.

[0068] The working principle and beneficial effects of the above technical solution are as follows: In this embodiment, a personalized database containing the corresponding conductive performance data of artificial thermal conductive films under different materials and different process conditions is first constructed; by analyzing the conductive performance data, the performance fluctuation parameters of different sintering production equipment and the same sintering production equipment in different production periods are identified; secondly, multiple types of sensors are arranged on the artificial thermal conductive film sintering production equipment to collect key parameters such as temperature, pressure and time in the sintering production equipment; multiple types of key parameters are fused through a data fusion algorithm to obtain a fused data set; the sintering production equipment is adjusted according to the performance fluctuation parameters; finally, the key parameters in the fused data set are input into the sintering quality evaluation model, and compared with the standard parameters of the personalized database to obtain a comparison result; among them, which key parameters contained in the comparison result deviate from the standard range are automatically fed back to the sintering production equipment according to the content of the comparison result, and corresponding adjustments are made. Step S100 of the above scheme builds a personalized database, collects conductive performance data under different materials and process conditions, and forms a standardized database; through data analysis, identifies the performance fluctuation parameters of different sintering production equipment and the same equipment in different production periods; identifies the law of performance fluctuation of sintering production equipment, and provides a reference basis for quality assessment and adjustment of sintering production equipment. Significance achieved: Provide reliable data support for quality assessment and process optimization; by identifying the performance fluctuation of sintering production equipment, provide a basis for sintering production equipment maintenance and process improvement, and reduce uncertainty in production. Step S200 Multi-type sensor data acquisition and fusion, deploy multi-type sensors on sintering production equipment, and collect key parameters such as temperature, pressure and time in real time; through data fusion algorithm, integrate multiple types of key parameters into a unified fusion data set to improve data integrity and accuracy; adjust the sintering production equipment according to the performance fluctuation parameters to ensure that the sintering production equipment operates in the best state. Significance achieved: Achieve all-round monitoring of the sintering process to ensure that key parameters are within a reasonable range; reduce the error of a single sensor and improve data reliability through data fusion; adjust the sintering production equipment according to the performance fluctuation parameters to improve the stability and consistency of the operation of the sintering production equipment. Step S300 Sintering quality evaluation and feedback adjustment, input the key parameters of the fused data set into the sintering quality evaluation model, and compare them with the standard parameters of the personalized database; identify key parameters that deviate from the standard range to provide a quantitative basis for quality evaluation; based on the comparison results, automatically feedback to the sintering production equipment to make corresponding adjustments to ensure stable sintering quality. Significance achieved: Ensure that the quality of the sintering process meets the standards and reduce the defective rate through real-time evaluation and feedback; realize automated adjustment of the production process to improve production efficiency and consistency.

[0069] In summary, this embodiment achieves accurate monitoring and optimization of the sintering process of artificial thermal conductive film through three steps: building a personalized database, collecting and fusing multi-type sensor data, and sintering quality assessment and feedback adjustment. Its significance is mainly reflected in: providing a reliable basis for quality assessment through data accumulation and analysis; achieving comprehensive monitoring of the sintering process through multi-type sensors and data fusion; and achieving intelligent adjustment of the sintering process through quality assessment models and automatic feedback mechanisms to improve product quality and production efficiency. This method can significantly improve production efficiency and product quality, and has important practical application value.

[0070] Example 2: Figure 2 As shown, based on Example 1, the process of analyzing the conductive performance data provided by the embodiment of the present invention includes the following steps:

[0071] S101: Build a hybrid storage architecture for a personalized database, where the basic data layer stores static data such as material properties and sintering production equipment parameters; the process data layer stores dynamic data during the sintering process, such as temperature, pressure and time series; and the performance data layer stores conductive performance test results and related indicators;

[0072] S102: Clean the material data, process data, equipment data and conductive performance data, and establish corresponding classification indexes, attribute indexes, numerical indexes and time indexes according to key fields;

[0073] S103: Extract key patterns related to conductive performance from the personalized database according to the index, and identify data points that deviate from normal patterns; decompose the performance fluctuation parameters corresponding to the data points into influencing factors at different levels, and decompose the changes in conductive performance into microstructural changes at the material level.

[0074] The working principle and beneficial effects of the above technical solution are as follows: This embodiment first builds a hybrid storage architecture of a personalized database, in which the basic data layer stores static data such as material properties and sintering production equipment parameters; the process data layer stores dynamic data in the sintering process, such as temperature, pressure and time series; the performance data layer stores the conductivity test results and related indicators; secondly, the material data, process data, equipment data and conductivity data are cleaned, and corresponding classification indexes, attribute indexes, numerical indexes and time indexes are established according to key fields; finally, key patterns related to conductivity are extracted from the personalized database according to the index, and data points that deviate from the normal pattern are identified; the performance fluctuation parameters corresponding to the data points are decomposed into influencing factors at different levels, and the changes in conductivity are decomposed into microstructural changes at the material level. Step S101 of the above scheme builds a hybrid storage architecture of a personalized database. By dividing the data into a basic data layer, a process data layer, and a performance data layer, it realizes the classified management of static data, dynamic data, and performance data, and improves the logical clarity and retrieval efficiency of data storage; combining the advantages of relational databases and non-relational databases, it can simultaneously process structured data (such as process parameters) and unstructured data (such as production logs, images), meeting diverse data storage needs; through hierarchical design, it clarifies the logical relationship between different data layers, providing a basis for data association analysis and pattern recognition. Significance: Hierarchical storage and hybrid architecture enable the database to efficiently store and manage massive amounts of data, providing a solid foundation for analysis; by clarifying the correlation between data, it creates conditions for multi-dimensional analysis of conductive properties and identification of influencing factors; the design of the dynamic data layer enables the database to record changes in the sintering process in real time, providing support for real-time monitoring and dynamic adjustment of the production process. Step S102: Data cleaning and index construction. Through data cleaning, duplicate, missing and abnormal data are removed, ensuring the integrity, consistency and accuracy of the data, and providing a high-quality data foundation for analysis. By establishing classification indexes, attribute indexes, numerical indexes and time indexes for key fields, the data retrieval speed and query efficiency are significantly improved, especially when processing large-scale data. The construction of indexes not only improves the retrieval efficiency, but also enhances the correlation between data, providing convenience for pattern recognition and factor decomposition. Significance: High-quality data is the prerequisite for the reliability of analysis results. Data cleaning ensures the scientificity and credibility of the analysis results. The construction of indexes enables the database to respond quickly to complex queries, improving the overall performance and user experience of the system. Through different types of indexes, data can be analyzed from multiple dimensions, providing support for the multi-factor correlation analysis of conductive properties.Step S103: Pattern recognition and factor decomposition. Through dynamic pattern recognition algorithms, key patterns related to conductive performance, such as periodic fluctuations, nonlinear trends, and abnormal patterns, are extracted from massive data, providing a scientific basis for performance analysis. Data points that deviate from normal patterns are identified, which can timely detect abnormalities in the production process and provide clues for fault diagnosis and performance optimization. Performance fluctuation parameters are decomposed into influencing factors at different levels, such as microstructural changes at the material level, to clarify the root causes of changes in conductive performance and provide direction for process optimization. Significance: Through pattern recognition and factor decomposition, the inherent laws of changes in conductive performance can be deeply explored, providing support for the scientific management of the production process. The identification of abnormal points can timely detect problems in the production process, providing a basis for dynamic adjustment and fault prevention of sintering production equipment. By clarifying the influencing factors, scientific suggestions can be provided for the optimization and adjustment of process parameters, thereby improving the conductive performance and production quality of the product.

[0075] In summary, this embodiment realizes efficient management and logical association of data by building a hybrid storage architecture; improves data quality and retrieval efficiency through data cleaning and index construction, creating conditions for multi-dimensional analysis; and deeply explores the inherent laws of changes in conductive properties through pattern recognition and influencing factor decomposition, providing a scientific basis for the optimization and adjustment of the production process. The three steps are closely linked and together constitute a complete process for conductive performance data analysis, which not only improves the efficiency and accuracy of data analysis, but also provides strong support for scientific management and process optimization of the production process. It embodies the deep integration of technology and practice, and has important practical application value and innovative significance.

[0076] Example 3: Figure 3 As shown, based on Example 1, the process of obtaining a fused data set provided in the embodiment of the present invention includes the following steps:

[0077] S201: normalize key parameters such as temperature, pressure and time collected by different sensors, and evaluate the influence of each key parameter on the sintering quality by calculating the covariance matrix; dynamically adjust the weight distribution using an adaptive weight adjustment factor according to the real-time status of the sintering production equipment;

[0078] The formula for normalization is:

[0079]

[0080] In the formula, x ij represents the original value of the jth key parameter in the i-th fused data set; μ j represents the mean value of the jth key parameter; σ j represents the standard deviation of the jth key parameter; x ij′represents the normalized parameter value; n represents the total number of fused data sets;

[0081] Align the time series of different sensors, and for the time series T of the jth key parameter j , the interpolation function is expressed as:

[0082] S j (t) = a j t 3 +b j t 2 +c j t+d j

[0083] In the formula, S j (t) represents the interpolation result of the jth key parameter at time t; a j ,b j ,c j ,d j represents the interpolation coefficient;

[0084] Covariance matrix expression:

[0085] C′=C+δ·C 2

[0086]

[0087] Where C represents the original covariance matrix, δ represents the high-order covariance coefficient; the expanded covariance matrix C′ is decomposed by eigenvalues, and a regularization term is introduced to prevent overfitting:

[0088] C′=VΛ′V T +ηI

[0089]

[0090] In the formula, Λ′ represents the expanded eigenvalue matrix, η represents the regularization coefficient; I represents the unit matrix;

[0091] The importance weight of each key parameter w j To indicate:

[0092]

[0093] In the formula, λ j represents the jth key parameter eigenvalue;

[0094] The adaptive weight adjustment factor is expressed as:

[0095] w j′ =w j ·(1+α j ΔPj )

[0096] ΔP j =|x ij′ -μ j′ |

[0097]

[0098] In the formula, ΔP j Indicates the real-time fluctuation range of the jth key parameter; μ j′ represents the real-time mean of the jth parameter; α j represents the adjustment factor;

[0099] S202: performing weighted fusion on the normalized key parameters according to dynamic weights to generate a preliminary fused data set, removing outliers in the fused data set, and performing smoothing on the fused data set without outliers;

[0100] in,

[0101]

[0102] In the formula, F i represents the fusion value of the i-th fusion data set;

[0103] The nonlinear fusion function is expanded to represent:

[0104]

[0105] Where β1, β2, β3 represent nonlinear fusion coefficients, which are determined by minimizing the fusion error; represents the trigonometric function term introduced to capture periodic changes; (F i -μ F ) 3 Represents the cubic polynomial term introduced to capture higher-order nonlinear relationships.

[0106] The nonlinear fusion coefficients β1, β2, and β3 are made dynamic so that they can be adjusted as the data distribution changes. The dynamic coefficient formula is expressed as:

[0107]

[0108] In the formula, γ k represents the dynamic adjustment factor; λ k is the eigenvalue of the kth nonlinear term; σ j′ Represents the real-time standard deviation of the jth key parameter;

[0109] Formula for removing outliers in the fused data set:

[0110]

[0111] In the formula, ∑ represents the covariance matrix of the fused data set; if Then remove the data point, is the 99% quantile of the chi-square distribution;

[0112] The formula for smoothing the fused data set is:

[0113]

[0114] In the formula, represents the smoothing value of the i-th fused data set; K i represents the Kalman gain, is the estimation error covariance at the previous moment, and R is the measurement noise covariance;

[0115] S203: standardizing the optimized fused data set to make it meet the input requirements of the sintering quality assessment model, and outputting the standardized fused data set to the sintering quality assessment model for comparison with standard parameters of the personalized database;

[0116] Fusion dataset normalization F i″ formula:

[0117]

[0118] In the formula, Represents the mean of the smoothed fused data set; Represents the standard deviation of the smoothed fused data set.

[0119] The working principle and beneficial effects of the above technical solution are as follows: In this embodiment, the key parameters such as temperature, pressure and time collected by different sensors are normalized, and the influence of each key parameter on the sintering quality is evaluated by calculating the covariance matrix; according to the real-time status of the sintering production equipment, the weight distribution is dynamically adjusted using the adaptive weight adjustment factor; secondly, the normalized key parameters are weighted and fused according to the dynamic weights to generate a preliminary fusion data set, the outliers in the fusion data set are removed, and the fusion data set with the outliers removed is smoothed; finally, the optimized fusion data set is standardized to meet the input requirements of the sintering quality assessment model, and the standardized fusion data set is output to the sintering quality assessment model for comparison with the standard parameters of the personalized database. Step S201 of the above solution is normalization and dynamic weight adjustment. By normalizing the key parameters such as temperature, pressure and time collected by different sensors, data of different dimensions are unified to the same scale; the normalization formula ensures that the mean of the data is 0 and the standard deviation is 1, thereby eliminating the influence of dimensional differences on subsequent analysis; the normalized data are comparable, which is convenient for weighted fusion and covariance analysis. Significance: Lay the foundation for the fusion of multi-source data and ensure the fairness and consistency of data from different sensors. Covariance matrix and eigenvalue decomposition evaluate the influence of each key parameter on the sintering quality by calculating the covariance matrix. Introduce high-order covariance terms and regularization terms to prevent overfitting and ensure the stability of the model; the covariance matrix reveals the correlation between parameters, and the eigenvalue decomposition extracts the importance of key parameters. Significance: Provide a basis for dynamic weight allocation to ensure that important parameters dominate the fusion process. Adaptive weight adjustment uses adaptive weight adjustment factors to dynamically adjust weight allocation according to the real-time status of sintering production equipment; the weight changes dynamically with data fluctuations, which can capture changes in real-time production status. Significance: Improve the flexibility and adaptability of fused data and ensure that the model can respond to changes in the production environment in a timely manner. Step S202 weighted fusion and outlier processing, weighted fusion of normalized key parameters according to dynamic weights to generate a preliminary fusion data set; the fused data set integrates multi-source information and can more comprehensively reflect the sintering quality. Significance: Provide a high-quality data basis for outlier detection and smoothing. Nonlinear fusion extension captures the periodicity and high-order nonlinear relationships in the data by introducing trigonometric function terms and cubic polynomial terms; it enhances the expressive power of the fusion model and can more accurately describe complex data features. Significance: Improves the accuracy and reliability of fused data and provides richer information for sintering quality assessment. Outlier removal removes outliers in the fused data set by calculating the chi-square distance, removes noise data, and ensures the purity of the fused data. Significance: Improves the reliability of data and avoids the interference of outliers on the evaluation results.Smoothing processing uses Kalman filtering to smooth the fused data set to reduce data fluctuations; the smoothed data is more stable and can better reflect the long-term trend of sintering quality. Significance: Provide high-quality data for standardization processing and model input. Step S203 standardization processing and model input, standardize the optimized fused data set to meet the input requirements of the sintering quality assessment model; the standardized data has a uniform scale and distribution, which is convenient for model processing and analysis. Significance: Ensure the standardization of model input and improve the accuracy and comparability of evaluation results; model input and comparison output the standardized fused data set to the sintering quality assessment model, and compare it with the standard parameters of the personalized database; through comparison, deviations and anomalies in sintering quality can be quickly identified. Significance: Provide a scientific basis for the optimization and adjustment of the production process to ensure the stability and consistency of sintering quality.

[0120] In summary, the generation process of the entire fusion data set of this embodiment converts multi-source heterogeneous data into a high-quality and high-reliability fusion data set through a series of technical means such as normalization, weighted fusion, outlier processing, smoothing and standardization. It not only improves the availability and analysis efficiency of data, but also provides a solid data foundation for the accurate evaluation and optimization of sintering quality. Its significance lies in the realization of intelligent processing of production data, promoting the refined management and high-quality development of sintering production; information density is improved through multi-source data fusion and optimization, generating a high information density data set containing key parameters such as temperature, pressure and time, providing comprehensive data support for sintering quality assessment. Dynamic adaptability Through dynamic weight adjustment and nonlinear fusion, the algorithm can adapt to the real-time state changes of sintering production equipment, and improve the accuracy and reliability of the fusion data set. Abnormal handling capability Through outlier detection and Kalman filtering based on Mahalanobis distance, the algorithm can effectively reduce noise interference and ensure the high quality of the fusion data set. Standardized output Through standardized processing, the fusion data set can be directly input into the sintering quality assessment model, simplifying the analysis process.

[0121] Example 4: Figure 4 As shown, based on Example 1, the process of constructing a sintering quality assessment model provided by the embodiment of the present invention comprises the following steps:

[0122] S301: Map the multi-type sensor data of the sintering production equipment into a multi-dimensional state space, and assign dynamic weights to each key parameter based on the historical data of different materials and process conditions in the personalized database; the action space adopts a composite structure, including two levels of macro adjustment (such as overall temperature setting) and micro optimization (such as local pressure fine-tuning), and the reward function adopts a hierarchical design;

[0123] Among them, macro adjustments are used to quickly respond to significant quality deviations, while micro optimization is used for refined control. The reward function adopts a hierarchical design. The basic layer rewards are based on the degree of deviation between key parameters and the standard range, and the optimization layer rewards provide additional incentives for long-term quality improvement through historical performance trend analysis.

[0124] S302: The adaptive exploration strategy dynamically adjusts the exploration rate according to the uncertainty of the current state; at the same time, the strategy gradient optimizes multiple goals such as sintering quality, equipment energy consumption and production efficiency, realizing the combination of action value evaluation and multi-objective optimization of strategy gradient; receiving sensor data in real time and updating the adaptive exploration strategy;

[0125] S303: If the adjusted sintering quality still does not meet expectations, the sintering quality evaluation model automatically triggers the self-correction mechanism, re-evaluates the definition of the state space and action space, and optimizes the reward function.

[0126] The working principle and beneficial effects of the above technical solution are as follows: In this embodiment, the multi-type sensor data of the sintering production equipment are first mapped into a multidimensional state space, and dynamic weights are assigned to each key parameter according to the historical data of different materials and process conditions in the personalized database; the action space adopts a composite structure, including two levels: macro adjustment (such as overall temperature setting) and micro optimization (such as local pressure fine-tuning); wherein, macro adjustment is used to quickly respond to significant quality deviations, and micro optimization is used for refined control; the reward function adopts a hierarchical design, the basic layer reward is based on the degree of deviation of the key parameters from the standard range, and the optimization layer reward is given additional incentives for long-term quality improvement through historical performance trend analysis; secondly, the adaptive exploration strategy dynamically adjusts the exploration rate according to the uncertainty of the current state; at the same time, the strategy gradient optimizes multiple objectives such as sintering quality, equipment energy consumption and production efficiency, and realizes the combination of action value evaluation and multi-objective optimization of strategy gradient; sensor data is received in real time and the adaptive exploration strategy is updated; finally, if the adjusted sintering quality still does not meet expectations, the sintering quality evaluation model automatically triggers the self-correction mechanism, re-evaluates the definition of state space and action space, and optimizes the reward function. Step S301 of the above scheme is the design of state space, action space and reward function; Multi-dimensional state space mapping By mapping multi-type sensor data into multi-dimensional state space, the model can fully capture the key parameters in sintering production and provide a high-quality data basis for decision-making. Dynamic weight allocation assigns dynamic weights to key parameters according to historical data, so that the model can adapt to different materials and process conditions and improve the generalization ability of the model; The combination of macro adjustment and micro optimization in composite action space design enables the model to respond quickly to significant quality deviations and achieve refined control, thereby improving the flexibility and efficiency of decision-making; The basic layer reward of the hierarchical reward function ensures that the key parameters are within the standard range, and the optimization layer reward encourages long-term quality improvement, providing a clear optimization goal for the model. Significance achieved: Data-driven precise modeling Through multi-dimensional state space and dynamic weight allocation, the model can more accurately reflect the actual situation of sintering production; Flexible and efficient decision-making ability Composite action space design enables the model to meet the needs of different scenarios and improve the adaptability and efficiency of decision-making. Multi-objective optimization-oriented hierarchical reward function provides the model with an optimization goal that takes into account both short-term quality and long-term performance, ensuring the sustainable development of the model. Step S302: Adaptive exploration strategy and multi-objective optimization. The adaptive exploration strategy dynamically adjusts the exploration rate according to the uncertainty of the current state, so that the model can learn efficiently in an unknown environment while avoiding waste of resources caused by excessive exploration. Multi-objective optimization is combined with strategy gradient optimization to optimize multiple objectives such as sintering quality, equipment energy consumption and production efficiency. The model can achieve multi-objective collaborative optimization in complex scenarios. The real-time update mechanism receives sensor data in real time and updates the adaptive exploration strategy, so that the model can quickly adapt to changes in the production environment and improve real-time decision-making capabilities.Significance achieved: The adaptive exploration strategy minimizes resource consumption while ensuring learning efficiency; Multi-objective collaborative optimization Through the combination of multi-objective optimization, the model can improve the sintering quality while taking into account equipment energy consumption and production efficiency to maximize the comprehensive benefits; Real-time adaptability The real-time update mechanism enables the model to quickly respond to changes in the production environment and improve the practicality and robustness of the model. Step S303 Self-correction mechanism triggering and optimization, self-correction mechanism triggering When the adjusted sintering quality still does not meet expectations, the model automatically triggers the self-correction mechanism, re-evaluates the definition of the state space and action space, and optimizes the reward function; Dynamic optimization capability Through the self-correction mechanism, the model can continuously optimize its own performance in actual production and improve long-term stability and robustness. Significance achieved: Dynamic optimization and self-improvement The self-correction mechanism enables the model to continuously optimize its own performance in actual applications and improve long-term stability and robustness; Continuous improvement and enhanced adaptability Through dynamic optimization of the state space, action space and reward function, the model can better adapt to the complex and changing production environment and ensure long-term efficient operation.

[0127] In summary, this embodiment provides the model with data-driven accurate modeling and flexible and efficient decision-making capabilities, laying the foundation for the model; through the combination of adaptive exploration strategy and multi-objective optimization, the learning efficiency and comprehensive optimization ability of the model are improved; through the self-correction mechanism, the dynamic optimization and self-improvement of the model are realized, ensuring the long-term stability and robustness of the model. The three steps together constitute the complete construction process of the sintering quality assessment model, enabling the model to achieve efficient and intelligent decision-making optimization in a complex and changeable sintering production environment.

[0128] Example 5: Figure 5 As shown, based on Example 4, the reward function provided by the embodiment of the present invention adopts a hierarchical design process, which includes the following steps:

[0129] S3011: Design the basic layer reward, select the key parameters for sintering quality, calculate the deviation between the actual value and the standard value of each key parameter; penalize the parameters with large deviations to ensure that the model tends to control the parameters within the standard range; through historical performance trend analysis, give additional incentives for long-term quality improvement to design optimization layer rewards;

[0130] Assume there are N key parameters, and the actual value of the jth key parameter is x j , the standard value is The deviation is

[0131] Base layer reward R base It is expressed as:

[0132]

[0133] Where w′ j is the dynamic weight of the jth key parameter, which is adjusted dynamically according to historical data and process conditions; the negative sign indicates that the greater the deviation, the smaller the reward (the greater the penalty);

[0134] Assume that within the time window T, the historical average value of the key parameter j is The current value is x j ;

[0135] Optimization layer reward R opt It can be expressed as:

[0136]

[0137] In the formula, α j is the long-term optimization weight of the jth key parameter, which is dynamically adjusted according to the historical performance trend; the positive sign indicates that the reward increases when the current value is better than the historical average;

[0138] S3012: Combine the base layer reward and the optimization layer reward to obtain a comprehensive reward function; to adapt to different materials and process conditions, the dynamic weight is adjusted according to historical data and current status;

[0139] R=R base +β·R opt

[0140] Where β is the weight coefficient of the optimization layer reward, which is used to balance short-term quality control and long-term performance improvement;

[0141] In order to adapt to different materials and process conditions, the dynamic weight w′ j and α j Need to be adjusted based on historical data and current status;

[0142]

[0143] In the formula, is the historical variance of the jth key parameter, reflecting its volatility; is the historical average change trend of the jth key parameter;

[0144] S3013: When the sintering quality assessment model triggers the self-correction mechanism, the reward function is re-evaluated and optimized, and the definition of key parameters and action space is adjusted according to the changes in the current production environment; the dynamic weights are recalculated according to the latest data; and the weight coefficients of the base layer reward and the optimization layer reward are adjusted according to the optimization results.

[0145] The working principle and beneficial effects of the above technical solution are as follows: First, the basic layer reward is designed in this embodiment, and the key parameters for sintering quality are selected to calculate the deviation between the actual value and the standard value of each key parameter; the parameters with large deviations are penalized to ensure that the model tends to control the parameters within the standard range; through historical performance trend analysis, the optimization layer reward is given additional incentives for long-term quality improvement; secondly, the basic layer reward and the optimization layer reward are combined to obtain a comprehensive reward function; to adapt to different materials and process conditions, the dynamic weight is adjusted according to historical data and current status; finally, when the sintering quality evaluation model triggers the self-correction mechanism, the reward function is re-evaluated and optimized, and the definition of key parameters and action space is adjusted according to the changes in the current production environment; the dynamic weight is recalculated according to the latest data; according to the optimization results, the weight coefficients of the basic layer reward and the optimization layer reward are adjusted. In the design of the basic layer reward and the optimization layer reward in step S3011 of the above scheme, by calculating the deviation between the actual value and the standard value of the key parameters and penalizing the parameters with large deviations, the basic layer reward can ensure that the model tends to control the parameters within the standard range, and effectively realize the control of short-term quality fluctuations. Significance: The introduction of the base layer reward provides the model with a clear optimization goal, namely, reducing the deviation of key parameters, thereby ensuring the stability and consistency of sintering quality; the adjustment of dynamic weights further enhances the adaptability of the model, enabling it to flexibly adjust the penalty intensity according to different process conditions and historical data. The optimization layer reward provides additional incentives for long-term quality improvement by comparing the current value with the historical average; when the current value is better than the historical average, the reward increases, thereby promoting the continuous optimization of the model in long-term performance. Significance: The introduction of the optimization layer reward makes up for the limitations of the base layer reward, focusing not only on short-term quality fluctuations, but also on long-term performance improvement; the design helps to promote continuous improvement and innovation in the sintering process. Step S3012 The construction of the comprehensive reward function combines the base layer reward and the optimization layer reward. The comprehensive reward function can take into account both short-term quality control and long-term performance improvement; by dynamically adjusting the weight coefficient, the model can find a balance between different goals according to actual needs. Significance: The design of the comprehensive reward function enables the model to flexibly respond to complex production environments, which not only ensures the stability of sintering quality, but also provides motivation for long-term optimization; the introduction of dynamic weights further enhances the adaptability and robustness of the model. Step S3013: Triggering of the self-correction mechanism and optimization of the reward function. When the sintering quality assessment model triggers the self-correction mechanism, the reward function will be re-evaluated and optimized according to changes in the current production environment; by adjusting the definition of key parameters and action space, and recalculating dynamic weights, the model can quickly adapt to new production conditions. Significance: The introduction of the self-correction mechanism significantly improves the adaptability and intelligence of the model; it can not only cope with changes in the production environment, but also dynamically adjust the optimization strategy according to the latest data to ensure that the sintering quality is always in the best state.

[0146] In summary, the hierarchical design of the reward function of this embodiment achieves the dual goals of short-term quality control and long-term performance improvement through the combination of base layer rewards and optimization layer rewards; the introduction of dynamic weights enables the model to flexibly adapt to different process conditions and production environments, while the self-correction mechanism further enhances the intelligence and adaptability of the model. It not only improves the stability and consistency of sintering quality, but also provides strong support for the continuous optimization and innovation of the process; the design process of the reward function takes into account both short-term quality control and long-term performance improvement through a hierarchical structure; the base layer rewards ensure that key parameters are within the standard range, and the optimization layer rewards encourage long-term quality improvement; through dynamic weight adjustment and self-correction mechanisms, the reward function can adapt to complex and changing production environments and provide clear optimization goals for the model.

[0147] Example 6: Figure 6 As shown, based on Example 4, the process of realizing the combination of action value evaluation and policy gradient multi-objective optimization provided by the embodiment of the present invention includes the following steps:

[0148] S3021: Determine sintering quality, equipment energy consumption and production efficiency as core optimization objectives, transform the multi-objective optimization problem into a unified evaluation system, and integrate different objectives into a comprehensive optimization function through weighting; action value evaluation is used to quantify the contribution of each action to multi-objective optimization, and combine the action value evaluation at the macro and micro levels to form a composite evaluation system;

[0149] S3022: Decompose the multi-objective optimization problem into multiple sub-objectives, each of which corresponds to an independent policy gradient optimization path; merge the policy gradients of each sub-objective into a comprehensive gradient by weighting to guide the update direction of the policy; dynamically adjust the weight of each sub-objective according to the current state and target priority;

[0150] S3023: Dynamically adjust the exploration rate according to the uncertainty of the current state and the optimization progress; receive sensor data in real time and update the exploration strategy.

[0151] The working principle and beneficial effects of the above technical solution are as follows: This embodiment first determines sintering quality, equipment energy consumption and production efficiency as core optimization targets, transforms the multi-objective optimization problem into a unified evaluation system, and integrates different targets into a comprehensive optimization function through weighting; action value evaluation is used to quantify the contribution of each action to multi-objective optimization, and combines the action value evaluations at the macro and micro levels to form a composite evaluation system; secondly, the multi-objective optimization problem is decomposed into multiple sub-targets, and each sub-target corresponds to an independent policy gradient optimization path; through weighting, the policy gradients of each sub-target are merged into a comprehensive gradient to guide the update direction of the strategy; according to the current state and target priority, the weight of each sub-target is dynamically adjusted; finally, according to the uncertainty of the current state and the optimization progress, the exploration rate is dynamically adjusted; sensor data is received in real time to update the exploration strategy. The step S3021 of the above scheme is to build a multi-objective optimization framework and evaluate the action value, which integrates the sintering quality, equipment energy consumption and production efficiency into a comprehensive optimization function, avoiding the local optimal problem that may be caused by single-objective optimization; by combining the action value evaluation at the macro and micro levels, it not only ensures the efficiency of global optimization, but also achieves the accuracy of local control; dynamically adjust the weight of each objective according to the current production status and priority, and ensure that the optimization process can flexibly adapt to the needs of different scenarios. The significance achieved: Through a unified evaluation system, comprehensive optimization of multiple objectives is achieved, which significantly improves the overall performance of sintering production; the composite evaluation system enables the model to focus on global and local optimization at the same time, improving the precision and adaptability of control; the dynamic weight mechanism enables the optimization process to be flexibly adjusted according to actual needs, enhancing the practicality and robustness of the model. Step S3022 is the multi-objective optimization of the policy gradient, which decomposes the multi-objective optimization problem into multiple sub-objectives and simplifies the complexity of the optimization problem; by weighting, the policy gradients of each sub-objective are merged into a comprehensive gradient, ensuring the unity and consistency of the optimization direction; dynamically adjust the weight of each sub-objective according to the current state and target priority, so that the optimization process can adapt to the needs of different scenarios. Significance achieved: Objective decomposition and gradient fusion enable multi-objective optimization problems to be solved efficiently, improving the convergence speed and optimization effect of the model; dynamic weight adjustment ensures the balance and coordination between sub-objectives, avoiding the conflict problems that may be caused by single-objective optimization; multi-objective optimization of policy gradients enables the model to dynamically adjust the optimization strategy according to actual needs, significantly improving the intelligence level of the model. The introduction of step S3023 adaptive exploration strategy dynamically adjusts the exploration rate according to the uncertainty of the current state and the optimization progress, ensuring the balance between exploration and utilization in the optimization process; by receiving sensor data in real time and updating the exploration strategy, the model can quickly respond to changes in the production environment.Significance achieved: Dynamically adjusting the exploration rate enables the model to increase exploration when uncertainty is high to discover better actions; reducing exploration when optimization tends to be stable to improve efficiency; the real-time feedback mechanism enables the model to quickly adapt to changes in the production environment, significantly improving the real-time and adaptability of the model; the introduction of the adaptive exploration strategy enables the optimization process to maintain efficiency and stability in a complex and changeable production environment, enhancing the robustness of the model.

[0152] In summary, this embodiment realizes the combination of multi-objective optimization of action value evaluation and policy gradient, and achieves the following overall significance: Comprehensiveness and refinement Through a unified evaluation system and a composite evaluation system, comprehensive optimization and refined control of sintering quality, equipment energy consumption and production efficiency are realized. Efficiency and intelligence Through target decomposition, gradient fusion and dynamic weight adjustment, efficient solution and intelligent adjustment of multi-objective optimization problems are realized. Real-time and adaptability Through adaptive exploration strategies and real-time feedback mechanisms, the model can quickly respond to changes in the production environment, significantly improving the real-time and adaptability of the model. Robustness and stability Through dynamic weight mechanisms and self-correction mechanisms, the model can maintain high efficiency and stability in complex and changeable production environments, enhancing the robustness of the model. It not only significantly improves the performance and practicality of the sintering quality evaluation model, but also provides innovative solutions for multi-objective optimization problems in complex industrial scenarios. This embodiment constructs a hierarchical and highly adaptable sintering quality assessment model by combining action value evaluation with multi-objective optimization of policy gradient; through the combination of multi-objective optimization framework, action value evaluation, policy gradient optimization, adaptive exploration strategy and self-correction mechanism, it realizes comprehensive optimization of sintering quality, equipment energy consumption and production efficiency, with significant technical creativity and practicality.

[0153] Example 7: Figure 7 As shown, based on Example 6, the process of dynamically adjusting the exploration rate and updating the exploration strategy provided by the embodiment of the present invention includes the following steps:

[0154] S30231: Evaluate the current state through multi-dimensional uncertainty quantification; the target conflict index is used to measure the mutual influence between the optimization goals, and the historical data deviation reflects the degree of deviation between the current observation value and the historical trend; through comprehensive calculation, a dynamic uncertainty score is generated;

[0155] S30232: Use nonlinear optimization progress evaluation to deeply analyze the optimization progress; the sub-goal convergence speed is used to evaluate the optimization efficiency of each sub-goal, and the strategy stability index measures the robustness of the current strategy under different states, dynamically identifies bottlenecks in the optimization process, and adjusts the exploration strategy;

[0156] S30233: Based on the uncertainty score and optimization progress evaluation results, adjust the exploration rate according to the target priority weight and environment complexity factor.

[0157] The working principle and beneficial effects of the above technical solution are as follows: the present embodiment first evaluates the current state through multi-dimensional uncertainty quantification; the target conflict index is used to measure the mutual influence between each optimization target, and the historical data deviation reflects the degree of deviation between the current observation value and the historical trend; a dynamic uncertainty score is generated through comprehensive calculation; secondly, the nonlinear optimization progress evaluation is used to deeply analyze the optimization progress; the sub-target convergence speed is used to evaluate the optimization efficiency of each sub-target, and the strategy stability index measures the robustness of the current strategy under different states, dynamically identifies the bottleneck in the optimization process, and adjusts the exploration strategy; finally, based on the uncertainty score and the optimization progress evaluation results, the exploration rate is adjusted according to the target priority weight and the environmental complexity factor. Step S30231 of the above scheme evaluates the current state through multi-dimensional uncertainty quantification, the target conflict index quantifies the mutual influence between each optimization target, and identifies the competition or synergy relationship between the targets; the historical data deviation evaluates the degree of deviation between the current observation value and the historical trend, reflecting the influence of environmental dynamics or data noise; the dynamic uncertainty score combines the above indicators to generate a quantitative value reflecting the uncertainty of the current state, providing a basis for adjusting the exploration rate. Significance achieved: Accurate state perception can fully perceive the complexity and dynamics of the current state through multi-dimensional uncertainty quantification, avoiding misjudgment caused by local information bias; dynamic adaptability provides a scientific basis for the adjustment of the exploration rate, ensuring that the system can flexibly respond to environmental changes and improve the robustness of optimization. Step S30232 uses nonlinear optimization progress evaluation to deeply analyze the optimization progress, and the sub-goal convergence speed evaluates the optimization efficiency of each sub-goal to identify bottlenecks or inefficient links in the optimization process; the strategy stability index measures the robustness of the current strategy under different states to ensure the reliability of the strategy in a dynamic environment; optimization bottleneck identification dynamically identifies bottlenecks in the optimization process and provides direction for the adjustment of the exploration strategy. Significance achieved: Optimization efficiency is improved by identifying the differences in sub-goal convergence speeds, and inefficient links can be optimized in a targeted manner to accelerate the overall optimization process; strategy robustness is enhanced by evaluating strategy stability, and the system can avoid strategy failure caused by environmental changes and ensure the continuity of the optimization process; dynamic strategy adjustment provides a scientific basis for updating the exploration strategy to ensure that the strategy can adapt to the ever-changing optimization needs. Step S30233 adjusts the exploration rate based on the uncertainty score and optimization progress evaluation results. The target priority weight dynamically adjusts the exploration rate according to the importance of each target, giving priority to high-priority targets. The environmental complexity factor flexibly adjusts the exploration rate according to the dynamics and complexity of the environment to avoid ineffective exploration or excessive exploration. The exploration rate adjustment combines the above factors to dynamically balance exploration and utilization to ensure the efficiency and effectiveness of the optimization process.Significance achieved: Resource optimization allocation can concentrate limited resources in the direction that needs exploration most by dynamically adjusting the exploration rate, thereby improving optimization efficiency; the global optimal solution approximation can avoid falling into the local optimum and gradually approach the global optimal solution by balancing exploration and utilization; environmental adaptability is enhanced by adjusting the exploration rate according to the environmental complexity factor, and can maintain efficient operation in complex and changing environments.

[0158] In summary, the precise perception state of this embodiment quantifies multi-dimensional uncertainty, comprehensively evaluates the current state, and provides a scientific basis for decision-making; the efficient optimization progress identifies bottlenecks and improves optimization efficiency through nonlinear optimization progress evaluation to ensure the robustness of the strategy; the dynamic balance exploration and utilization optimizes resource allocation through adaptive exploration rate adjustment, and gradually approaches the global optimal solution. It not only improves the efficiency and effect of multi-objective optimization, but also demonstrates its powerful application value in complex environments through dynamic adaptability and robustness.

[0159] Example 8: Figure 8 As shown, based on Example 4, the process of re-evaluating the definition of the state space and the action space provided in this embodiment of the present invention includes the following steps:

[0160] S3031: Based on the current quality deviation and optimization goal, re-evaluate the contribution of each parameter in the multi-dimensional state space and identify the key parameters that have a significant impact on quality; assign new dynamic weights to key parameters based on real-time sensor data and historical performance trends; incorporate external environmental variables and equipment status into the state space;

[0161] S3032: Redefine the amplitude and priority of macro adjustment actions according to the severity of the current quality deviation; control actions for local parameter local temperature fine-tuning and pressure gradient adjustment;

[0162] S3033: Re-adjust the weights of key parameters and standard range deviations based on the distribution characteristics of current quality deviations; introduce a dynamic threshold mechanism to dynamically adjust the standard range based on historical performance trends; and provide additional incentives for long-term quality improvement through historical performance trend analysis.

[0163] The working principle and beneficial effects of the above technical solution are as follows: This embodiment first re-evaluates the contribution of each parameter in the multidimensional state space based on the current quality deviation and optimization target, and identifies the key parameters that have a significant impact on quality; assigns new dynamic weights to key parameters based on real-time sensor data and historical performance trends; incorporates external environmental variables and equipment status into the state space; secondly, based on the severity of the current quality deviation, redefines the amplitude and priority of macro-adjustment actions; performs control actions on local temperature fine-tuning and pressure gradient adjustment of local parameters; finally, based on the distribution characteristics of the current quality deviation, readjusts the weights of key parameters and standard range deviations; introduces a dynamic threshold mechanism to dynamically adjust the standard range based on historical performance trends; and provides additional incentives for long-term quality improvement through historical performance trend analysis. In step S3031 of the above scheme, the state space is re-evaluated. The parameter contribution analysis identifies the key parameters that have a significant impact on quality, accurately focuses on the factors that contribute the most to quality deviation, and improves the pertinence of the state space; the dynamic weight adjustment assigns new dynamic weights to key parameters based on real-time sensor data and historical performance trends to ensure that the state space can more accurately reflect the current production status; the inclusion of environment and equipment status includes external environmental variables (such as temperature and humidity) and equipment status (such as aging degree) into the state space to enhance the comprehensiveness and adaptability of the model. Significance: Accurate perception By optimizing the state space, the model can more accurately perceive the current production status and provide a more reliable basis for action decisions; the adaptability improvement incorporates the environment and equipment status, so that the model can better cope with complex and changing production environments and improve its robustness; efficiency optimization focuses on key parameters, reduces the interference of redundant information, and improves the operating efficiency of the model. Step S3032 Redefine the action space, optimize the macro-adjustment action. Redefine the amplitude and priority of the macro-adjustment action according to the severity of the current quality deviation to ensure that it can quickly respond to significant deviations; expand the micro-optimization action to add control actions for local parameters (such as local temperature fine-tuning, pressure gradient adjustment) to improve the refined control ability of the action space. Significance: Rapid response. By optimizing the macro-adjustment action, the model can quickly correct significant quality deviations and reduce production losses; refined control extends the micro-optimization action, so that the model can fine-tune local parameters to improve the stability and consistency of the overall quality; multi-level control combines macro and micro actions to achieve multi-level and multi-dimensional control strategies and improve the comprehensive performance of the model. Step S3033 Optimize the reward function, adjust the deviation weight. According to the distribution characteristics of the current quality deviation, readjust the weight of the deviation between the key parameters and the standard range to ensure that the reward function can more accurately reflect the severity of the quality deviation; introduce a dynamic threshold mechanism to dynamically adjust the standard range according to the historical performance trend to avoid reward deviations caused by fixed thresholds.Long-term performance incentives provide additional incentives for long-term quality improvement (such as stability and consistency) through historical performance trend analysis, guiding the model to focus on long-term optimization goals. Significance: Accurate evaluation optimizes the deviation weights and introduces dynamic thresholds, so that the reward function can more accurately evaluate the current quality status and provide more reliable feedback for the model; long-term optimization provides additional incentives for long-term performance, guiding the model to focus on long-term quality improvement while quickly responding to short-term deviations; multi-objective balance dynamically adjusts the reward function to ensure that the model can balance multiple goals such as quality, energy consumption and efficiency, thereby improving its overall performance.

[0164] In summary, this embodiment significantly improves the overall performance of the model through precise perception, rapid response and refined control; enhances the adaptability of the model in complex and changing environments by incorporating environmental and equipment status and dynamically adjusting weights and thresholds; guides the model to focus on long-term quality improvement while responding quickly to short-term deviations through long-term performance incentives and multi-objective balance; reduces production losses and improves production efficiency and product quality by optimizing control strategies and reward functions. It not only demonstrates the creative value of technological innovation, but also provides a more efficient and reliable intelligent solution for sintering production.

[0165] Example 9: Fig. 9 As shown, based on Example 8, the process of dynamically adjusting the standard range according to the historical performance trend provided by the embodiment of the present invention includes the following steps:

[0166] S30331: Based on historical performance data, identify the regular characteristics of its changes over time and extract key features in historical performance trends; use trend models to predict the possible range of changes in key parameters in the future;

[0167] S30332: Set the initial threshold range based on the statistical characteristics of historical performance data as a benchmark for dynamic adjustment; compare the actual value of the current key parameter with the historical performance trend to evaluate the degree of deviation;

[0168] S30333: Based on the real-time performance evaluation results and combined with trend prediction information, the threshold range is dynamically adjusted; if the current parameter value is higher than the historical trend prediction value, the upper threshold limit is increased; if it is lower than the prediction value, the lower threshold limit is lowered.

[0169] The working principle and beneficial effects of the above technical solution are as follows: This embodiment first identifies the regular characteristics of the historical performance data that change over time and extracts the key characteristics in the historical performance trend; uses the trend model to predict the possible range of changes of key parameters in the future; secondly, sets the initial threshold range according to the statistical characteristics of the historical performance data as a benchmark for dynamic adjustment; compares the actual value of the current key parameter with the historical performance trend to evaluate the degree of deviation; finally, based on the real-time performance evaluation results and combined with the trend prediction information, dynamically adjusts the threshold range; if the current parameter value is higher than the historical trend prediction value, the upper threshold limit is increased; if it is lower than the prediction value, the lower threshold limit is lowered. Step S30331 of the above solution, regularity feature recognition, identifies the regular characteristics of the key parameters that change over time (such as periodicity, trend, volatility, etc.) by analyzing the historical performance data, and provides a theoretical basis for dynamic adjustment; trend prediction uses the trend model to predict the possible range of changes of key parameters in the future, and provides a forward-looking basis for the adjustment of dynamic thresholds. Significance achieved: Improve the prediction ability of the model Through trend prediction, the model can predict the change trend of key parameters in advance and reduce misjudgment caused by sudden changes; Enhance adaptability Based on the regular characteristics of historical data, the model can better adapt to the complex and changeable production environment and improve robustness. Step S30332 Set the initial threshold range according to the statistical characteristics of historical performance data as the benchmark for dynamic adjustment; Compare the actual value of the current key parameter with the historical performance trend to evaluate its degree of deviation; Initial threshold setting is based on the statistical characteristics of historical performance data (such as mean, standard deviation, etc.), set the initial threshold range to provide a benchmark for dynamic adjustment; Real-time performance evaluation compares the actual value of the current key parameter with the historical performance trend to evaluate its degree of deviation and provide real-time basis for dynamic adjustment. Significance achieved: Ensure the rationality of the threshold The initial threshold range is set based on the statistical characteristics of historical data, which can reflect the normal range of changes in key parameters and avoid setting the threshold too broad or narrow; Real-time monitoring and feedback Through real-time performance evaluation, the model can promptly detect abnormal changes in key parameters and provide timely feedback for dynamic adjustment. Step S30333 Dynamic threshold adjustment dynamically adjusts the threshold range according to the real-time performance evaluation results and trend prediction information, so that it can reflect the changes in the current production status; threshold smoothing avoids sudden changes in the threshold range through dynamic adjustment, ensuring its stability and continuity. Significance achieved: The dynamic threshold can more accurately reflect the actual range of changes in key parameters, reducing the possibility of misjudgment and missed judgment; through dynamic adjustment, the model can adapt to changes in the production environment, improve robustness and practicality; the dynamic threshold mechanism can guide the model to focus on long-term performance improvement and achieve continuous improvement in production quality.

[0170] In summary, through historical performance trend analysis and dynamic threshold adjustment, the model of this embodiment can more accurately reflect the changing rules of key parameters and adapt to complex and changeable production environments; real-time performance evaluation and dynamic adjustment mechanisms can timely discover and correct abnormal changes in key parameters to ensure the stability of the production process; through trend prediction and dynamic threshold adjustment, the model can focus on long-term performance improvement and provide technical support for continuous improvement of production quality; combining historical performance trends with dynamic threshold adjustment breaks through the limitations of traditional fixed thresholds and reflects the creativity and practicality of the technical solution. It not only improves the performance of the model, but also provides new ideas and methods for the intelligent optimization of production systems, which has important technical value and practical significance.

[0171] Example 10: Fig.10 As shown, based on Example 9, the process of comparing the actual value of the current key parameter with the historical performance trend provided by the embodiment of the present invention includes the following steps:

[0172] S303321: Time-align the actual values ​​of current key parameters with historical performance trends; standardize the actual values ​​of current key parameters and historical performance trends to eliminate dimension and unit differences; extract key features from historical performance trends as a benchmark for comparison;

[0173] S303322: Match the actual value of the current key parameter with the key features of the historical performance trend to identify their performance differences in different dimensions; calculate the absolute difference between the actual value of the current key parameter and the predicted value of the historical performance trend to reflect the magnitude of the deviation; compare the absolute deviation with the fluctuation range of the historical performance trend to calculate the relative deviation rate to reflect the severity of the deviation;

[0174] S303323: Analyze the consistency between the actual value change trend of the current key parameters and the historical performance trend to determine whether they deviate from the expected direction; comprehensively score the degree of deviation of the actual value of the current key parameters based on the multi-dimensional indicators of absolute deviation and relative deviation rate; and divide the degree of deviation into different levels according to the comprehensive scoring results.

[0175] The working principle and beneficial effects of the above technical solution are as follows: the present embodiment firstly time-aligns the actual value of the current key parameter with the historical performance trend; standardizes the actual value of the current key parameter and the historical performance trend to eliminate the dimension and unit differences; extracts key features from the historical performance trend as a comparison benchmark; secondly, matches the actual value of the current key parameter with the key features of the historical performance trend to identify the performance differences in different dimensions; calculates the absolute difference between the actual value of the current key parameter and the predicted value of the historical performance trend to reflect the magnitude of the deviation; compares the absolute deviation with the fluctuation range of the historical performance trend, calculates the relative deviation rate to reflect the severity of the deviation; finally, analyzes the consistency between the actual value change trend of the current key parameter and the historical performance trend to determine whether it deviates from the expected direction; based on the absolute deviation and relative deviation rate multi-dimensional indicators, comprehensively scores the deviation degree of the actual value of the current key parameter; and divides the deviation degree into different levels according to the comprehensive scoring result. Step S303321 of the above solution aligns and preprocesses data to ensure that the actual value of the current key parameter and the historical performance trend are on the same time scale to avoid comparison errors caused by time deviation. Significance: Provide an accurate time benchmark for comparative analysis and enhance the reliability of evaluation results. Standardization eliminates the dimension and unit differences between the actual value of the current key parameter and the historical performance trend, making the two comparable. Significance: Improve the fairness and consistency of the comparison and avoid misjudgment caused by different data units. Key feature extraction provides a clear reference basis for matching and deviation evaluation and enhances the pertinence of the evaluation. Step S303322 Feature matching and deviation quantification comprehensively reveal the difference between the actual value of the current key parameter and the historical trend, providing multi-dimensional information for deviation evaluation; intuitively quantify the gap between the actual value of the current key parameter and the historical trend, providing basic data for evaluation; on the basis of considering the historical fluctuation range, further evaluate the severity of the deviation to avoid underestimating the degree of deviation due to large historical fluctuations. Step S303323 Trend consistency analysis and comprehensive evaluation evaluate whether the actual value of the current key parameter is consistent with the historical trend from the perspective of trend, providing a basis for the analysis of the cause of deviation; through the comprehensive evaluation of multi-dimensional indicators, avoid the limitations of a single indicator and improve the comprehensiveness and accuracy of the evaluation results; provide a clear level basis for dynamic adjustment, so as to facilitate the adoption of corresponding adjustment strategies for different degrees of deviation.

[0176] In summary, this embodiment ensures the accuracy of comparative analysis and avoids misjudgment due to data deviation through time alignment, standardization processing and key feature extraction; covers all aspects of deviation degree assessment from feature matching to deviation quantification, to trend consistency analysis and comprehensive evaluation, ensuring the comprehensiveness of the evaluation results; based on real-time performance data and historical trend information, it can dynamically evaluate the degree of deviation and provide real-time basis for dynamic adjustment; through deviation level classification, it provides clear guidance for practical applications and facilitates taking corresponding adjustment measures for different degrees of deviation. It not only improves the accuracy and comprehensiveness of deviation degree assessment, but also provides a scientific basis for dynamically adjusting the standard range, which has important technical value and practical significance.

[0177] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention belong to the scope of equivalent technologies of the present invention, the present invention is also intended to include these modifications and variations.

Claims

1. A detection method for evaluating the sintering quality of an artificial thermal conductive film, characterized in that: The following steps are involved: Build a personalized database containing the corresponding conductivity performance data of artificial thermal conductive films under different materials and different process conditions; through the analysis of conductivity performance data, identify the performance fluctuation parameters of different sintering production equipment and the same sintering production equipment in different production periods; Multiple types of sensors are installed on the artificial thermal conductive film sintering production equipment to collect key parameters such as temperature, pressure and time in the sintering production equipment; The data fusion algorithm is used to fuse multiple types of key parameters to obtain a fused data set; the sintering production equipment is adjusted according to the performance fluctuation parameters; The key parameters in the fused data set are input into the sintering quality assessment model and compared with the standard parameters in the personalized database to obtain the comparison results. Among them, the comparison results include which key parameters deviate from the standard range, and the comparison results are automatically fed back to the sintering production equipment for corresponding adjustments.

2. The method for evaluating the sintering quality of an artificial thermally conductive film according to claim 1, characterized in that: The process of analyzing the conductivity data includes the following steps: Build a hybrid storage architecture for personalized databases, where the basic data layer stores static data on material properties and sintering production equipment parameters; the process data layer stores dynamic data during the sintering process; The performance data layer stores the test results of the conductive performance and its related indicators; Clean the material data, process data, equipment data and conductive performance data, and establish corresponding classification index, attribute index, numerical index and time index according to key fields; According to the index, key patterns related to conductive properties are extracted from the personalized database, and data points that deviate from normal patterns are identified; the performance fluctuation parameters corresponding to the data points are decomposed into influencing factors at different levels, and the changes in conductive properties are decomposed into microstructural changes at the material level.

3. The method for evaluating the sintering quality of an artificial thermally conductive film according to claim 1, characterized in that: The process of obtaining the fused data set includes the following steps: The key parameters of temperature, pressure and time collected by different sensors are normalized, and the influence of each key parameter on the sintering quality is evaluated by calculating the covariance matrix; According to the real-time status of the sintering production equipment, the weight distribution is dynamically adjusted using an adaptive weight adjustment factor; The normalized key parameters are weighted fused according to dynamic weights to generate a preliminary fused data set, outliers in the fused data set are removed, and the fused data set without outliers is smoothed; The optimized fused data set is standardized to meet the input requirements of the sintering quality assessment model, and the standardized fused data set is output to the sintering quality assessment model for comparison with the standard parameters of the personalized database.

4. The method for evaluating the sintering quality of an artificial thermally conductive film according to claim 1, characterized in that: The process of building a sintering quality assessment model includes the following steps: Mapping multi-type sensor data of sintering production equipment into a multi-dimensional state space, assigning dynamic weights to each key parameter based on historical data of different materials and process conditions in a personalized database; The action space adopts a composite structure, including macro adjustment and micro optimization, and the reward function adopts a hierarchical design; The adaptive exploration strategy dynamically adjusts the exploration rate according to the uncertainty of the current state; at the same time, the strategy gradient optimizes multiple goals such as sintering quality, equipment energy consumption and production efficiency, realizing the combination of action value evaluation and multi-objective optimization of strategy gradient; receives sensor data in real time and updates the adaptive exploration strategy; If the adjusted sintering quality still does not meet expectations, the sintering quality evaluation model automatically triggers the self-correction mechanism, re-evaluates the definition of the state space and action space, and optimizes the reward function.

5. The method for evaluating the sintering quality of an artificial thermally conductive film according to claim 4, characterized in that: The reward function adopts a hierarchical design process, which includes the following steps: Design the base layer reward, select the key parameters for sintering quality, and calculate the deviation between the actual value and the standard value of each key parameter; Through historical performance trend analysis, additional incentives are given to long-term quality improvement through design optimization layer rewards; The base layer reward and the optimization layer reward are combined to obtain a comprehensive reward function. To adapt to different materials and process conditions, the dynamic weight is adjusted according to historical data and current status. When the sintering quality assessment model triggers the self-correction mechanism, the reward function is re-evaluated and optimized, and the definition of key parameters and action space is adjusted according to the changes in the current production environment; the dynamic weights are recalculated based on the latest data; According to the optimization results, adjust the weight coefficients of the base layer rewards and optimization layer rewards.

6. The method for evaluating the sintering quality of an artificial thermal conductive film according to claim 4, characterized in that: The process of combining action value evaluation with multi-objective optimization of policy gradients includes the following steps: Determine sintering quality, equipment energy consumption and production efficiency as the core optimization objectives, transform the multi-objective optimization problem into a unified evaluation system, and integrate different objectives into a comprehensive optimization function through weighting; Action value evaluation is used to quantify the contribution of each action to multi-objective optimization, combining macro- and micro-level action value evaluations to form a composite evaluation system; Decompose the multi-objective optimization problem into multiple sub-objectives, each of which corresponds to an independent policy gradient optimization path; by weighting, merge the policy gradients of each sub-objective into a comprehensive gradient to guide the update direction of the strategy; dynamically adjust the weight of each sub-objective according to the current state and goal priority; Dynamically adjust the exploration rate based on the uncertainty of the current state and the optimization progress; receive sensor data in real time and update the exploration strategy.

7. The method for evaluating the sintering quality of an artificial thermally conductive film according to claim 6, characterized in that: The process of dynamically adjusting the exploration rate and updating the exploration strategy includes the following steps: The current state is evaluated through multi-dimensional uncertainty quantification; the target conflict index is used to measure the mutual influence between the optimization targets, and the historical data deviation reflects the degree of deviation between the current observation value and the historical trend; through comprehensive calculation, a dynamic uncertainty score is generated; Nonlinear optimization progress evaluation is used to deeply analyze the optimization progress; sub-goal convergence speed is used to evaluate the optimization efficiency of each sub-goal, while the strategy stability index measures the robustness of the current strategy under different states, dynamically identifies bottlenecks in the optimization process, and adjusts the exploration strategy; Based on the uncertainty score and optimization progress evaluation results, the exploration rate is adjusted according to the target priority weight and the environment complexity factor.

8. The method for evaluating the sintering quality of an artificial thermally conductive film according to claim 4, characterized in that: The process of re-evaluating the definitions of the state space and action space consists of the following steps: Based on the current quality deviation and optimization goals, re-evaluate the contribution of each parameter in the multi-dimensional state space and identify the key parameters that have a significant impact on quality; assign new dynamic weights to key parameters based on real-time sensor data and historical performance trends; incorporate external environmental variables and equipment status into the state space; Redefine the magnitude and priority of macro-adjustment actions according to the severity of the current quality deviation; fine-tune local parameters, local temperature, and control actions for pressure gradient adjustment; According to the distribution characteristics of the current quality deviation, the weights of the deviations between key parameters and the standard range are readjusted; a dynamic threshold mechanism is introduced to dynamically adjust the standard range according to historical performance trends; and additional incentives are given for long-term quality improvement through historical performance trend analysis.

9. The method for evaluating the sintering quality of an artificial thermally conductive film according to claim 8, characterized in that: The process of dynamically adjusting the standard range based on historical performance trends includes the following steps: Based on historical performance data, identify the regularity of its changes over time and extract key features in historical performance trends; Use trend models to predict the possible range of changes in key parameters in the future; Compare the actual values ​​of current key parameters with historical performance trends to assess their degree of deviation; Based on the real-time performance evaluation results and combined with trend forecast information, the threshold range is dynamically adjusted; if the current parameter value is higher than the historical trend forecast value, the upper threshold limit is increased; If it is lower than the predicted value, lower the lower threshold.

10. The method for evaluating the sintering quality of an artificial thermally conductive film according to claim 9, characterized in that: The process of comparing the current actual values ​​of key parameters with historical performance trends includes the following steps: Time-align the actual values ​​of current key parameters with historical performance trends; standardize the actual values ​​of current key parameters and historical performance trends to eliminate dimension and unit differences; Extract key features from historical performance trends as a baseline for comparison; Match the actual value of the current key parameter with the key features of the historical performance trend to identify their performance differences in different dimensions; calculate the absolute difference between the actual value of the current key parameter and the predicted value of the historical performance trend to reflect the magnitude of the deviation; compare the absolute deviation with the fluctuation range of the historical performance trend to calculate the relative deviation rate to reflect the severity of the deviation; Analyze the consistency between the actual value change trend of the current key parameters and the historical performance trend to determine whether they deviate from the expected direction; comprehensively score the degree of deviation of the actual value of the current key parameters based on the multi-dimensional indicators of absolute deviation and relative deviation rate; According to the comprehensive scoring results, the degree of deviation is divided into different levels.

Images