Large model-driven data governance decision support system

Abstract

This invention relates to a large-model-driven data governance decision support system, belonging to the fields of intelligent manufacturing and industrial automation control technology. It includes: a physical field evolution modeling module, which establishes a real-time global stress field based on collected multimodal state datasets and outputs a time-series stress field dataset; a dynamic drift assessment module, which receives the time-series stress field dataset, compares it with a preset reference stress field dataset, and calculates the drift risk level; a defect evolution prediction module, which inputs the time-series stress field dataset and collected defect-related feature datasets into a fusion prediction model to generate a latent defect risk index; and a closed-loop adaptive control module, which integrates the drift risk level and the latent defect risk index to generate an adaptive control gain, calculates the deviation between the time-series and reference stress field datasets, and generates control commands to adjust production parameters based on the gain and deviation. This invention achieves a leap from discrete sensor data to continuous dynamic physical field models, providing a high-fidelity data foundation for decision-making.

Description

Technical Field

[0001] This invention relates to the fields of intelligent manufacturing and industrial automation control, specifically a large-model-driven data governance decision support system. Background Technology

[0002] In the production process of high-end manufacturing, traditional quality control and process monitoring methods mainly rely on discrete, single-point sensor data and offline material sampling inspection. These methods have problems such as insufficient coverage and poor real-time performance in reflecting the overall production process. They cannot construct a panoramic view of the internal stress state of materials during the production process, nor can they capture the dynamic process of its changes over time and space.

[0003] This situation results in quality control measures being inherently lagging and reactive. Managers are unable to effectively distinguish between temporary fluctuations in the production process and systemic drifts that may lead to product failure, and they also lack the ability to predict future defects. This makes production adjustments often only carried out after problems occur, making it difficult to provide scientific preventive control strategies.

[0004] The aforementioned situation and shortcomings are due to the limitations of data acquisition methods and analysis models. The limited number of sensor points on the production line leads to insufficient spatial resolution of the data, while offline and lagging detection methods cannot meet the needs of real-time monitoring. At the same time, the system lacks a comprehensive understanding and modeling of the coupling effect of multiple physics fields, and cannot integrate scattered measurement data into a continuous and dynamic physical field.

[0005] As a result, when potential quality risks arise in the production process, the system cannot provide early warnings or accurately locate the root cause of the problem, leading to unnecessary waste and failing to achieve forward-looking, refined closed-loop control of product quality and process stability.

[0006] The information disclosed in the background section above is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0007] The purpose of this invention is to provide a large model-driven data governance decision support system to solve the problems mentioned in the background art.

[0008] The technical solution of the present invention includes: a physical field evolution modeling module, which collects a multimodal state dataset, establishes a real-time global stress field based on the multimodal state dataset, and outputs a time-series stress field dataset;

[0009] The dynamic drift assessment module receives the time-series stress field dataset, compares it with the preset reference stress field dataset, and calculates the drift risk level.

[0010] The defect evolution prediction module collects a defect-related feature dataset, and uses the time-series stress field dataset and the defect-related feature dataset as inputs to the fusion prediction model to generate a latent defect risk index.

[0011] The closed-loop adaptive control module integrates the drift risk level and the latent defect risk index to generate an adaptive control gain, calculates the deviation between the time-series stress field dataset and the reference stress field dataset, and generates control commands to adjust production parameters based on the adaptive control gain and the deviation.

[0012] Preferably, the multimodal state dataset includes: polarization delay data, material thickness data, surface temperature data, and material transport velocity data;

[0013] The physical field evolution modeling module reconstructs the mechanical stress field based on the polarization delay data, the material thickness data, and the surface temperature data.

[0014] The physical field evolution modeling module calculates the thermal stress field based on the surface temperature data;

[0015] The real-time global stress field is formed by the superposition of the mechanical stress field and the thermal stress field.

[0016] Preferably, the physical field evolution modeling module further includes:

[0017] Based on the material transport velocity data, calculate the advection term describing the stress field transport effect with material motion;

[0018] By simulating the stress relaxation within the material, the diffusion term is obtained;

[0019] Based on the difference between the newly calculated stress and the stress field at the previous moment, a source term is set to drive the model update.

[0020] The physical field evolution modeling module establishes the dynamic evolution equation of the stress field based on the advection term, the diffusion term, and the source term;

[0021] The physical field evolution modeling module updates the real-time global stress field based on the stress field dynamic evolution equation.

[0022] Preferably, the dynamic drift evaluation module includes:

[0023] Calculate the difference between the time-series stress field dataset and the reference stress field dataset to generate a real-time deviation;

[0024] The relative drift index is calculated based on the real-time deviation and the reference stress field dataset.

[0025] The absolute over-limit index is calculated based on the real-time deviation and the preset maximum allowable stress deviation.

[0026] The determination of the drift risk level depends on the relative drift index and the absolute over-limit index.

[0027] Preferably, the dynamic drift evaluation module further includes:

[0028] Compare the absolute exceedance index with the quality compliance threshold;

[0029] Compare the relative drift index with the process stability warning threshold;

[0030] When the absolute exceedance index is greater than the quality compliance threshold, the drift risk level is determined as a quality exceedance warning.

[0031] When the absolute exceedance index is not greater than the quality compliance threshold and the relative drift index is greater than the process stability warning threshold, the drift risk level is determined as a process instability warning.

[0032] When the absolute exceedance index is not greater than the quality compliance threshold and the relative drift index is not greater than the process stability warning threshold, the drift risk level is determined to be qualified.

[0033] Preferably, the defect-associated feature dataset includes: acoustic emission signal feature data acquired by an acoustic sensor array, and surface visual feature data acquired by a line scan camera;

[0034] The latent defect risk indicators include: stress gradient risk factor and latent defect index.

[0035] Preferably, the defect evolution prediction module further includes:

[0036] The stress gradient risk factor is calculated based on the spatial gradient and temporal rate of change of the time-series stress field dataset.

[0037] The defect evolution prediction module takes the time-series stress field dataset, the acoustic emission signal feature data, the surface visual feature data, and the stress gradient risk factor as inputs to the fusion prediction model to output the latent defect index.

[0038] Preferably, the closed-loop adaptive control module includes:

[0039] The deviation between the time-series stress field dataset and the reference stress field dataset is input into the inverse model to generate the control target error vector.

[0040] Establish a decision logic that maps the drift risk level and the latent defect risk index to the adaptive control gain;

[0041] The closed-loop adaptive control module generates the adaptive control gain based on the decision logic.

[0042] Preferably, the closed-loop adaptive control module further includes:

[0043] The control target error vector is multiplied by the adaptive control gain to generate the control command;

[0044] The control command is the final control adjustment vector.

[0045] Preferably, the closed-loop adaptive control module further includes:

[0046] The final control adjustment vector is added to the current production parameter setpoint vector to generate an updated setpoint vector;

[0047] The closed-loop adaptive control module sends the updated setpoint vector to the production line execution unit to achieve closed-loop control.

[0048] This invention improves upon the existing technology by providing a large-model-driven data governance decision support system, and has the following improvements and advantages:

[0049] 1. It has achieved a leapfrog improvement from discrete sensing data to continuous dynamic physical field models, enabling managers to gain insight into the internal state of materials in the production process from a global and dynamic perspective for the first time, providing an unprecedented high-dimensional and high-fidelity data foundation for all subsequent decisions;

[0050] 2. This invention can output a clear drift risk level, such as a process instability warning or a quality over-limit warning. This distinction enables the system to take corresponding measures for different risks: mild preventive fine-tuning for the former and decisive corrective intervention for the latter, thereby greatly improving the accuracy and efficiency of control.

[0051] 3. It can issue early warnings at the nascent stage of defects or even before, achieving true predictive quality assurance;

[0052] 4. The entire production system is intelligent and adaptive, capable of responding to various production disturbances in the most efficient and robust manner, thereby improving the stability of product quality. Attached Figure Description

[0053] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0054] Figure 1This is a flowchart of the system of the present invention. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0056] Example 1:

[0057] Please see Figure 1 This invention provides a large-model-driven data governance decision support system technical solution, comprising: a physical field evolution modeling module, which collects multimodal state datasets, establishes a real-time global stress field based on the multimodal state datasets, and outputs a time-series stress field dataset; a dynamic drift assessment module, which receives the time-series stress field dataset, compares it with a preset reference stress field dataset, and calculates the drift risk level; a defect evolution prediction module, which collects a defect correlation feature dataset, uses the time-series stress field dataset and the defect correlation feature dataset as inputs to a fusion prediction model, and generates a latent defect risk index; and a closed-loop adaptive control module, which integrates the drift risk level and the latent defect risk index to generate an adaptive control gain, calculates the deviation between the time-series stress field dataset and the reference stress field dataset, and generates control commands to adjust production parameters based on the adaptive control gain and the deviation.

[0058] In this embodiment, a large model-driven data governance decision support system constructs a complete closed loop from data perception to decision execution through the sequential collaborative operation of its internal physical field evolution modeling module, dynamic drift evaluation module, defect evolution prediction module, and closed-loop adaptive control module. This system transforms discrete multimodal sensing data on the production line into a continuous, dynamic, and visualized global stress field time-series data stream, which not only accurately reflects the physical state of the material at any given moment but also profoundly reveals the evolution trend of stress as it flows along the production line.

[0059] The system achieves a two-dimensional dynamic assessment of production process stability and product quality compliance by quantitatively comparing the real-time stress field with the ideal state of the gold batch. The system integrates multi-source information such as internal stress, acoustic emission, and surface vision, realizing a technological leap from passive detection to proactive prediction. It can predict potential hidden defects that may form at a future point in time. All assessment and prediction analysis results are transformed into precise, quantitative control commands that are perfectly matched with the risk level, automatically adjusting production parameters, thereby achieving forward-looking, refined, and automated closed-loop control of product quality, significantly improving the yield rate and process stability in the high-end manufacturing field.

[0060] Example 2

[0061] The multimodal state dataset includes: polarization delay data, material thickness data, surface temperature data, and material transport velocity data; the physics field evolution modeling module reconstructs the mechanical stress field based on the polarization delay data, material thickness data, and surface temperature data; the physics field evolution modeling module calculates the thermal stress field based on the surface temperature data; the real-time global stress field is formed by superimposing the mechanical stress field and the thermal stress field.

[0062] The physics field evolution modeling module also includes: calculating the advection term to describe the stress field transport effect with material motion based on material transport velocity data; simulating stress relaxation inside the material to obtain the diffusion term; setting the source term to drive model updates based on the difference between the currently calculated new stress and the stress field at the previous moment; establishing the dynamic evolution equation of the stress field based on the advection term, diffusion term, and source term; and updating the real-time global stress field based on the dynamic evolution equation of the stress field.

[0063] In a specific implementation, the physical field evolution modeling module serves as the core of the system's perception and modeling. It constructs a dynamic stress field evolution model coupled with thermodynamic effects, decomposes the total stress field into the sum of mechanical stress and thermal stress, and uses a comprehensive evolution equation to accurately describe the dynamic changes of the stress field with material movement.

[0064] The calculation of the thermal stress field aims to quantify the internal stress caused by temperature changes;

[0065] ;

[0066] This formula originates from the theory of thermoelasticity. Its technical motivation lies in the fact that temperature changes in materials during the production process are the key physical factors that lead to the generation and uneven distribution of internal stress. In order to achieve accurate modeling of the total stress field, the passively measured temperature field data is transformed into stress field components with clear physical meaning, thereby incorporating the thermal effect as a calculable variable into the overall model.

[0067] in, Indicates the spatial location at time t. The thermal stress tensor at that location; The Young's modulus, which represents the material, is a core inherent property that characterizes the material's ability to resist elastic deformation. Its value is determined through preliminary material mechanical property experiments such as tensile tests, or by directly consulting the material specifications provided by the supplier. The coefficient of thermal expansion is an inherent thermophysical property that measures the degree to which the dimensions of a material change with temperature. Its value is determined by experimental methods such as thermomechanical analysis (TMA) or obtained by consulting authoritative material property handbooks. It is the stress-free reference temperature, which usually refers to the curing temperature during material molding or a set zero-stress reference process temperature. This is a key process setting parameter. It is the spatial position at time t that is collected in real time by a non-contact infrared sensor array. The surface temperature of the material;

[0068] This formula is applied in the physical field evolution modeling module to collect real-time surface temperature data. Combined with known material constants, the thermal stress components can be directly calculated. The technical effect is that it realizes the accurate quantification of thermal stress, which enables the model to clearly distinguish and evaluate the stress problems caused by uneven cooling or fluctuations in ambient temperature, providing key physical basis for subsequent stress analysis and defect tracing.

[0069] The core of tomographic reconstruction of mechanical stress field lies in using optical measurement data to invert the mechanical stress inside the material. This technical solution is based on the principle of photoelasticity and the idea of ​​computed tomography. Its technical motivation is that it is extremely difficult to directly measure the internal stress field, but the integral of the birefringence field, which is proportional to the stress field, along a specific path (i.e., the polarization retardation) can be measured non-contactly by optical means. Through tomographic reconstruction algorithm, the internal birefringence field distribution is inverted using the measured multi-angle polarization retardation data. According to the photoelasticity law, the reconstructed birefringence field is transformed into a mechanical stress field.

[0070] ;

[0071] ;

[0072] in, It is the birefringence at time t and spatial location (x,y); Representative operators for inverse tomographic reconstruction operations such as filtering and back projection; At time t, along the detection angle The original polarization delay data measured in the direction; the technical principle of this formula is that the collected multi-angle polarization delay data is regarded as the birefringence field inside the material. Along different angles The Raydon transform results are obtained. By applying a filtered back-projection algorithm to the projection data at each angle, the two-dimensional birefringent field can be reconstructed. After obtaining the birefringence field, the mechanical stress can be calculated according to the photoelastic law. .in, It is a temperature-dependent stress-optics coefficient, which characterizes a material's ability to convert mechanical stress into optical anisotropy, and its relationship with temperature. The dependency relationship was determined through prior calibration experiments;

[0073] To determine the stress-optical coefficient With temperature The dependency relationship can be calibrated using the following method: place the material sample in a precisely temperature-controlled heating chamber with a transparent window; apply a known, uniform uniaxial tensile stress to the sample using an external device. Using the same light source and polarization detector as this system, the optical path delay through the sample was measured. At multiple different temperature points Repeat this process, according to the fundamental formula of photoelasticity:

[0074] ; ;

[0075] The coefficient values ​​at different temperatures were calculated, and the function was finally obtained through curve fitting. ;

[0076] In the system, the formula fuses multimodal data from the optical sensing system—namely, polarization delay, thickness, and temperature—to reconstruct a high-resolution two-dimensional mechanical stress field. The significant technological advantage of this approach lies in its ability to achieve real-time, comprehensive, and visualized monitoring of internal mechanical stress on flexible material production lines. This overcomes the limitations of traditional single-point or offline measurements, providing an unprecedentedly powerful tool for understanding and controlling stresses introduced by processes such as mechanical tension and rolling.

[0077] The stress field dynamic evolution equation, which is the culmination of the module, describes the complete dynamics of the total stress field in time and space.

[0078] ;

[0079] This formula is a physics-based transport equation, whose structure draws on the convection-diffusion equations in fluid mechanics and heat transfer. The technical motivation is that static stress field snapshots are insufficient to describe the full picture of continuous production processes. As materials flow on the production line, the internal stress field is both transported and relaxed due to changes in the internal structure, while new stresses are constantly generated. This equation aims to establish a unified mathematical framework that can describe these complex dynamic processes, thereby achieving a leap from stress field snapshots to dynamic representations of stress fields.

[0080] in, Represents the total stress field (Right now The rate of change over time; This is the translational term, used to describe the stress field as it changes with the material at a velocity. The transport effect produced by motion, among which Monitored by a laser Doppler velocimeter, It is a gradient operator; This is a diffusion term used to simulate stress relaxation and diffusion phenomena caused by viscoelasticity within the material. It is the Laplace operator, and D is the stress diffusion coefficient, which is a physical quantity that characterizes the rate of stress relaxation of a material. Its unit is m² / s. It is calibrated through targeted material experiments such as stress relaxation experiments, or set empirically based on the viscoelastic model of the material.

[0081] Stress diffusion coefficient The calibration can be determined through the following experiment: Select a test area on the production line, and use external excitation to create a local, instantaneous stress concentration at the center of the area; then, use the physical field evolution modeling module of this invention to continuously monitor the stress peak value over time. and space The dissipation process will collect time-series stress field data. By fitting the analytical solution of the two-dimensional diffusion equation, the expression of the two-dimensional diffusion equation is:

[0082]

[0083] This represents the change of the stress field σ with time t, thus allowing us to solve for the optimal diffusion coefficient D.

[0084] For the source term, it is defined as:

[0085] ;

[0086] Since σ_total(t) = σ_mech(t) + σ_thermal(t), this formula is equivalent to:

[0087] ;

[0088] Represents the stress field newly calculated at the current moment. Stress field at the previous moment The difference, in a physical sense, is the effective new information that drives the model to continuously update towards the real state, ensuring that the model can keep up with changes in the production state in real time;

[0089] This equation is the computational core of the physical field evolution modeling module. Based on the advection, diffusion, and source terms, it is solved numerically, for example using the finite element method or finite difference method, to dynamically update the real-time global stress field of the entire production area. The core technical effect is that it truly transforms discrete multimodal sensing data into a continuous, dynamic, and visualized dynamic representation of the global stress field. This dynamic representation not only reflects the current state but also reflects the complete history and future trend of stress evolution along the production line, providing the highest-dimensional and most insightful data input for all subsequent evaluation and prediction modules.

[0090] Example 3

[0091] The dynamic drift assessment module includes: calculating the difference between the time-series stress field dataset and the reference stress field dataset to generate a real-time deviation; calculating the relative drift index based on the real-time deviation and the reference stress field dataset; calculating the absolute over-limit index based on the real-time deviation and the preset maximum allowable stress deviation; and determining the drift risk level depends on the relative drift index and the absolute over-limit index.

[0092] The dynamic drift assessment module also includes: comparing the absolute exceedance index with the quality compliance threshold; comparing the relative drift index with the process stability warning threshold; when the absolute exceedance index is greater than the quality compliance threshold, the drift risk level is determined as a quality exceedance warning; when the absolute exceedance index is not greater than the quality compliance threshold and the relative drift index is greater than the process stability warning threshold, the drift risk level is determined as a process instability warning; when the absolute exceedance index is not greater than the quality compliance threshold and the relative drift index is not greater than the process stability warning threshold, the drift risk level is determined as acceptable.

[0093] The dynamic drift evaluation module of the technical solution in this embodiment is designed to conduct a comprehensive real-time evaluation of the production status. It achieves dual monitoring of process stability and product quality compliance by calculating two core indicators: the relative drift index and the absolute over-limit index. In this embodiment, in order to conduct a refined evaluation of the material, the entire global stress field is discretized into multiple sub-regions, and the subscript i represents a specific sub-region.

[0094] Relative drift index The calculations are used to assess the relative stability of the production process;

[0095] ;

[0096] This formula originates from the concept of measuring process variability in statistical process control. The technical motivation is that it is not enough to know only the absolute value of the deviation; it is also necessary to understand the magnitude of this deviation relative to the normal state. A deviation that is small for a high-stress region may be a significant fluctuation for a low-stress region. A standardized, dimensionless index is needed to objectively evaluate the relative degree of process fluctuation.

[0097] in, Let be the relative drift index of region i; It is determined by the real-time total stress field Subtract the reference stress field The obtained real-time bias tensor; It is a preset reference stress field, which is set based on the production data of the gold batch or the ideal finite element simulation results; The Frobenius norm represents the matrix. It is calculated as the square root of the sum of the squares of all the elements of the matrix, and can comprehensively reflect the overall magnitude of the stress tensor.

[0098] This index is continuously calculated in the dynamic drift assessment module to quantify the fluctuation range of the current production state compared to the ideal state. The technical effect is to provide a sensitive quantitative indicator of process stability. In a stable process, this index should remain within a small range. The continuous increase or drastic fluctuation of the index, even if the quality red line has not been reached, indicates that the process is becoming unstable, providing a direct basis for subsequent first-level early warning - process instability, thereby achieving preventive early intervention.

[0099] Absolute Overlimit Index The calculation is used to assess the absolute compliance of product quality;

[0100] ;

[0101] The design concept of this formula comes directly from the concept of specification limits in quality management. The technical motivation is that any production process must meet the final product quality specifications. A direct and clear indicator is needed to measure whether the current stress deviation has approached or exceeded the maximum range allowed by the product specifications, so as to determine whether there is a quality risk in the product.

[0102] in, Let be the absolute overlimit index of region i; Same as the previous formula, it is the real-time bias tensor; It is the maximum allowable stress deviation tensor in region i, set according to the final product quality specification. It defines the permissible fluctuation boundary of real-time stress relative to the reference reference.

[0103] This index is a decisive criterion for quality compliance, when When this occurs, it clearly means that the stress deviation in the area has exceeded the product specifications, constituting a quality defect. Its technical effect is that it transforms the abstract quality specifications into a quantitative indicator that can be calculated and monitored in real time. It provides a decisive and unambiguous triggering condition for the secondary early warning - quality over-limit, ensuring that any situation that exceeds the quality bottom line can be immediately identified and corrective intervention can be initiated.

[0104] The key adjustable parameters in this module are: This refers to the relative drift warning threshold, whose setting logic is clear and implementable: Those skilled in the art analyze historical production data from known quality-compliant batches, particularly during periods of normal process fluctuations within these batches, and calculate the threshold values ​​for these periods. The distribution of the index will Setting the threshold to a high quantile of the distribution, such as the 95th percentile, ensures that it is sensitive enough to issue an early warning when abnormal fluctuations first appear in the process, but before they reach a level that affects product quality, reflecting the preventive principle of risk management.

[0105] Example 4

[0106] The defect-related feature dataset includes: acoustic emission signal feature data acquired by an acoustic sensor array, and surface visual feature data acquired by a line scan camera; latent defect risk indicators include: stress gradient risk factor, and latent defect index;

[0107] The defect evolution prediction module also includes: calculating the stress gradient risk factor based on the spatial gradient and time change rate of the time-series stress field dataset; the defect evolution prediction module takes the time-series stress field dataset, acoustic emission signal feature data, surface visual feature data and stress gradient risk factor as input to the fusion prediction model to output the latent defect index.

[0108] The core task of the defect evolution multimodal monitoring and prediction module in this embodiment is to upgrade from passive detection to active prediction. It uses deep learning prediction by fusing multi-source data to discover latent defects that will form at some point in the future.

[0109] Stress gradient risk factor The calculations aim to quantify the risks posed by stress concentration and its rate of deterioration;

[0110] ;

[0111] This formula is based on the fundamental principle of fracture mechanics in materials mechanics, namely that material failure often begins in areas of stress concentration, and the intensification of stress concentration is a precursor to crack propagation. The technical motivation is that the amplitude of the stress field alone is insufficient to fully assess the risk of defects. The spatial gradient of the stress field represents the degree of stress concentration, while the time rate of change of the stress field amplitude reflects the dynamic trend of state deterioration. Combining the two can provide a more comprehensive assessment of the risk.

[0112] in, Let be the stress gradient risk factor for region i at time t; in the first term... It is the norm of the stress field gradient, which directly quantifies the degree of stress concentration in that region; the second term... The time-varying rate of stress norm (state deterioration rate) is expressed as the material transport velocity. Normalization was performed to convert it into a stress change gradient along the material's motion direction, thus enabling it to be consistent with the spatial gradient of the first term in terms of physical dimensions. and It is a weighting coefficient. This factor serves as a key input feature of the fusion prediction model. Its technical effect is to provide the model with profound insights into the trend of stress state deterioration. It enables the model not only to see where the stress is high, but also to combine the areas where the stress is most concentrated and the areas where the stress increases the fastest, thereby more accurately locating the high-risk areas most likely to produce internal defects such as microcracks.

[0113] Weighting coefficient and It is a dimensionless empirical parameter, the determination of which depends on the specific material properties and failure mode; for brittle materials that are more susceptible to stress concentration under static loads, The weight will be higher; for scenarios where dynamic fatigue accumulation is more critical, The weights should be increased accordingly; the optimal values ​​of these weights are determined through hyperparameter optimization, specifically to improve the fusion prediction model. The optimization objective is to achieve the final prediction accuracy on the historical defect sample dataset. Methods such as grid search or Bayesian optimization are employed to optimize this accuracy. and To optimize the model and maximize its prediction accuracy, these two coefficients are also used to balance two terms with different physical dimensions and combine them into a unified risk score.

[0114] Latent Defect Index The calculation of this value is the core output of this module, representing the comprehensive confidence level of future defects.

[0115] ;

[0116] This formula represents a deep learning fusion prediction model. The technical motivation is that a single indicator is difficult to detect early defects caused by the combined effect of multiple factors. For example, internal microcracks may only manifest as weak acoustic emission signals and slight stress field changes. Any single signal may be drowned out by noise. It is necessary to use a powerful nonlinear model to fuse multi-dimensional evidence such as internal stress, acoustics, and vision in order to effectively identify such complex and latent defect patterns.

[0117] in, It is the latent defect index of region i at time t, and its value range is normalized to between 0 and 1, representing the confidence of defect occurrence; This represents a trained deep learning model, such as a time-series model based on a Transformer or Gated Recurrent Unit (GRU) architecture, which is trained to identify complex association patterns in multimodal data.

[0118] A deep learning network containing gated recurrent units and an attention mechanism can be used; the specific architecture can be set as follows: it takes four input data sequences from the past N time steps: real-time stress field... Acoustic emission signal characteristics Surface visual features and stress gradient risk factor The features are fed into four parallel GRU layers to extract their respective time-dependent features. A multi-head attention layer then weights and fuses these four feature streams, enabling the model to dynamically focus on the data source that contributes most to defect prediction at a specific time. The fused feature vector is then passed through a fully connected layer and a sigmoid activation function to output a latent defect index ranging from [0, 1]. The training dataset for this model consists of historical production data. Each data sample contains multimodal data over a continuous N time steps. The labels are set based on whether manually verified defects appear within M time steps after the current time period. For example, 1 indicates a defect, and 0 indicates no defect. Here, N is the time step size of the input sequence, representing the length of the time window during which the model reviews historical data; M is the time step size for predicting the future, used to define the training labels. The values ​​of N and M can be determined through experimental optimization based on the characteristic time of the production process and the typical cycle of defect evolution.

[0119] The model's input includes the real-time stress field from Module 1. Acoustic emission signal characteristics from the high-frequency acoustic sensor array, corresponding region i Surface microstructure features of region i from a high-resolution line scan camera And the stress gradient risk factor calculated in this module. ;

[0120] This model analyzes multimodal data sequences from the past N time steps to output a final judgment on the risk of latent defects at the current moment. Its technical advantage lies in achieving a leap from passive detection to proactive prediction; it can not only discover existing defects but also predict latent defects that will form at some future point in time, providing quantifiable risk indicators. and As a core input, it is passed to the closed-loop adaptive control logic module, providing a solid foundation for triggering more targeted predictive control strategies at different levels.

[0121] Example 5

[0122] The closed-loop adaptive control module includes: inputting the deviation between the time-series stress field dataset and the reference stress field dataset into the inverse model to generate the control target error vector; establishing decision logic that maps the drift risk level and latent defect risk index into adaptive control gain; and generating adaptive control gain based on the decision logic.

[0123] The closed-loop adaptive control module also includes: multiplying the control target error vector with the adaptive control gain to generate a control command; wherein the control command is the final control adjustment vector;

[0124] The closed-loop adaptive control module also includes: adding the final control adjustment vector to the current production parameter setpoint vector to generate an updated setpoint vector; the closed-loop adaptive control module sends the updated setpoint vector to the production line execution unit to realize control closed loop;

[0125] The production parameter closed-loop adaptive control logic module of this invention serves as the decision-making and execution terminal of the entire system. Its core function is to integrate the evaluation and prediction results of all upstream modules, dynamically calculate the optimal control adjustment amount, determine the adjustment intensity according to the risk level, and ultimately achieve a truly adaptive closed-loop control.

[0126] Control target error vector The calculations aim to determine the ideal direction and magnitude of the adjustment;

[0127] ;

[0128] The core of this formula is the inverse model. The technical motivation lies in directly addressing stress deviations. Deducing the production parameters that need to be adjusted is a complex, nonlinear inverse problem. By establishing an inverse physical model or training a neural network to fit this inverse relationship, the idealized amount of production parameter adjustment required to completely eliminate the current stress deviation can be directly calculated.

[0129] in, It is a vector that contains ideal adjustment values ​​for multiple production parameters such as tension, speed, and temperature; It is an inverse physics model or a pre-trained neural network whose input is stress deviation and whose output is the parameter error that causes the deviation; It is the real-time stress deviation at the current moment, provided by Module 2;

[0130] Inverse model Preferably, this is achieved by training a neural network; the training data is generated as follows: a forward physical simulation model that matches the actual production line is constructed using finite element analysis software. This model can generate the model based on the input production parameters. Accurate simulation output of the corresponding steady-state global stress field ; through standard parameters of gold batches Numerous, systematic random perturbations are performed in the vicinity, generating tens of thousands of parameter perturbation-stress deviation data pairs, i.e. ,in , ; based on stress deviation The parameter perturbation that causes this bias is used as input to the neural network. As output, train an inverse model such as a multilayer perceptron. Once training is complete, the model can directly and quickly estimate the amount of ideal parameter adjustment needed to eliminate the stress deviation based on real-time stress deviation. ;

[0131] This step is the first step in the control demand calculation. The technical effect is that it directly translates the problem found by the upstream module, namely the stress deviation, into a solution that the downstream module can understand, namely the parameter adjustment amount, which provides a clear target and direction for the subsequent generation of control commands.

[0132] Adaptive gain matrix The purpose of generating this is to determine the aggressiveness of the control system's response;

[0133] ;

[0134] This formula originates from adaptive control theory. The technical motivation is that traditional fixed-gain controllers cannot cope with complex and time-varying operating conditions. The response strength of an intelligent system should match the risk level: for small process fluctuations, smooth and small-amplitude fine-tuning should be used to maintain stability; while for serious quality overruns or high-risk defect predictions, rapid and large-amplitude corrective interventions must be taken. This is the core of achieving this dynamic adjustment;

[0135] in, It is a diagonal matrix, and its diagonal elements are... Adjustment gains corresponding to different production parameters; functions It is a decision logic, typically implemented by an enforceable set of fuzzy logic rules or a rule engine, which will take the two drift indices from Module 2. And two defect risk indicators from Module 3 These four risk indicators serve as inputs and are mapped to specific gain values.

[0136] This function is the adaptive core of this module; its technical effect lies in constructing a true adaptive closed-loop control system. The decision logic g operates based on a predefined set of rules; for example, if the input... A value greater than 1 corresponds to a level 2 warning, and g is... Assign higher values ​​to the diagonal elements, such as those in the range of 0.8 to 1.5; conversely, if... Not greater than 1 but Exceed For a Level 1 warning, g is assigned a lower value, such as within the range of 0.2 to 0.5. This dynamic gain mechanism ensures that the control response is both precise and reasonable.

[0137] This logic employs a hierarchical priority and weighted combination approach, comprehensively... The gain matrix is ​​determined by the maximum value of four risk indicators across all regions i. ;

[0138] Highest priority: If the latent defect index exceeds the critical value, for example... Or the absolute overlimit index is greater than 1 ( If the system determines that there is a serious quality risk, then... The diagonal elements are set in a higher value range, such as [0.8, 1.5]; secondary priority: if the first-level condition is not met, but the stress gradient risk factor exceeds its warning threshold ( ) or the relative drift index exceeds its warning threshold ( The system's decision-making process is prone to instability or potential defect evolution risks. The diagonal elements are set in a lower value range, such as [0.2, 0.5]; Normal state: If none of the above conditions are met, a very small maintenance gain is set or no adjustment is made; Within a defined value range, the specific gain value can be proportional to the magnitude of the risk indicator exceeding the threshold, for example:

[0139] ;

[0140] in: This represents taking the absolute overlimit index among all discrete subregions i. The maximum value means that the control system will determine the adjustment intensity based on the most risky area on the entire material, thereby achieving more precise adaptive control.

[0141] The settings can be referenced from historical data; the batches that ultimately produce defects are those before the defects are formed. The distribution of factors is analyzed, and specific quantiles are used as early warning thresholds.

[0142] Final control adjustment vector The calculation and execution of control commands constitute the final formation and closed loop of control instructions.

[0143] ;

[0144] These two formulas are the basic execution logic of a standard digital control system. The motivation is to combine the ideal adjustment amount with the adjustment force determined by risk to generate the final executable instruction, update the system setpoint, and complete one control iteration.

[0145] in, This is the core output of this module, namely the final, executable control adjustment vector; It is a vector of current production parameter setpoints. These are the updated settings;

[0146] The final calculation of the control adjustment vector is the output of this module, which determines the ideal adjustment direction and magnitude. The degree of adjustment determined by the current risk level Multiplying these components yields a precise and reasonable final adjustment instruction, which is then sent to various execution units on the production line, such as motors and heaters, to update the system's setpoints. The completion of this process signifies that the entire system has achieved a complete, intelligent, and adaptive control cycle from perception, modeling, evaluation, prediction to decision-making and execution.

[0147] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A large-model-driven data governance decision support system, characterized by: include: The physical field evolution modeling module collects multimodal state datasets, establishes a real-time global stress field based on the multimodal state datasets, and outputs a time-series stress field dataset. The dynamic drift assessment module receives the time-series stress field dataset, compares it with the preset reference stress field dataset, and calculates the drift risk level. The defect evolution prediction module collects a defect-related feature dataset, and uses the time-series stress field dataset and the defect-related feature dataset as inputs to the fusion prediction model to generate a latent defect risk index. The closed-loop adaptive control module integrates the drift risk level and the latent defect risk index to generate an adaptive control gain, calculates the deviation between the time-series stress field dataset and the reference stress field dataset, and generates control commands to adjust production parameters based on the adaptive control gain and the deviation. The multimodal state dataset includes: polarization delay data, material thickness data, surface temperature data, and material transport velocity data; The physical field evolution modeling module reconstructs the mechanical stress field based on the polarization delay data, the material thickness data, and the surface temperature data. The physical field evolution modeling module calculates the thermal stress field based on the surface temperature data; The real-time global stress field is formed by the superposition of the mechanical stress field and the thermal stress field; The dynamic drift evaluation module includes: Calculate the difference between the time-series stress field dataset and the reference stress field dataset to generate a real-time deviation; The relative drift index is calculated based on the real-time deviation and the reference stress field dataset. The absolute over-limit index is calculated based on the real-time deviation and the preset maximum allowable stress deviation. The determination of the drift risk level depends on the relative drift index and the absolute over-limit index; The defect evolution prediction module also includes: Based on the spatial gradient and temporal rate of change of the time-series stress field dataset, the stress gradient risk factor is calculated. The defect evolution prediction module takes the time-series stress field dataset, acoustic emission signal feature data, surface visual feature data, and stress gradient risk factor as inputs to the fusion prediction model to output a latent defect index. The closed-loop adaptive control module includes: The deviation between the time-series stress field dataset and the reference stress field dataset is input into the inverse model to generate the control target error vector. Establish a decision logic that maps the drift risk level and the latent defect risk index to the adaptive control gain; The closed-loop adaptive control module generates the adaptive control gain based on the decision logic; The closed-loop adaptive control module also includes: The control target error vector is multiplied by the adaptive control gain to generate the control command; The control command is the final control adjustment vector.

2. The large model-driven data governance decision support system according to claim 1, characterized in that, The physical field evolution modeling module also includes: Based on the material transport velocity data, calculate the advection term describing the stress field transport effect with material motion; By simulating the stress relaxation within the material, the diffusion term is obtained; Based on the difference between the newly calculated stress and the stress field at the previous moment, a source term is set to drive the model update. The physical field evolution modeling module establishes the dynamic evolution equation of the stress field based on the advection term, the diffusion term, and the source term; The physical field evolution modeling module updates the real-time global stress field based on the stress field dynamic evolution equation.

3. The large model-driven data governance decision support system according to claim 2, characterized in that, The dynamic drift evaluation module also includes: Compare the absolute exceedance index with the quality compliance threshold; Compare the relative drift index with the process stability warning threshold; When the absolute exceedance index is greater than the quality compliance threshold, the drift risk level is determined as a quality exceedance warning. When the absolute exceedance index is not greater than the quality compliance threshold and the relative drift index is greater than the process stability warning threshold, the drift risk level is determined as a process instability warning. When the absolute exceedance index is not greater than the quality compliance threshold and the relative drift index is not greater than the process stability warning threshold, the drift risk level is determined to be qualified.

4. The large model-driven data governance decision support system according to claim 1, characterized in that, The defect-related feature dataset includes: acoustic emission signal feature data acquired by an acoustic sensor array, and surface visual feature data acquired by a line scan camera; The latent defect risk indicators include: stress gradient risk factor and latent defect index.

5. The large model-driven data governance decision support system according to claim 4, characterized in that, The closed-loop adaptive control module also includes: The final control adjustment vector is added to the current production parameter setpoint vector to generate an updated setpoint vector; The closed-loop adaptive control module sends the updated setpoint vector to the production line execution unit to achieve closed-loop control.

Images