A trusted process window construction method and system for industrialization verification of scientific and technological achievements

Abstract

The application provides a trusted process window construction method and system for industrialization verification of scientific and technological achievements, and relates to the technical field of industrial process control, and comprises the following steps: S1, obtaining achievement technical information corresponding to the scientific and technological achievements to be verified and industrial data of a target industrial process; S2, performing calibration driven by credibility; S3, performing credibility quantification verification based on at least one of a device disturbance component, an environmental disturbance component and a node residual error; S4, generating a self-adaptive backflow flow direction control signal bound with physical properties of a process node; S5, performing a joint iterative determination process to update intrinsic parameters, independent physical error components and process control variables; S6, calculating a compliance probability of a terminal quality index falling within a process specification; and S7, outputting a set of process control variables as a trusted process window of the scientific and technological achievements in the target industrial process. The application can form a closed-loop technical system from achievement verification, production line amplification to abnormality troubleshooting.

Description

Technical Field

[0001] This invention relates to the field of industrial process control technology, and more particularly to a method and system for constructing a reliable process window for the industrialization verification of scientific and technological achievements. Background Technology

[0002] When scientific and technological achievements are transformed from laboratory research and development to industrial mass production, they need to undergo a crucial stage of industrialization verification. The core task of industrialization verification is to confirm whether the scientific and technological achievements can stably produce industrial products that meet quality specifications under the equipment conditions, environmental fluctuations, batch differences of incoming materials, and testing chain constraints of the target industrial production line. This verification process involves cross-scale mapping of laboratory mechanism models to industrial scenarios, range screening of process parameters, and fusion and utilization of multi-source heterogeneous data. The reliable process window refers to the set of process control variables that can ensure that the quality indicators of the end product meet the specifications with a probability no less than a preset value. This window not only includes the value range of the control variables, but also covers equipment tolerances, allowable environmental fluctuation ranges, and abnormal triggering conditions. It is the core criterion for determining whether a scientific and technological achievement can enter the industrial execution stage.

[0003] In existing technologies, the industrialization verification of scientific and technological achievements mainly relies on methods such as trial and error, conventional model fitting, single-point optimization, digital twin modeling, statistical process control, and sampling simulation. Trial and error methods approximate feasible process ranges by successively adjusting process parameters and observing changes in product quality; this method is highly dependent on the personal experience of engineers and has a lengthy trial production cycle. Conventional model fitting methods use experimental or field data to establish regression or response surface models between process parameters and quality indicators, thereby solving for parameter combinations that meet specifications. Single-point optimization methods search for a single optimal process parameter point with the goal of optimizing the quality indicator. Digital twin modeling methods construct virtual mirror models of industrial processes, simulating product quality performance under different process conditions in a virtual environment. Statistical process control methods use control charts to monitor the production line's operating status and issue alarm signals when quality indicators exceed control limits. Sampling simulation methods evaluate the robustness of process schemes by randomly sampling model parameters or disturbance terms.

[0004] The aforementioned existing technologies have several shortcomings. There is a lack of calculable mapping between the mechanistic information obtained in the laboratory stage and the actual conditions of the industrial production line. Laboratory sample data, simulation data, pilot-scale data, and field observation data are used independently without being incorporated into a unified verification framework. This leads to frequent performance degradation or yield collapse of technical solutions that pass laboratory verification after entering the industrial production line. The residuals between model predictions and actual production line observations are difficult to attribute, making it impossible to distinguish the contribution of equipment noise, environmental disturbances, measurement link deviations, timing drift, and model structural deficiencies. When anomalies occur, engineers cannot determine whether to supplement experimental data, adjust the model structure, maintain equipment status, or redefine process boundary conditions. The determination of the process window relies excessively on single-point optimal values ​​or empirical intervals, lacking a systematic reverse-engineering method based on probabilistic compliance criteria. It ignores the tail risk caused by the superposition of uncertainties such as equipment tolerances, environmental fluctuations, and incoming material batch drift, resulting in an actual compliance rate far lower than expected during field execution. When quality anomalies occur during continuous trial production or mass production, the lack of technical means to trace back from the final quality defect to the process node makes it difficult for engineers to quickly locate the failure point and the source of the dominant disturbance. The troubleshooting process usually relies on step-by-step inspection and repeated testing, which is time-consuming and costly. These problems collectively restrict the technological progress of scientific and technological achievements from laboratory mechanisms to reliable verification and stable mass production on industrial production lines. Summary of the Invention

[0005] In view of this, the purpose of this invention is to propose a method and system for constructing a reliable process window for the industrialization verification of scientific and technological achievements, so as to solve the technical problem that existing laboratory data cannot be included in a unified verification framework.

[0006] The technical means employed in this invention are as follows: A method for constructing a reliable process window for the industrialization verification of scientific and technological achievements includes the following steps: S1. Obtain the technical information of the scientific and technological achievements to be verified and the industrial data of the target industrial process. Transform the laboratory mechanism information and data characteristics carried in the technical information of the achievements into industrial scenario model information through multi-source heterogeneous elution operators. The industrial scenario model information includes state evolution path, control action channel, physical boundary conditions and interface uncertainty. S2. Based on industrial data, perform credibility-driven calibration on the intrinsic model and observation model in the industrial scenario model information to obtain the posterior distribution of parameters and the node residuals that characterize the mismatch between the model and the actual production line. S3. Explicit error information is extracted from the node residuals, and the equipment disturbance component and environmental disturbance component are decomposed. Trusted quantification verification is performed based on at least one of the equipment disturbance component, environmental disturbance component and node residuals. If the trusted quantification verification does not meet the preset conditions, S4 is executed. If the trusted quantification verification meets the preset conditions, S5~S7 are executed. S4. Convert at least one result of the credible quantification verification into an over-limit index. Determine the dominant over-limit item based on the over-limit amount of each over-limit index relative to the preset upper limit threshold. Determine the input object to be reinforced in S1 or the model object to be reinforced in S2 based on the dominant over-limit item. Generate an adaptive backflow direction control signal. The adaptive backflow direction control signal is used to instruct the input object to be reinforced or the model object to be reinforced to be reinforced. Return the reinforced object to S1 or S2 to replace the corresponding original input or original model object. S5. Perform a joint iterative decision process to update intrinsic parameters, independent physical error components, and process control variables; S6. Based on the process topology of the target industrial process, perform full-distribution uncertainty propagation on the state evolution path, interface uncertainty, updated independent physical error components and updated process control variables, and calculate the probability that the terminal quality index falls into the process specification. S7. Inversely calculate the set of process control variables that make the probability of achieving the target satisfy the preset target probability, and output the set of process control variables as the credible process window of the scientific and technological achievement in the target industrial process.

[0007] Furthermore, in S1, the technical information and industrial data of the results are derived from a multi-source heterogeneous dataset, which includes laboratory sample test data, mechanism simulation data, pilot production line trial production data, and on-site observation data of the target industrial line. The multi-source heterogeneous elution operator is used to extract the interface mapping relationship of each data source under different spatial scales and time periods, remove or replace the real equipment identity, real process formula and customer production identifier, and calculate the interface state covariance as the uncertainty input amplified across scales.

[0008] Furthermore, in S2, the industrial scenario model information is represented by the following state equations and observation equations:

[0009]

[0010] in, Indicates the first The state variables of each process node Indicates the first Control variables for each process node Represents the observed variable. Indicates intrinsic parameters, Indicates the observed parameters, and These represent process disturbances and observational disturbances, respectively.

[0011] Furthermore, in S2, when the industrial scenario model information flows across process nodes, a standard specification interface card carrying the state expectation and error covariance matrix is ​​used; the standard specification interface card includes at least three of the following: state estimate, state covariance, unit, reference benchmark, timestamp, process area identifier, batch identifier, equipment capability identifier, scale conservation conversion field, and data washing field, wherein the equipment capability identifier is used to replace the real equipment identity information.

[0012] Furthermore, in S3, the explicit error information extraction of the nodal residuals includes the following steps: The node residuals are mapped to a multidimensional physical evolution space jointly defined by the process node state basis, interface uncertainty covariance, and physical boundary topology. After physical error decoupling operator and functional consistency constraint, at least one of the following is obtained: equipment disturbance component, environmental disturbance component, time-varying drift component, detection link disturbance component, systematic offset component, and mechanism structure mismatch component. Each component obtained by decomposition is used to characterize the source of physical mismatch when the scientific and technological achievements are scaled up across the target industrial process.

[0013] Furthermore, in S3, the credible quantification verification includes at least one of error closure verification, prediction coverage consistency verification, equipment disturbance strength verification, environmental coupling disturbance verification, systematic offset verification, time series drift trend verification, and tail risk contribution verification; the threshold, weight, and judgment template of the credible quantification verification are configured according to industry, equipment family, material system, or process type.

[0014] Furthermore, the trusted quantification verification also includes: A multidimensional confidence metric tensor is constructed based on one or more error components. The multidimensional confidence metric tensor is compared with a preset confidence condition. When the multidimensional confidence metric tensor meets the preset confidence condition, the calibrated model is determined to proceed to S7.

[0015] Furthermore, in S5, the joint iterative determination process includes the following steps: Fix the error components obtained from the extraction and decomposition of the explicit error information and the process control variables, and update the intrinsic parameters and observation parameters; Fix the updated intrinsic parameters and observation parameters, and update the error components obtained from the decomposition. Fix the updated intrinsic parameters, observed parameters, and error components, and update the process control variables; When the changes in the comprehensive failure measure, the probability of achieving the target, and the multidimensional credibility metric tensor satisfy the preset convergence conditions, the output is a stable and acceptable state.

[0016] Furthermore, the comprehensive loss measure is determined based on a weighted combination of model consistency loss, error closure loss, credibility metric constraint loss, target achievement loss, and industry constraint loss, with each weight configured according to industry, equipment family, material system, or process type.

[0017] This invention also provides a reliable process window construction system for verifying the industrialization of scientific and technological achievements, comprising: The data receiving module is used to acquire the technical information of the scientific and technological achievements to be verified and the industrial data of the target industrial process. The industrial scene mapping module is used to transform the laboratory mechanism information and data characteristics carried in the technical information of the results into industrial scene model information containing state evolution path, control action channel, physical boundary conditions and interface uncertainty through multi-source heterogeneous elution operators. The model calibration module is used to calibrate the intrinsic model and observation model in the industrial scenario model information based on industrial data, and obtain the posterior distribution of parameters and the node residuals that characterize the mismatch between the model and the actual production line. The error information extraction module is used to perform explicit error information extraction on the node residuals, and at least decompose them into equipment disturbance components and environmental disturbance components. The credibility evaluation module is used to perform credibility quantification verification based on at least one of the device disturbance component, environmental disturbance component, and node residual; The dynamic backflow module is used to calculate the normalized over-limit of each verification index and determine the dominant over-limit item when the credibility metric verification does not meet the preset conditions. It generates an adaptive backflow direction control signal according to the preset mapping between the dominant over-limit item and the data source, device, detection, boundary condition or model. The reinforced dataset, device status, observation parameters, boundary conditions or intrinsic model are used as the update input of the corresponding module. The iterative decision module is used to perform a joint iterative decision process to update intrinsic parameters, independent physical error components and process control variables when the credibility quantification verification meets the preset conditions. The propagation and solution module is used to propagate the fully distributed uncertainty according to the process topology of the target industrial process and calculate the probability of the final quality index falling into the process specification. The window output module is used to inversely calculate the set of process control variables that make the probability of compliance meet the preset target probability and to determine the set of process control variables as the reliable process window.

[0018] Compared with the prior art, the present invention has the following advantages: This invention utilizes a multi-source heterogeneous elution operator to unify laboratory mechanism information, sample test data, mechanism simulation data, pilot production line data, and target industrial line field observation data from scientific and technological achievements into industrial scenario model information that includes state evolution paths, control action channels, physical boundary conditions, and interface uncertainties. This achieves a computable mapping from laboratory conditions to industrial conditions. Based on this, the invention maps the node residuals generated after model calibration to a multi-dimensional physical evolution space. Through a physical error decoupling operator, independent physical error components are obtained, including intrinsic equipment disturbances, environmental coupling disturbances, time-varying drift, detection link disturbances, systematic offsets, and mechanism-structure mismatches. This ensures that model prediction biases are no longer indivisible black-box values ​​but can be attributed to specific physical sources. This mechanism overcomes the predicament of unexplainable errors and unattributable anomalies in traditional industrialization verification, providing a quantitative means for identifying physical mismatches during cross-scale scaling.

[0019] This invention constructs a multidimensional reliability metric tensor based on independent physical error components. It performs at least one reliability quantification verification among error closure verification, prediction coverage consistency verification, and equipment disturbance intensity verification. The reliability metric tensor is compared with preset reliability conditions, serving as a quantitative gate for determining whether the model enters the window inverse calculation step. When the reliability verification fails to meet the preset conditions, the system blocks the window output process and inputs the characteristics of the out-of-limit error components into an adaptive decision operator. This generates a backflow direction control signal bound to the physical attributes of process nodes, driving directional reinforcement of physical boundaries, experimental samples, equipment status, or detection links. This mechanism forcibly filters out unreliable model predictions at the engineering level, incorporating model calibration, error decomposition, and reliability evaluation into the same closed-loop control framework, avoiding the direct use of unverified models for industrial validation decisions.

[0020] This invention performs a joint iterative judgment process when the credibility verification meets preset conditions. It alternately updates intrinsic parameters, independent physical error components, and process control variables until the comprehensive failure rate and target compliance probability satisfy convergence conditions. Subsequently, it performs full-distribution uncertainty propagation on the state evolution path, interface uncertainty, error components, and control variables according to the process topology to obtain the probability distribution of the final quality index meeting specifications. It then inversely calculates the set of process control variables that make the compliance probability meet the preset target probability as the reliable process window output. This window includes at least two types: control variable intervals, equipment tolerances, environmental fluctuation ranges, and key risk contribution rankings. These can be directly converted into equipment setpoint intervals, allowable tolerances, environmental control ranges, and online detection frequencies for industrial line execution. This mechanism establishes the determination of the process window based on probabilistic compliance criteria and convergence control principles, unlike traditional methods that rely on single-point optimal values ​​or empirical intervals. It solves the problem in existing technologies where the actual compliance rate is lower than expected during on-site execution of the process window. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a flowchart of the method of the present invention.

[0023] Figure 2 This is a schematic diagram of the system structure of the present invention. Detailed Implementation

[0024] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0025] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0026] To facilitate understanding by those skilled in the art regarding the deployment of this invention in engineering systems, several broad terms are explained below. These terms are used only to explain the functional boundaries and engineering implementation of this invention, and do not require that actual products use the same names, nor do they limit the scope of protection of the claims.

[0027] The scientific and technological achievements in this invention refer to technical information with a clear technological source, which may originate from material systems, process principles, equipment principles, control principles, experimental laws, simulation models, or prototypes. These achievements are not abstract scientific discoveries themselves, but rather technical information that can be mapped to industrial process state variables, control variables, observation variables, boundary conditions, and target specifications.

[0028] The industrialization verification in this invention refers to verifying, under target industrial processes, target equipment conditions, and target quality specifications, whether the scientific and technological achievements possess an executable and stably reproducible set of process control variables. The output of this process is not a commercial evaluation conclusion, but rather a reliable process window that can be used for industrial execution.

[0029] In this invention, the reliable process window refers to the set of process control variables obtained when the model reliability, error closure, coverage, and target achievement probability all meet preset conditions. The reliable process window may include control variable intervals, variable covariance, equipment tolerances, environmental fluctuation ranges, and abnormal backflow conditions.

[0030] This invention can be deployed as a physical engineering architecture, namely, an industrialization verification architecture that takes the intrinsic mechanism of scientific and technological achievements as its source, industrial physical scenarios as its carrier, and error decoupling, credibility metric, feedback control, and uncertainty window inverse calculation as its core components. In practical engineering systems, this architecture can be called a Problem-to-Asset Engine, abbreviated as PAE.

[0031] The dynamic quantitative evaluation access control described in this invention refers to a control mechanism that makes a technical judgment on whether a scientific and technological achievement should enter the next stage of industrial verification based on multidimensional credibility metric tensors, error closure degree, target achievement probability, and industrial execution constraints. In engineering systems, this access control can correspond to the readiness level control of scientific and technological achievements or industrial objects, which can be abbreviated as ORL; this level control is used to describe the technical verification stage and does not constitute the commercial evaluation rules themselves.

[0032] The multidimensional reliability metric tensor described in this invention refers to a model reliability metric quantification object induced by independent physical error components. It can include physical action dimensions such as intrinsic device disturbances, environmental coupled disturbances, systematic offsets, temporal evolution trends, prediction coverage deviations, and tail risk contributions. This object is used to control whether the model can enter the window for inverse calculation, whether feedback control signals need to be triggered, and whether it possesses the reliability to support industrial pilot production.

[0033] The Incremental Metric Layer (IML) criterion described in this invention refers to a criterion that, in a cross-scale state evolution path, projects perturbation information with different dimensions, sampling frequencies, and physical sources onto a unified mechanism space for consistency measurement by constructing a mechanism metric tensor G(X). The perturbation information may include pump flow fluctuations, hot plate edge temperature differences, segmented drying humidity perturbations, equipment actuator aging, detection link offsets, or batch material fluctuations. This criterion is used to explain the physical correspondence between residual projection, error component extraction, and reliable metric verification in a multi-dimensional physical evolution space. Its specific metric parameters, projection matrix, threshold library, and industry templates can be reserved as engineering configuration items or trade secrets.

[0034] The physical error decoupling operator described in this invention refers to a processing mechanism that maps node residuals to a multidimensional physical evolution space and extracts independent physical error components such as time-varying drift, environmental coupling disturbances, intrinsic equipment disturbances, detection link disturbances, systematic offsets, and mechanistic mismatches. In engineering implementation, this mechanism can manifest as one or more of the following: error ledger generation, residual attribution, credibility constraint construction, or uncertainty allocation.

[0035] The multi-source heterogeneous elution operator described in this invention refers to a processing mechanism that unifies the scale, normalizes the units, aligns the interfaces, replaces equipment capability labels, and extracts uncertainties for laboratory data, simulation data, pilot-scale data, and target industrial line data. Through this mechanism, the true identity of equipment, the true process formula, and the customer's site identification can be replaced by capability labels, normalized variables, or dimensionless parameters, thereby enabling the model to complete cross-scale mapping while protecting industrial secrets.

[0036] The cross-scale intrinsic state evolution path model described in this invention refers to a set of models that uniformly transform laboratory mechanisms, sample data, simulation data, pilot-scale data, and target industrial line data from scientific and technological achievements into state variables, control variables, observation variables, boundary conditions, interface states, and their uncertainties. In actual system implementation, this cross-scale intrinsic state evolution path model can also be called a scenario equation package or scenario exchange protocol, abbreviated as SEP. Its main function is to remove hardware identity, customer formulas, and field secrets, while retaining the physical interaction relationships that can be used for industrial verification.

[0037] The standard interface card described in this invention refers to a universal interface contract used in cross-node data assembly in industrial topologies for the expected state of flow, error covariance matrix, scale conservation conversion fields, and data washing fields. Through this interface card, the uncertainty boundary of the output of the preceding node is explicitly issued and absorbed by the following node, avoiding black box collapse caused by undefined boundary coupling between the output of the previous process and the input of the next process.

[0038] The physical readiness control access control described in this invention refers to a dynamic technology access control system that uses independent physical error components, error closure degree, reliability metric, target achievement probability, and industrial execution constraints as inputs to determine whether a scientific and technological achievement is ready for pilot testing, continuous trial production, or mass production verification. This access control system is not intended for commercial maturity evaluation, but rather to control whether the status of the model, data, and equipment is sufficient to support the next stage of industrial physical verification.

[0039] like Figure 1 As shown, this invention provides a method for constructing a reliable process window for the industrialization verification of scientific and technological achievements, including the following steps: The system first acquires the technical information corresponding to the scientific and technological achievement to be verified. For example, in thin film preparation processes, the technical information may include film formation mechanism, fluid transport laws, heat treatment laws, laboratory sample thickness data, simulation samples, and target thickness uniformity indicators. Subsequently, the system acquires on-site observation data, equipment status data, environmental data, and process specifications of the target industrial process.

[0040] The technical information and industrial data are derived from multi-source heterogeneous datasets, including laboratory sample test data, mechanism simulation data, pilot production line data, and on-site observation data from the target industrial line. The system extracts the interface mapping relationships of each data source under different spatial scales and time periods, and calculates the interface state covariance, using it as the uncertainty input in the cross-scale amplification process. Therefore, the mechanism information formed under laboratory conditions is not directly equivalent to production line conditions, but rather undergoes mapping through boundary conditions, equipment capabilities, material batches, environmental fluctuations, and testing links.

[0041] Industrial scenario model information may include: System boundaries, such as equipment sections, process sections, material sections, or testing sections; State variables, such as film thickness, temperature, concentration, pressure, surface energy, residual stress, or defect density; Control variables, such as rotational speed, flow rate, temperature, pressure, time, power, scanning speed, or ratio; Observed variables, such as sensor readings, detection patterns, quality indicators, or end-product defect indicators; Boundary conditions, such as thermal boundaries, force boundaries, flow boundaries, chemical boundaries, or time boundaries; Interface status information, such as the state estimate and covariance from the output of the previous process node to the next process node.

[0042] Furthermore, the cross-node data assembly of the industrial topology is processed through a standard specification interface card carrying the state expectation and error covariance matrix. This standard specification interface card has standard data washing fields, scale conservation conversion fields, capability label fields, and interface uncertainty fields, enabling the explicit distribution of uncertainty boundaries from preceding nodes and their absorption by subsequent nodes, thereby preventing black-box collapse caused by undefined boundary coupling.

[0043] No. The state update relationship of a process node can be represented as:

[0044] The observation relationship can be expressed as:

[0045] in, Representing the intrinsic model, Represents the observation model, Represents state variables, Indicates control variables, Represents the observed variable. and These represent the parameters to be calibrated. and These represent process disturbances and observational disturbances, respectively.

[0046] Interface status information can be represented as:

[0047] in, This represents the state estimate. Represents the state covariance. Indicates the unit. Indicates the reference standard, Represents a timestamp. Indicates batch identifier, This indicates the equipment's capability.

[0048] The system calibrates the intrinsic model and the observation model based on experimental samples, simulation samples, and field observation data. The calibration goal is not merely to obtain a single optimal parameter, but to obtain the posterior distribution or covariance information of the parameters for subsequent uncertainty propagation.

[0049] The calibrated predicted value can be expressed as:

[0050] The nodal residuals can be expressed as:

[0051] To improve error attribution capability, the system maps node residuals to a multi-dimensional physical evolution space and performs explicit error decomposition using a physical error decoupling operator. For example, the node residuals can be represented as:

[0052] in, Indicates the system bias term. This represents the device drift term that changes over time. Indicates equipment noise. Represents the environmental noise term. Indicates the measured noise term. This indicates a missing element in the structure.

[0053] The aforementioned error components can be obtained through one or more of the following methods: repeated sample experiments, cross-batch experiments, window scanning experiments, equipment calibration, sensor calibration, hierarchical statistical models, time series decomposition, or random effects models.

[0054] The multidimensional physical evolution space can be characterized using a non-Euclidean geometric space with a mechanism-regulatory tensor. The system extracts the intrinsic state expectation, interface state covariance, boundary condition type, equipment capability label, and observation time series of each process node, and inputs the residual time series observed from the actual production line as external flow loads into the multidimensional physical evolution space.

[0055] Furthermore, in an implementation based on an incremental metric criterion, the system constructs a mechanistic metric tensor G(X) using state variables, control variables, observation variables, equipment capability labels, physical boundary conditions, and interface state covariance as mechanism space coordinates. This mechanistic metric tensor is used to perform equivalent scaling on industrial disturbances of different dimensions, enabling equipment noise, environmental noise, detection offset, and slow variable drift to be measured within the same non-Euclidean geometric mechanism space. For example, pump flow fluctuations can be considered as local disturbance loads along the equipment execution channel, and segmented drying humidity disturbances can be considered as anisotropic disturbance loads along the environmental boundary topology. Both are projected onto a unified mechanism space via the mechanistic metric tensor G(X), and then their consistency is measured based on the projection direction, projection intensity, and coupling relationship with the state evolution path. The above process provides a physically abstract path that can be implemented by those skilled in the art, but does not require the disclosure of specific numerical matrices, neural network weights, threshold libraries, or industry configuration templates.

[0056] The execution process of the functional consistency constraint includes: constructing a local physical basis based on the state evolution path, constructing a local metric based on the interface uncertainty covariance, calculating the projection component of the residual gradient in the local metric space, and determining the corresponding independent physical error component based on the coupling relationship between the projection component and the time basis, equipment state basis, environmental boundary basis, detection link basis, and structural mismatch basis. Projections exhibiting a continuous nonlinear steep descent or slow shift along the time basis can be identified as time-varying drift evolution characteristics; anisotropic fluctuations coupled with the plant physical boundary topology or environmental state variables can be identified as environmental disturbance characteristics; local random fluctuations related to equipment capability labels or actuator states can be identified as equipment intrinsic noise characteristics; deviations related to observation links, detection frequencies, or sensor states can be identified as detection link disturbance characteristics; residual projections that cannot be explained by the above components and are related to the intrinsic model structure can be identified as mechanistic structure mismatch characteristics.

[0057] The above representation method provides a physical decoupling path that can be implemented by those skilled in the art, but does not limit the specific numerical solver, threshold library, neural network structure, weight matrix or industry template; in actual implementation, projection decomposition, hierarchical statistical model, state space model, Bayesian calibration, random effects model or other equivalent physical constraint solving mechanism can be used.

[0058] The system constructs a multidimensional confidence metric tensor based on explicit error decomposition information. This multidimensional confidence metric tensor can be represented by a model confidence evaluation vector as follows:

[0059] The equipment noise ratio, ambient noise ratio, and offset ratio can be expressed as follows:

[0060]

[0061]

[0062] in, This indicates the allowable error or quality tolerance for this process node. Trend factor Coverage can be obtained from a trend scoring function of the residuals with respect to time, batch, equipment status, or control variables. This indicates the proportion of actual observed values ​​that fall within the predicted confidence interval.

[0063] Variance closure can be expressed as:

[0064] When the multidimensional confidence metric tensor meets the preset confidence conditions, the error closure degree is not lower than the closure threshold, and the coverage meets the coverage requirements, the system determines that the current model and error decomposition information can enter the process window to determine the steps. Otherwise, the system determines the dominant over-limit item according to the normalized over-limit quantity, and generates an adaptive backflow direction control signal according to the above preset mapping; after performing the corresponding reinforcement, the updated object replaces the corresponding input in S1 or S2 and returns to S1, specifically as follows: when the credibility quantification verification does not meet the preset conditions, the results of each credibility quantification verification in S3 are uniformly converted into over-limit index value M_j; the over-limit index value M_j includes at least one of error unclosure degree, prediction out-of-band rate, equipment disturbance ratio, environmental disturbance ratio, systematic offset ratio, time series drift trend factor and tail risk contribution rate, wherein the error unclosure degree is 1 minus the error closure degree, the prediction out-of-band rate is 1 minus the prediction coverage rate, and T_j is the preset upper limit threshold corresponding to the j-th over-limit index; calculate the normalized over-limit quantity q_j=max{0,(M_j T_j) / T_j}, select the index with the largest normalized exceedance and its corresponding error component as the dominant exceedance term; determine the reinforcement object based on the dominant exceedance term, the reinforcement object includes the technical information or industrial data in S1, and at least one of the physical boundary conditions, intrinsic model, intrinsic parameters, observation model, observation parameters and interface uncertainty in S2; wherein, exceeding the error unclosure limit corresponds to supplementing industrial data for error decomposition and re-estimating the error components, exceeding the predicted out-of-band rate or detection link disturbance limit corresponds to updating the observation model, observation parameters or interface uncertainty, and exceeding the equipment disturbance ratio limit corresponds to updating the equipment in the industrial data. If the process disturbance parameters in the state data or intrinsic model exceed the limit, the environmental disturbance ratio should be updated to update the environmental data or physical boundary conditions in the industrial data. If the systematic offset ratio exceeds the limit, the physical boundary conditions, intrinsic model or intrinsic parameters should be updated. If the time-series drift trend factor exceeds the limit, the time-varying terms or mechanism structure terms in the intrinsic model should be updated. If the tail risk contribution rate exceeds the limit, the boundary condition data of the corresponding process node should be supplemented or the interface uncertainty should be updated. An adaptive backflow direction control signal is generated based on the object to be reinforced. The reinforced object replaces the corresponding input in S1 or S2 and returns to S1. S1 to S3 are repeated until the credibility quantification verification meets the preset conditions. The system employs a joint iterative decision-making process to simultaneously update intrinsic parameters, error decomposition information, and process control variables. The comprehensive failure rate is defined as:

[0065] in, This represents the model consistency loss. Indicates the error closure loss. This represents the loss due to credibility constraints. Indicates the loss incurred in achieving the objective. This indicates industrial constraint loss.

[0066] In the t-th iteration, the system executes the following steps sequentially:

[0067]

[0068]

[0069] in, Indicates the first Explicit error decomposition information for each iteration This represents the error decomposition operator. This represents the domain of safety control variables.

[0070] The system is considered to have reached a stable and acceptable state when the following conditions are met simultaneously:

[0071]

[0072] in, This indicates the probability that the terminal quality indicators meet the specifications.

[0073] The target industrial process includes multiple process nodes. The system assembles these nodes according to the process topology to obtain the propagation relationship of the final quality indicators: Y_final=(G_K T_(K 1→K) … T_(1→2) G_1)(S_0,U_(1:K)) Where G_k represents the calibrated model of the k-th process node, T_(k→k+1) represents the interface mapping from the k-th process node to the (k+1)-th process node, S_0 represents the initial state, and U_(1:K) represents the set of process control variables from the 1st to the Kth process nodes.

[0074] symbol" "" is a commonly used symbol for function composition or mapping composition in mathematics, control theory, system modeling, and engineering calculations. Let mappings A: X→Y and B: Y→Z, then the composite mapping is denoted as B. A, and defined as (B) A(x) = B(A(x)), meaning that the right-hand mapping A is performed first, followed by the left-hand mapping B. Therefore, G_K in the formula of this invention... T_(K 1→K) … T_(1→2) G_1 represents the sequential connection of each process node model and its interface mapping according to the process topology of the target industrial process. The output of the previous mapping is used as the input of the next mapping, and finally the terminal quality index Y_final is obtained.

[0075] The system performs joint sampling or approximate propagation of the parameter posterior distribution, interface state covariance, error component distribution, initial state, and control variables to obtain:

[0076] The terminal compliance probability is:

[0077] A reliable process window is defined as the set of process control variables that are satisfied when the target quality specification, model reliability, and error closure condition are all met. Its inverse solution process is as follows. For example, a reliable process window can be represented as:

[0078] in, Indicates a reliable process window. This indicates the allowable failure rate. The system can use one or more of the following methods to solve for the reliable process window: grid search, Bayesian optimization, constraint optimization, importance sampling, adaptive sampling, or Monte Carlo sampling.

[0079] When the confidence condition is not met, the system does not directly output the process window. Instead, it calculates the normalized over-limit of each confidence verification index, determines the dominant over-limit item, and outputs an adaptive backflow direction control signal to the data source end, equipment end, detection end, boundary condition end or model end according to the above-mentioned preset mapping between bias, trend, equipment disturbance, environmental disturbance, detection link disturbance and predicted coverage and reinforcement object.

[0080] After the adaptive backflow direction control signal is executed, the new sample is incorporated into the original dataset to form D′, the updated device status forms E′, the updated observation model and observation parameters form H′ and ψ′, the updated boundary conditions form B′, and the updated intrinsic model and intrinsic parameters form F′ and θ′. The above updated objects replace the corresponding inputs in S1 or S2 respectively. The system returns to S1 and re-executes S1 to S3 until the credibility quantification verification meets the preset conditions.

[0081] When the target industrial process experiences abnormal terminal quality indicators, sudden defect changes, decreased yield, or increased out-of-band rate during continuous trial production or batch production, the system does not directly attribute the abnormality to a single device or a single control variable. Instead, it first obtains the process node status data, equipment status data, environmental data, online detection data, and digital process parameter table corresponding to the abnormal batch.

[0082] The system compares the state evolution path of the abnormal batch with the trusted process window to identify the abnormal state evolution path that deviates from the trusted process window; then, the node residual corresponding to the abnormal state evolution path is input into the multidimensional physical evolution space, and multiple independent physical abnormal components are obtained through the physical error decoupling operator.

[0083] Based on the process topology of the target industrial process, the system ranks the contribution of abnormal risks to each process node, cross-node interface, equipment status, environmental boundary, and detection link. For example, in the thin film coating scenario, if the deviation of the edge area of ​​the final film thickness continues to increase, the system can output the thermal boundary of the heat treatment edge, the stability of the coating liquid supply, or the edge detection link as candidate tracing directions; in the continuous mixing-forming-curing scenario, if the areal density variation and porosity increase simultaneously, the system can output the pumping link, the particle size drift of the powder batch, or the humidity boundary of the drying section as candidate tracing directions.

[0084] Through the above methods, this invention can not only positively determine the reliable process window during the technology transfer stage, but also reverse troubleshooting and physical attribution of abnormal quality indicators during industrial operation. This reverse tracing process uses multi-dimensional physical error decoupling and topological risk contribution ranking as technical basis, without relying on subjective judgment of anomaly causes based on human experience.

[0085] The following examples illustrate how the present invention is implemented when technological achievements are incorporated into industrial processes. The numerical values ​​in these examples are anonymized or normalized exemplary values ​​used to describe calculation processes and engineering actions, and do not constitute limitations on specific industry thresholds, customer equipment parameters, or actual process formulations. In actual implementation, thresholds, weights, and parameter mapping tables can be configured according to equipment families, material systems, process types, and corporate confidentiality requirements.

[0086] Example 1: Application of improved film uniformity results to industrial coating-heat treatment production lines In this embodiment, the technological achievement to be verified is a thin film uniformity improvement technique obtained on a laboratory scale. This achievement has been proven to reduce film thickness non-uniformity on laboratory samples, but the laboratory equipment, substrate size, environmental disturbances, and measurement frequency all differ from the target industrial production line. The target industrial process includes five process nodes: incoming material conditioning, coating, segmented heat treatment, cooling, and online inspection. The final quality indicators include average film thickness, film thickness non-uniformity, defect density, and edge region deviation.

[0087] The aforementioned data was uniformly categorized into four types of anonymized data sources: laboratory sample test data, mechanism simulation samples, pilot production line samples, and on-site observation samples from the target industrial line. The system performs unit normalization, scale alignment, capability label replacement, and interface uncertainty extraction on these four types of data sources to avoid disclosing the true identity of the equipment, the actual process formula, and customer site identification.

[0088]

[0089] The system maps scientific and technological achievements into industrial scenario model information. Taking a process node as an example, state variables may include local film thickness, solvent concentration, surface temperature, and equivalent viscosity; control variables may include coating speed, liquid supply intensity, processing time, heat treatment temperature, and ambient humidity; and observation variables may include average film thickness, in-plane non-uniformity index, and defect density.

[0090]

[0091] The in-plane non-uniformity index is characterized by the dispersion of the film thickness field within the effective process area, and the defect density is obtained by normalizing the results of online detection or sampling detection; all of the above variables can be represented by dimensionless variables or capability labels to avoid disclosing the real process formula.

[0092]

[0093] For example, the coating node can be described by a combination of an intrinsic coating evolution model and an observation model. The intrinsic coating evolution model reflects the effects of fluid spreading, liquid supply disturbance and thermal boundary, while the observation model reflects the mapping of the online film thickness detection link to the real state.

[0094]

[0095] Calibration parameters are stored in the form of posterior distribution or covariance information. Process disturbances and observation disturbances are incorporated into subsequent interface uncertainty and reliability metric constraints, respectively. After the first calibration, the system found a stable bias in the film thickness at the edge region, and the bias-related physical error characteristics exceeded a preset threshold. The system inputs this characteristic into an adaptive decision operator and outputs a data source return flow direction control signal pointing to edge thermal boundary reinforcement, volatile boundary reinforcement, and local sampling density reinforcement.

[0096] The system re-estimates the node residuals based on the supplementary data and maps the residuals into multiple independent physical error components, including systematic offset characteristics, time-varying drift evolution characteristics, equipment intrinsic disturbance characteristics, environmental decoupling disturbance characteristics, detection link disturbance characteristics, and mechanism structure mismatch characteristics.

[0097]

[0098] After reassessment, the error closure level met the preset closure condition, the predicted coverage level met the preset coverage requirement, and the trend-related physical disturbances were lower than the preset trend condition. The error contribution ranking showed that the environmental humidity disturbance at the heat treatment node and the liquid supply fluctuation at the coating node were the main sources of risk.

[0099] The probability of the terminal quality indicators simultaneously meeting the target film thickness, in-plane uniformity, and defect density specifications is calculated.

[0100]

[0101] When the changes in the overall loss measure, the probability of achieving the target, and the confidence metric all meet the preset convergence conditions, the system enters the propagation of fully distributed uncertainty.

[0102] The system inversely calculates the set of control variables that satisfy the probability of achieving the target, the error closure condition, and the industrial execution constraints.

[0103]

[0104] The set of control variables mentioned above constitutes the reliable process window for thin film processing, which includes constraints such as allowable failure rate, error closure condition, and predicted coverage condition.

[0105] In this embodiment, the industrialization process of the system output is shown in Table 1: Table 1. Comparison of Industrialization Verification Process for Improving Thin Film Uniformity

[0106] Through the above process, the improvement in film uniformity is no longer directly amplified from single-point parameters in the laboratory, but is transformed into a reliable process window for the target industrial line. The industrial execution end does not obtain a single optimal setpoint, but rather a table of process execution parameters including coating speed, liquid supply volume, heat treatment temperature, heat treatment time, ambient humidity range, detection frequency, and abnormal reflux conditions. When the bandgap rate increases in subsequent batches, the system can prioritize checking the stability of the liquid supply and the humidity control in the heat treatment section based on risk contribution, thereby supporting the transition of the results from pilot-scale verification to stable industrial operation.

[0107] Example 2: Transformation of high-solids functional slurry results into a continuous mixing-forming-curing production line In this embodiment, the technological achievement to be verified is a high-solids functional slurry and its curing process. This achievement exhibits high film strength and low porosity under laboratory conditions, but in continuous industrial production lines, it is easily affected by powder batches, mixing shear, pumping fluctuations, drying rates, and ambient humidity, leading to fluctuations in coating areal density, increased local porosity, and decreased adhesion. The target industrial process includes six process nodes: powder feeding, continuous mixing, pump coating, segmented drying, thermosetting, and online testing.

[0108] In step one, the system acquires the formulation mechanism, laboratory rheological data, curing kinetic data, powder particle size distribution, simulation samples, and target industrial production line data of the results to be verified. The target industrial production line data includes powder batch characteristics, mixing speed, mixing time, pump flow rate, coating line speed, drying section temperature, curing time, online areal density test results, porosity test results, and adhesion sampling test results. To avoid disclosing the actual formulation, all proportions and equipment settings in this embodiment are represented using normalized variables.

[0109] The system constructs state variables and control variables for a continuous production line. State variables may include solid volume fraction, apparent viscosity, agglomeration size, load per unit area, material temperature, and curing state; control variables may include mixing speed, mixing time, pump flow rate, coating line speed, drying temperature, and curing time.

[0110]

[0111] The system calibrates the laboratory rheological model with data from the continuous mixing equipment. Initially, the system found that the physical error component of the online surface density residual did not meet the preset closure condition. The system inputs the characteristics of the excessive error component into an adaptive decision operator, outputting backflow control signals for pumping stability, feed mass flow rate, and repeatable sample data source. After equipment condition reinforcement and sample reinforcement, the intensity of equipment-related disturbances decreased, the error closure improved, and the coverage met the preset requirements.

[0112] The probability of meeting the target specifications is determined by the probability that the areal density variation, porosity, and adhesion all meet the target specifications simultaneously.

[0113]

[0114] The target specifications include an upper limit for areal density variation, an upper limit for porosity, and a lower limit for adhesion. The threshold values ​​for each specification can be configured based on the material system, equipment family, and customer quality standards.

[0115] The system not only determines whether the terminal's compliance probability has reached the target, but also whether the reliable process window size is greater than the preset minimum executable window size, in order to avoid outputting a fragile optimal solution that only holds true under single-point conditions.

[0116]

[0117] When the window size is too small, the system does not directly output industrial execution parameters, but instead suggests reducing the upper limit of linear speed, enhancing pumping stability, or improving the environmental control capabilities of the drying section to expand the executable area.

[0118] The system outputs a normalized reliable process window and its engineering transformation results. Exemplary outputs include: normalized mixing speed range, normalized mixing time range, pump flow rate to linear velocity ratio range, first drying section temperature range, curing time range, batch allowable powder particle size range, and ambient humidity allowable range. The system also outputs a risk contribution ranking, showing that under the current equipment capacity, pump flow rate fluctuation, batch powder particle size drift, and drying section humidity disturbance are the top three risk factors.

[0119] In this embodiment, the industrialization process of the results achieved based on the above method is shown in Table 2: Table 2. Industrialization Implementation Flowchart for High Solids Functional Slurry Results

[0120] In this embodiment, laboratory results on high-solids functional slurries are transformed into a process window executable for continuous industrial production lines. This process window includes not only mixing, pumping, coating, drying, and curing parameters, but also powder batch acceptance conditions, environmental control conditions, online detection frequency, and abnormal reflux conditions. When batch deviations in powder particle size lead to increased porosity risk, the system outputs candidate reflux directions for incoming material re-inspection or drying window contraction; when fluctuations in pump flow rate lead to increased areal density out-of-band risk, the system outputs candidate reflux directions for pump link state reinforcement or flow closed-loop control enhancement. Thus, the scientific and technological achievements are transformed from laboratory parameters into a reliable process window that can be executed, monitored, and iterated on an industrial production line.

[0121] Example 3: like Figure 2 As shown, this invention also provides a reliable process window construction system for verifying the industrialization of scientific and technological achievements. This system includes a data receiving module, an industrial scene mapping module, a model calibration module, an error decomposition module, a reliability evaluation module, a dynamic backflow module, an iterative judgment module, a propagation solution module, a window output module, and an anomaly tracing module.

[0122] The data receiving module receives technical information and industrial data. The industrial scene mapping module generates state evolution paths, control channels, observation variables, boundary conditions, and interface uncertainties using multi-source heterogeneous elution operators, and assembles cross-node data via a standard interface card carrying state expectations and error covariance matrices. The model calibration module outputs the posterior distribution of parameters and node residuals. The error information extraction module performs explicit error information extraction on node residuals, decomposing at least one type of equipment disturbance component or environmental disturbance component, and can further extract multiple independent physical error components in the multidimensional physical evolution space. The credibility evaluation module performs at least one credibility quantification verification based on equipment disturbance components, environmental disturbance components, node residuals, or combinations thereof. The dynamic backflow module generates adaptive backflow direction control signals based on the characteristics of out-of-specification error components and the physical properties of process nodes. The iterative judgment module updates intrinsic parameters, independent physical error components, and control variables. The propagation solution module calculates the probability that the terminal quality indicators meet specifications. The window output module outputs a credible process window and a digital process parameter table. The anomaly tracing module is used to map the evolution path of the abnormal state to a multi-dimensional physical evolution space when abnormal quality indicators occur in continuous trial production or batch production. Based on the independent physical anomaly components and process topology, it outputs suspected failure nodes, suspected failure interfaces, dominant physical anomaly components and candidate maintenance or review actions.

[0123] This invention can also be implemented using an electronic device. The electronic device may include a processor, a memory, an input interface, an output interface, and a communication interface. The memory stores a computer program, which, when executed by the processor, implements the steps of the above-described method.

[0124] The present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described method.

[0125] The present invention also provides a computer program product, the computer program product comprising a computer program or computer instructions, wherein the computer program or computer instructions, when executed by a processor, implement the steps of the above method.

[0126] The key protection points of this invention are: mapping the technical information of the results to industrial field data into a computable state evolution path and interface uncertainty through a multi-source heterogeneous elution operator; realizing cross-node assembly through a standard specification interface card carrying state expectations and error covariance matrices; performing explicit error information extraction on node residuals to obtain at least equipment disturbance components or environmental disturbance components, and further mapping them to a multi-dimensional physical evolution space in subordinate embodiments; performing at least one reliable quantification verification based on error information to control whether the model enters the window inverse calculation; generating a backflow direction control signal bound to the physical properties of process nodes through an adaptive decision operator; inversely calculating the reliable process window through joint iterative judgment and propagation of fully distributed uncertainty; and performing reverse tracing of production line defects and anomalies based on explicit decomposition of industrial residuals when industrial operation is abnormal. The above protection points do not depend on specific threshold libraries, weight matrices, projection matrices, industry templates, customer equipment identities, or real process formulas.

[0127] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing a reliable process window for the industrialization verification of scientific and technological achievements, characterized in that, Includes the following steps: S1. Obtain the technical information of the scientific and technological achievements to be verified and the industrial data of the target industrial process. Transform the laboratory mechanism information and data characteristics carried in the technical information of the achievements into industrial scenario model information through multi-source heterogeneous elution operators. The industrial scenario model information includes state evolution path, control action channel, physical boundary conditions and interface uncertainty. S2. Based on industrial data, perform credibility-driven calibration on the intrinsic model and observation model in the industrial scenario model information to obtain the posterior distribution of parameters and the node residuals that characterize the mismatch between the model and the actual production line. S3. Perform explicit error information extraction on the node residuals, decompose them into equipment disturbance components and environmental disturbance components, and perform a reliable quantification verification based on at least one of the equipment disturbance components, environmental disturbance components and node residuals. If the trusted metric verification does not meet the preset conditions, execute S4; if the trusted metric verification meets the preset conditions, execute S5~S7. S4. Convert at least one result of the credible quantification verification into an over-limit index. Determine the dominant over-limit item based on the over-limit amount of each over-limit index relative to the preset upper limit threshold. Determine the input object to be reinforced in S1 or the model object to be reinforced in S2 based on the dominant over-limit item. Generate an adaptive backflow direction control signal. The adaptive backflow direction control signal is used to instruct the input object to be reinforced or the model object to be reinforced to be reinforced. Return the reinforced object to S1 or S2 to replace the corresponding original input or original model object. S5. Perform a joint iterative decision process to update intrinsic parameters, independent physical error components, and process control variables; S6. Based on the process topology of the target industrial process, perform full-distribution uncertainty propagation on the state evolution path, interface uncertainty, updated independent physical error components and updated process control variables, and calculate the probability that the terminal quality index falls into the process specification. S7. Inversely calculate the set of process control variables that make the probability of achieving the target satisfy the preset target probability, and output the set of process control variables as the credible process window of the scientific and technological achievement in the target industrial process.

2. The method for constructing a reliable process window for industrialization verification of scientific and technological achievements according to claim 1, characterized in that, In S1, the technical information and industrial data of the results are derived from multi-source heterogeneous datasets, which include laboratory sample test data, mechanism simulation data, pilot production line data, and on-site observation data of the target industrial line. The multi-source heterogeneous elution operator is used to extract the interface mapping relationship of each data source under different spatial scales and time periods, remove or replace the real equipment identity, real process formula and customer production identifier, and calculate the interface state covariance as the uncertainty input amplified across scales.

3. The method for constructing a reliable process window for industrialization verification of scientific and technological achievements according to claim 1, characterized in that, In S2, the industrial scenario model information is represented by the following state equations and observation equations: in, Indicates the first The state variables of each process node Indicates the first Control variables for each process node Represents the observed variable. Indicates intrinsic parameters, Indicates the observed parameters, and These represent process disturbances and observational disturbances, respectively.

4. The method for constructing a reliable process window for industrialization verification of scientific and technological achievements according to claim 1, characterized in that, In S2, when the industrial scenario model information flows across process nodes, a standard specification interface card carrying the state expectation and error covariance matrix is ​​used. The standard specification interface card includes at least three of the following: state estimate, state covariance, unit, reference benchmark, timestamp, process area identifier, batch identifier, equipment capability identifier, scale conservation conversion field, and data washing field. The equipment capability identifier is used to replace the real equipment identity information.

5. The method for constructing a reliable process window for industrialization verification of scientific and technological achievements according to claim 1, characterized in that, In S3, the explicit error information extraction for nodal residuals includes the following steps: The node residuals are mapped to a multidimensional physical evolution space jointly defined by the process node state basis, interface uncertainty covariance, and physical boundary topology. After physical error decoupling operator and functional consistency constraint, at least one of the following is obtained: equipment disturbance component, environmental disturbance component, time-varying drift component, detection link disturbance component, systematic offset component, and mechanism structure mismatch component. The components obtained from the decomposition are used to characterize the sources of physical mismatch when scientific and technological achievements are scaled up across target industrial processes.

6. The method for constructing a reliable process window for industrialization verification of scientific and technological achievements according to claim 1, characterized in that, In S3, the credible quantification verification includes at least one of error closure verification, prediction coverage consistency verification, equipment disturbance strength verification, environmental coupling disturbance verification, systematic offset verification, time series drift trend verification, and tail risk contribution verification; the threshold, weight, and judgment template of the credible quantification verification are configured according to industry, equipment family, material system, or process type.

7. The method for constructing a reliable process window for industrialization verification of scientific and technological achievements according to claim 6, characterized in that, The trusted quantification verification also includes: A multidimensional confidence metric tensor is constructed based on one or more error components. The multidimensional confidence metric tensor is compared with a preset confidence condition. When the multidimensional confidence metric tensor meets the preset confidence condition, the calibrated model is determined to proceed to S7.

8. The method for constructing a reliable process window for industrialization verification of scientific and technological achievements according to claim 1, characterized in that, In S5, the joint iterative determination process includes the following steps: Fix the error components obtained from the extraction and decomposition of the explicit error information and the process control variables, and update the intrinsic parameters and observation parameters; Fix the updated intrinsic parameters and observation parameters, and update the error components obtained from the decomposition. The updated intrinsic parameters, observed parameters, and error components are fixed, and the process control variables are updated accordingly. When the changes in the comprehensive failure measure, the probability of achieving the target, and the multidimensional credibility metric tensor satisfy the preset convergence conditions, the output is a stable and acceptable state.

9. The method for constructing a reliable process window for industrialization verification of scientific and technological achievements according to claim 8, characterized in that, The comprehensive loss measure is determined by a weighted combination of model consistency loss, error closure loss, credibility metric constraint loss, target achievement loss, and industry constraint loss. The weights of each item are configured according to the industry, equipment family, material system, or process type.

10. A reliable process window construction system for verifying the industrialization of scientific and technological achievements, characterized in that, include: The data receiving module is used to acquire the technical information of the scientific and technological achievements to be verified and the industrial data of the target industrial process. The industrial scene mapping module is used to transform the laboratory mechanism information and data characteristics carried in the technical information of the results into industrial scene model information containing state evolution path, control action channel, physical boundary conditions and interface uncertainty through multi-source heterogeneous elution operators. The model calibration module is used to calibrate the intrinsic model and observation model in the industrial scenario model information based on industrial data, and obtain the posterior distribution of parameters and the node residuals that characterize the mismatch between the model and the actual production line. The error information extraction module is used to perform explicit error information extraction on the node residuals, and at least decompose them into equipment disturbance components and environmental disturbance components. The credibility evaluation module is used to perform credibility quantification verification based on at least one of the device disturbance component, environmental disturbance component, and node residual; The dynamic backflow module is used to calculate the normalized over-limit of each verification index and determine the dominant over-limit item when the credibility metric verification does not meet the preset conditions. It generates an adaptive backflow direction control signal according to the preset mapping between the dominant over-limit item and the data source, device, detection, boundary condition or model. The reinforced dataset, device status, observation parameters, boundary conditions or intrinsic model are used as the update input of the corresponding module. The iterative determination module is used to perform a joint iterative determination process to update intrinsic parameters, independent physical error components and process control variables when the credibility quantification verification meets the preset conditions. The propagation and solution module is used to propagate the fully distributed uncertainty according to the process topology of the target industrial process and calculate the probability of the final quality index falling into the process specification. The window output module is used to inversely calculate the set of process control variables that make the probability of compliance meet the preset target probability and to determine the set of process control variables as the reliable process window.

Images