Method, device and equipment for constructing process prediction model with experience constraint and storage medium

Abstract

The application discloses a process prediction model construction method and device fusing experience constraints, equipment and a storage medium, and relates to the technical field of intelligent manufacturing. The method comprises the following steps: acquiring process parameters in a manufacturing process, quality indexes corresponding to the process parameters, and regular constraint relationships between the process parameters and the quality indexes; an initial process prediction model is constructed, a quality prediction value is obtained according to the process parameters and the initial process prediction model; a model prediction gradient is determined according to the quality prediction value and the process parameters, and an experience constraint loss term is determined according to the model prediction gradient and the regular constraint relationships; and the initial process prediction model is trained according to the experience constraint loss term, the process parameters and the quality indexes, so as to obtain a target process prediction model. The application improves the processing quality prediction precision and realizes the collaborative improvement of the consistency between process parameter optimization and experience rules.

Description

Technical Field

[0001] This application relates to the field of intelligent manufacturing technology, and in particular to a method, apparatus, equipment and storage medium for constructing process prediction models that integrate empirical constraints. Background Technology

[0002] In the manufacturing process, to achieve accurate prediction and control of processing quality, existing technologies have widely adopted purely data-driven models such as neural networks. These methods automatically learn the mapping relationship between process parameters and quality indicators by fitting sample data, thus improving the ability to describe nonlinear mapping relationships to a certain extent compared to traditional statistical modeling methods.

[0003] However, purely data-driven models rely entirely on the data distribution characteristics of the training samples. When the number of samples is limited or the distribution is unbalanced, the model's fitting ability in sparse regions drops sharply. More importantly, this type of method lacks the ability to recognize and constrain the inherent technological experience of the processing, and cannot integrate technological experience knowledge into the model's learning process. This results in the model's predictions reflecting only the statistical characteristics of the samples, rather than the actual physical changes. The predictions are prone to deviating from actual processing patterns, which is particularly prominent in scenarios involving extrapolation of data distribution, such as expanding the search range of process parameters or optimizing process parameters.

[0004] Therefore, the current solution has a technical problem in manufacturing processes where the prediction results of the data-driven model are prone to deviating from the actual processing rules. Summary of the Invention

[0005] The main objective of this application is to provide a method, apparatus, equipment, and storage medium for constructing a process prediction model that integrates empirical constraints, aiming to solve the technical problem that the prediction results of data-driven models are prone to deviate from the actual processing rules in process manufacturing.

[0006] To achieve the above objectives, this application proposes a method for constructing a process prediction model that integrates empirical constraints. The method includes: The process parameters in the manufacturing process, the quality indicators corresponding to the process parameters, and the regular constraint relationship between the process parameters and the quality indicators are obtained. An initial process prediction model is constructed, and the quality prediction value is obtained based on the process parameters and the initial process prediction model. The model prediction gradient is determined based on the predicted quality value and the process parameters, and the empirical constraint loss term is determined based on the relationship between the model prediction gradient and the rule constraint. The initial process prediction model is trained based on the empirical constraint loss term, the process parameters, and the quality index to obtain the target process prediction model.

[0007] In one embodiment, determining the model prediction gradient based on the predicted quality value and the process parameters, and determining the empirical constraint loss term based on the relationship between the model prediction gradient and the regular constraint, includes: The model prediction gradient is determined based on the predicted quality value and the process parameters. Based on the aforementioned constraint relationships, determine the empirical trend direction corresponding to the process parameters; Based on the gradient predicted by the model, the predicted change direction of the process parameters is determined. Based on the predicted direction of change and the empirical law direction, the empirical constraint loss term is determined.

[0008] In one embodiment, determining the empirical constraint loss term based on the predicted change direction and the empirical law direction includes: Based on the gradient predicted by the model, the predicted change magnitude of the process parameters is determined. When the predicted change direction is inconsistent with the empirical law direction, an empirical penalty value is determined based on the predicted change amplitude. Based on the empirical penalty value, determine the empirical constraint loss term.

[0009] In one embodiment, training the initial process prediction model based on the empirical constraint loss term, the process parameters, and the quality index to obtain the target process prediction model includes: Based on the predicted quality value and the quality index, determine the data prediction loss item; The total loss term is determined based on the empirical constraint loss term and the data prediction loss term. Using the total loss term as a constraint, the initial process prediction model is trained using the process parameters and the quality index to obtain the target process prediction model.

[0010] In one embodiment, after the step of training the initial process prediction model based on the empirical constraint loss term, the process parameters, and the quality index to obtain the target process prediction model, the method further includes: Construct a bi-objective optimization problem with the predicted quality value as the first optimization objective and the empirical constraint loss term as the second optimization objective; Based on the preset search strategy and the target process prediction model, the bi-objective optimization problem is solved to obtain the target process parameters.

[0011] In one embodiment, after the step of training the initial process prediction model based on the empirical constraint loss term, the process parameters, and the quality index to obtain the target process prediction model, the method further includes: The quality prediction value is used as the first explanatory objective, and the empirical constraint loss term is used as the second explanatory objective. Based on a preset attribution method, determine the first contribution degree of each process parameter to the first explanatory objective and the second contribution degree to the second explanatory objective; Based on the first contribution level and the second contribution level, the comprehensive contribution level of each process parameter is determined, and the process parameters are adjusted according to the comprehensive contribution level.

[0012] In one embodiment, constructing an initial process prediction model and obtaining a quality prediction value based on the process parameters and the initial process prediction model includes: Obtain an initial neural network structure and a preset activation function, wherein the initial neural network structure includes an input layer, at least one hidden layer, and an output layer; Based on the initial neural network structure and the activation function, an initial process prediction model is constructed; The process parameters are input into the input layer, the hidden layer, and the output layer of the initial process prediction model for forward propagation to obtain the quality prediction value.

[0013] Furthermore, to achieve the above objectives, this application also proposes a process prediction model construction apparatus that integrates empirical constraints, the apparatus comprising: The data acquisition module is used to acquire process parameters during the manufacturing process, quality indicators corresponding to the process parameters, and the regular constraint relationship between the process parameters and quality indicators. A prediction module is constructed to build an initial process prediction model and obtain the quality prediction value based on the process parameters and the initial process prediction model. The empirical constraint module is used to determine the model prediction gradient based on the predicted quality value and the process parameters, and to determine the empirical constraint loss term based on the relationship between the model prediction gradient and the rule constraint. The constraint training module is used to train the initial process prediction model based on the empirical constraint loss term, the process parameters, and the quality index to obtain the target process prediction model.

[0014] Furthermore, to achieve the above objectives, this application also proposes a process prediction model construction device that integrates empirical constraints. The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. The computer program is configured to implement the steps of the process prediction model construction method that integrates empirical constraints as described above.

[0015] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the process prediction model construction method with fused empirical constraints as described above.

[0016] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the process prediction model construction method with fused empirical constraints as described above.

[0017] One or more technical solutions proposed in this application have at least the following technical effects: This application transforms the obtained regular constraint relationship into an empirical constraint on the gradient direction of the model prediction, and uses this empirical constraint as one of the supervision signals to participate in the model training. This enables the model to not only fit the sample data, but also to make the trend of the predicted value with the change of process parameters consistent with the empirical regular direction, thereby reducing the degree of deviation of the prediction result from the actual processing law and improving the reliability of the prediction result. Attached Figure Description

[0018] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart illustrating an embodiment of the process prediction model construction method incorporating empirical constraints provided in this application; Figure 2 A schematic diagram of the data-experience hybrid driven prediction model structure provided in an embodiment of the process prediction model construction method integrating empirical constraints of this application; Figure 3 A flowchart of dual-objective interpretable analysis is provided for an embodiment of the process prediction model construction method that incorporates empirical constraints in this application. Figure 4 A flowchart illustrating Embodiment 2 of the process prediction model construction method incorporating empirical constraints in this application; Figure 5 A flowchart illustrating Embodiment 3 of the method for constructing a process prediction model that incorporates empirical constraints in this application; Figure 6A flowchart for optimizing process parameters is provided in an embodiment of the process prediction model construction method that incorporates empirical constraints in this application. Figure 7 A flowchart illustrating the modeling optimization process of an embodiment of the process prediction model construction method incorporating empirical constraints provided in this application. Figure 8 This is a schematic diagram of the module structure of the process prediction model construction device that integrates empirical constraints according to an embodiment of this application; Figure 9 This is a schematic diagram of the equipment structure of the hardware operating environment involved in the process prediction model construction method that incorporates empirical constraints in the embodiments of this application.

[0021] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0022] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0023] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0024] The main solution of this application embodiment is as follows: acquiring process parameters during the manufacturing process, quality indicators corresponding to the process parameters, and the regular constraint relationships between the process parameters and quality indicators; constructing an initial process prediction model; obtaining quality prediction values ​​based on the process parameters and the initial process prediction model; determining the model prediction gradient based on the quality prediction values ​​and the process parameters; and determining an empirical constraint loss term based on the model prediction gradient and the regular constraint relationships; training the initial process prediction model based on the empirical constraint loss term, the process parameters, and the quality indicators to obtain a target process prediction model.

[0025] Because existing pure data-driven models only aim to reduce the error between predicted and true values ​​during training, their learning process depends entirely on the distribution of sample data. When the number of samples is limited or the distribution is unbalanced, the model lacks the ability to recognize and constrain areas outside the data distribution, and the prediction results are prone to deviate from the actual processing rules, failing to provide stable and reliable prediction results.

[0026] This application provides a solution that acquires and utilizes existing regularity constraints in the processing process, transforming them into empirical constraints on the direction of the model's predicted gradient. This allows the model to be simultaneously driven by sample data and constrained by empirical regularity during training, thereby maintaining the trend of the predicted results consistent with the actual processing regularity while fitting the sample data. This effectively reduces the degree to which the predicted results deviate from the actual processing regularity and improves the reliability of the predicted results.

[0027] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or an electronic device or processor capable of performing the above functions. The following description uses a processor as an example to illustrate this embodiment and the subsequent embodiments. All actions involving the acquisition of signals, information, or data in this application are performed in accordance with the relevant data protection regulations of the locality and with authorization from the owner of the corresponding device.

[0028] Based on this, embodiments of this application provide a method for constructing a process prediction model that integrates empirical constraints, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the process prediction model construction method that incorporates empirical constraints according to this application.

[0029] In this embodiment, the method for constructing the process prediction model that integrates empirical constraints includes steps S10 to S40: Step S10: Obtain the process parameters in the manufacturing process, the quality indicators corresponding to the process parameters, and the regular constraint relationship between the process parameters and the quality indicators.

[0030] It should be noted that process parameters are adjustable processing variables that affect processing quality. Quality indicators are quantitative indicators used to evaluate processing quality, including one or more of the following: surface quality indicators, dimensional accuracy indicators, processing efficiency indicators, and material performance indicators. The regularity constraint relationship is the correspondence between the direction of change of each process parameter and the direction of change of the quality indicator.

[0031] Specifically, first, based on the specific processing procedure, identify the processing variables that affect the processing quality. There should be at least two of these processing variables, which are then used as process parameters. For example, in robotic belt grinding, abrasive grit size, contact force, belt linear speed, and feed rate are used as process parameters. Next, determine the type of index used to evaluate the processing quality, and use this index type as the quality index. For example, in robotic belt grinding, the surface roughness after machining is used as the quality index.

[0032] Secondly, process parameter data and corresponding processing quality inspection data are collected during the processing. During the collection process, for each combination of process parameters, the corresponding processing quality inspection results are collected, forming paired sample data. The collected data is preprocessed to remove outliers, such as deleting outliers that significantly deviate from the normal range. Missing data is compensated for, for example, by using interpolation or mean substitution to fill in missing values. The data is normalized to map process parameters of different dimensions to the same numerical range, thus eliminating the impact of dimensional differences on model training. The training set and validation set are divided according to a preset ratio.

[0033] Finally, process parameters and their corresponding quality indices are extracted from the preprocessed data to establish a data sample set relating process parameters and processing quality. Each sample in this set contains a combination of process parameters and its corresponding quality index value. For each process parameter, the direction of change between the two is determined based on the actual trend of the quality index change when the parameter changes during processing. If the quality index decreases when the parameter increases, then there is an inverse relationship between the parameter and the quality index; if the quality index increases when the parameter increases, then there is a positive relationship. This direction of change is used as a constraint relationship.

[0034] Understandably, since purely data-driven models rely solely on sample data for learning, they lack the ability to utilize inherent technological experience in the processing. When the sample size is limited or unevenly distributed, it's difficult to guarantee that the prediction results conform to actual processing patterns. This step, by acquiring process parameters, quality indicators, and the constraints between them, provides a data foundation for introducing empirical constraints in subsequent model training. This allows the model training process to utilize both sample data and empirical knowledge simultaneously, thereby improving the final model's ability to follow actual processing patterns.

[0035] Step S20: Construct an initial process prediction model, and obtain the quality prediction value based on the process parameters and the initial process prediction model.

[0036] It should be noted that the initial process prediction model is a neural network-based model used to establish the mapping relationship between process parameters and quality indicators. This model receives process parameters as input and outputs corresponding quality prediction values. The quality prediction values ​​are the prediction results of the initial process prediction model for the quality indicators corresponding to the current combination of input process parameters.

[0037] In one feasible implementation, step S20 may include: obtaining an initial neural network structure and a preset activation function, wherein the initial neural network structure includes an input layer, at least one hidden layer, and an output layer; constructing an initial process prediction model based on the initial neural network structure and the activation function; and inputting the process parameters into the input layer, the hidden layer, and the output layer of the initial process prediction model for forward propagation to obtain a quality prediction value.

[0038] It should be noted that the initial neural network structure is a feedforward fully connected neural network structure. A feedforward fully connected neural network is a network structure in which every neuron between adjacent layers is interconnected, and the signal propagates unidirectionally from the input layer to the output layer. The preset activation functions include hidden layer activation functions and output layer activation functions. The hidden layer activation functions are continuously differentiable activation functions, including the tanh (hyperbolic tangent) activation function or the sigmoid (S-shaped) activation function, and the output layer activation function is a linear activation function.

[0039] Specifically, the neural network structure parameters are set, including the number of nodes in the input layer, the number of hidden layers and the number of nodes in each layer, and the number of nodes in the output layer. The number of nodes in the input layer is set to the number of process parameters. One or more hidden layers are set, with the number of nodes in each hidden layer determined according to actual needs. The number of nodes in the output layer is set to the number of quality indicators. The activation function of the hidden layers is set to a continuously differentiable activation function, and the activation function of the output layer is set to a linear activation function. Based on the obtained neural network structure and activation functions, an initial process prediction model is constructed.

[0040] Representing process parameters in vector form:

[0041] in Let represent the i-th process parameter, and n be the number of process parameters.

[0042] The process parameters are input into the input layer of the initial process prediction model. The input layer passes the process parameters to the first hidden layer. The hidden layer performs a weighted summation of the received data, then applies a nonlinear transformation using an activation function, and passes the transformation result to the next layer. After layer-by-layer transformation through all hidden layers, the output layer performs a linear transformation on the output of the last hidden layer, outputting the predicted quality value y.

[0043] Where x is a process parameter, This is the initial process prediction model.

[0044] Understandably, since subsequent steps require calculating the gradient of the quality prediction value with respect to process parameters to determine whether the model's prediction direction conforms to empirical rules, and gradient calculation requires the model to be differentiable, this step employs a feedforward fully connected neural network structure and sets a continuously differentiable activation function to make the model differentiable, providing a model foundation for subsequent gradient calculation. Simultaneously, by inputting process parameters into the model and performing forward propagation to obtain the quality prediction value, a concrete function output object is provided for subsequent gradient calculation, allowing the partial derivatives of the quality prediction value with respect to process parameters to be quantified. If the model is not differentiable, the gradient cannot be determined subsequently, and thus it is impossible to determine whether the model's prediction direction is consistent with empirical rules. Therefore, this step is a prerequisite for the subsequent implementation of empirical constraints.

[0045] Step S30: Determine the model prediction gradient based on the predicted quality value and the process parameters, and determine the empirical constraint loss term based on the relationship between the model prediction gradient and the rule constraint.

[0046] It should be noted that the model prediction gradient is the partial derivative of the quality prediction value with respect to the process parameters, used to characterize the direction and rate of change of the quality prediction value when each process parameter changes. The empirical constraint loss term is a loss term used to measure the degree of deviation between the direction of the model prediction gradient and the direction limited by the regular constraint relationship.

[0047] Specifically, automatic differentiation techniques are used to calculate the partial derivatives of the predicted quality value with respect to each process parameter. These partial derivatives are then used as the model prediction gradient, with one model prediction gradient corresponding to each process parameter. The empirical gradient direction for each process parameter is determined based on the constraint relationships; the empirical gradient direction can be positive or negative. The model prediction gradient direction for each process parameter is compared with its corresponding empirical gradient direction. If the model prediction gradient direction matches the empirical gradient direction, the empirical penalty value for that process parameter is set to zero. If the model prediction gradient direction does not match the empirical gradient direction, the empirical penalty value for that process parameter is determined based on the magnitude of the model prediction gradient; a larger magnitude results in a larger empirical penalty value. The empirical penalty values ​​for all process parameters are then summed, and the sum of these empirical penalty values ​​is used as the empirical constraint loss term. .

[0048] Taking the robotic belt grinding process as an example, the process parameters include abrasive grit size AS, contact force CF, belt linear velocity BLS, and feed rate FR, while the quality index is surface roughness. Based on experience with belt abrasive grinding processes, the following rules and constraints apply: When AS increases Reduce, that is When CF increases Reduce, that is When BLS increases Reduce, that is When FR increases Increase, that is .

[0049] Therefore, the empirically constrained loss term for:

[0050] Understandably, since pure data-driven models only aim at data fitting and lack a constraint mechanism to determine whether the model's predictive behavior conforms to empirical rules, this step compares the model's predicted gradient direction with the empirical gradient direction. When they are inconsistent, a penalty value is determined based on the gradient magnitude. This transforms qualitative empirical rules into quantifiable constraint signals, providing empirical-level supervision information for subsequent model training. Specifically, directional consistency determines whether a penalty is triggered, and the magnitude determines the intensity of the penalty; together, they constitute a complete constraint judgment logic. If only direction judgment is available without magnitude quantification, all directional errors will be treated equally, making it impossible to distinguish the degree of deviation. If only magnitude is available without direction judgment, it is impossible to determine whether the model's prediction violates empirical rules. Therefore, this step, by combining the two dimensions of direction and magnitude, enables empirical constraints to accurately identify and penalize predictive behaviors that violate empirical rules.

[0051] Step S40: Train the initial process prediction model based on the empirical constraint loss term, the process parameters, and the quality index to obtain the target process prediction model.

[0052] It should be noted that the target process prediction model is a model obtained by training the initial process prediction model, and is used to output the corresponding quality prediction value based on the input process parameters.

[0053] In one feasible implementation, step S40 may include: determining a data prediction loss term based on the predicted quality value and the quality index; determining a total loss term based on the empirical constraint loss term and the data prediction loss term; and training the initial process prediction model using the process parameters and the quality index with the total loss term as a constraint to obtain a target process prediction model.

[0054] It should be noted that the data prediction loss term measures the difference between the predicted quality value and the quality metric. The larger the difference between the predicted quality value and the quality metric, the larger the value of the data prediction loss term; conversely, the smaller the difference, the smaller the value of the data prediction loss term. The total loss term is a combination of the data prediction loss term and the empirical constraint loss term, used to simultaneously constrain the model's fit to the sample data and its adherence to empirical rules during model training.

[0055] Specifically, the difference between the predicted quality value and the quality indicator is calculated, and the data prediction loss item is determined based on this difference. Predict the loss term from the data. With experience-constrained loss terms Perform a weighted summation to determine the total loss item. :

[0056] in This is the empirical constraint weighting coefficient, used to adjust the weight of the empirical constraint loss term in the total loss term.

[0057] Total loss item As the optimization objective, the gradient of the total loss term with respect to each trainable parameter in the initial process prediction model is calculated, and the values ​​of each trainable parameter are updated along the gradient descent direction using this gradient. The process of forward propagation, calculating the total loss, calculating the gradient, and updating parameters is repeated, using a set of process parameters and their corresponding quality indices in each iteration. In each iteration, the quality prediction value, the data prediction loss term, and the total loss term are recalculated, and the model parameters are updated accordingly. As the number of iterations increases, the value of the total loss term gradually decreases. Iteration stops when the value of the total loss term meets a preset condition, and the current model is output as the target process prediction model.

[0058] Understandably, training a model solely with the data prediction loss term can lead to predictions deviating from empirical patterns, while training solely with the empirical constraint loss term can result in the model failing to fit the sample data. This step combines the data prediction loss term and the empirical constraint loss term into a total loss term, allowing the model to be dually supervised by both sample data and empirical patterns during training. This ensures that the model fits the sample data while maintaining adherence to actual processing patterns. Without the data prediction loss term, the model cannot learn the mapping relationships in the sample data; without the empirical constraint loss term, the model lacks the ability to follow empirical patterns. Therefore, the combination of both is key to achieving a data-experience hybrid driving mechanism.

[0059] In another feasible implementation, after step S40, the method may further include: using the quality prediction value as a first explanatory target and the empirical constraint loss term as a second explanatory target; determining the first contribution degree of each process parameter to the first explanatory target and the second contribution degree to the second explanatory target according to a preset attribution method; determining the comprehensive contribution degree of each process parameter according to the first contribution degree and the second contribution degree, and adjusting the process parameters according to the comprehensive contribution degree.

[0060] It should be noted that the explanatory objectives include a primary explanatory objective and a secondary explanatory objective, which respectively reflect the impact of changes in each process parameter on the prediction results and the degree to which empirical rules are followed. The preset attribution method is the integral gradient method, used to attribute the model output to each input parameter. The contribution level represents the magnitude of the influence of each process parameter on the quality prediction value or the empirical constraint loss term; the larger the value, the more significant the influence of that process parameter on the objective.

[0061] Specifically, the quality prediction value F output by the target process prediction model is used as the primary explanatory target. The empirically constrained loss term is used. As a second explanatory objective, the integral gradient method is used to calculate the contribution of each process parameter to the predicted quality value. For the i-th process parameter... Starting from the baseline x' and ending at the current input x, the integral is performed along the path from the baseline to the current input. The integral of the partial derivative of the quality prediction value F with respect to the i-th process parameter along this path is calculated, and multiplied by the difference between the current value of the process parameter and the baseline value to obtain the integral gradient value of the process parameter with respect to the quality prediction value. :

[0062] in, This represents the i-th process parameter. Indicates the reference input. This represents the processing quality prediction output of a data-experience hybrid driven prediction model, i.e., a target process prediction model. This is the current input parameter vector. For the preset baseline vector, These are the continuous interpolation parameters from the baseline to the current input.

[0063] The integral gradient method is used to calculate the contribution of each process parameter to the empirical constraint loss term. For the i-th process parameter... Starting from the baseline x' and ending at the current input x, the integral gradient value of this process parameter with respect to the empirical constraint loss term. :

[0064] in, This represents the empirical constraint loss term in the data-experience hybrid driven prediction model. The calculated integral gradient value is used as the second contribution level. The absolute values ​​of the first and second contribution levels for each process parameter are added together to obtain the overall contribution level of that process parameter. The process parameters are then ranked according to their overall contribution levels, with priority given to adjusting those parameters having higher overall contribution levels.

[0065] Understandably, by calculating the contribution of each process parameter to the predicted quality value and the empirical constraint loss term separately using the integral gradient method, the overall impact of each process parameter on the two objectives can be obtained. If only the predicted quality value is provided without the contribution of each parameter, operators cannot determine which parameter adjustment will have the most significant effect; if only the contribution of a single objective is provided, the impact on empirical consistency may be overlooked. The fusion of the contribution levels of both objectives allows operators to prioritize adjusting the process parameters with the greatest impact based on the ranking of overall contribution levels, thereby optimizing the process parameters in a targeted manner.

[0066] This embodiment provides a method for constructing a process prediction model that integrates empirical constraints. By transforming empirical process rules into directional constraints on the model's prediction gradient, the model is simultaneously supervised by both sample data and empirical rules during training. This solves the technical problem that pure data-driven models, lacking empirical constraints, lead to prediction results that deviate from actual processing rules, thereby improving the reliability of the prediction results. Furthermore, by using the quality prediction value and the empirical constraint loss term as dual explanatory objectives, the integral gradient method is used to calculate the contribution of each process parameter to the two objectives and merge them into a comprehensive contribution. This achieves a quantifiable explanation of the model's prediction behavior and the effect of empirical constraints, providing a basis for the targeted optimization and adjustment of process parameters.

[0067] For example, to help understand the implementation process of the process prediction model construction method that integrates empirical constraints obtained in this embodiment in combination with the above embodiments, please refer to... Figure 2 , Figure 2 A data-experience hybrid-driven prediction model structure diagram is provided, which illustrates a method for constructing a process prediction model that integrates empirical constraints. Specifically: First, the process parameters are input into the data-experience hybrid driving prediction model's processing quality prediction module. The processing quality prediction module performs forward propagation calculations on the input process parameters and outputs the corresponding quality prediction values.

[0068] Secondly, based on the difference between the predicted quality values ​​and the quality indicators, the data prediction loss term is determined.

[0069] Simultaneously, process parameters and quality prediction values ​​are input into the automatic differential calculation module. The automatic differential technique is used to calculate the partial derivative of the quality prediction value with respect to each process parameter, which serves as the model prediction gradient. This model prediction gradient is then input into the empirical gradient constraint module. Simultaneously, the processing experience is transformed into empirical rule directions and input into the empirical gradient constraint module. The empirical gradient constraint module compares the direction of the model prediction gradient with the empirical rule direction. When the two directions are inconsistent, an empirical penalty value is determined based on the gradient magnitude. The empirical penalty values ​​corresponding to all process parameters are then aggregated to construct the empirical constraint loss term.

[0070] Then, the data prediction loss term and the empirical constraint loss term are input into the total loss function module. The total loss function module performs a weighted sum of the data prediction loss term and the empirical constraint loss term and outputs the total loss term.

[0071] Finally, the total loss term is input into the model parameter update module. The model parameter update module aims to minimize the total loss term by calculating its gradient with respect to each trainable parameter in the model and updating the values ​​of each trainable parameter along the gradient descent direction. This process is repeated until the total loss term meets a preset condition, at which point the iteration stops, and the processing quality prediction surrogate model, i.e., the target process prediction model, is output.

[0072] For example, to help understand the implementation process of the process prediction model construction method that integrates empirical constraints obtained in this embodiment in combination with the above embodiments, please refer to... Figure 3 , Figure 3 A dual-objective interpretable analysis flowchart is provided for a method of constructing a process prediction model that integrates empirical constraints. Specifically: First, obtain samples of process parameters to be interpreted, and input these samples into the trained data-experience hybrid prediction model.

[0073] Secondly, the data-experience hybrid driven prediction model performs forward propagation calculations on the input process parameter samples, outputting the corresponding quality prediction values ​​and empirical constraint loss terms. Based on the quality prediction values ​​and empirical constraint loss terms, the integral gradient contribution values ​​of each process parameter to the quality prediction values ​​and the integral gradient contribution values ​​of each process parameter to the empirical constraint loss terms are calculated.

[0074] Then, integral gradient analysis of the processing quality target and integral gradient analysis of the empirical consistency target were performed separately. In the integral gradient analysis of the processing quality target, the process parameters were ranked according to their integral gradient contribution to the quality prediction value, resulting in a quality contribution ranking of the process parameters. The larger the contribution value of the process parameter, the more significant its impact on the quality prediction result. In the integral gradient analysis of the empirical consistency target, the process parameters were ranked according to their integral gradient contribution to the empirical constraint loss term, resulting in an empirical consistency contribution ranking of the process parameters. The larger the contribution value of the process parameter, the more significant its impact on the degree to which the empirical constraint is satisfied.

[0075] Finally, the quality contribution ranking and empirical consistency contribution ranking of process parameters are comprehensively analyzed, and the results of the two types of contribution ranking are merged to obtain the comprehensive contribution degree of each process parameter to the two objectives. The bi-objective integral gradient interpretable analysis results are output to guide the directional adjustment of process parameters.

[0076] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to that in Embodiment 1 above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 4 Step S30 may include steps S31 to S34: Step S31: Determine the model prediction gradient based on the predicted quality value and the process parameters.

[0077] Specifically, automatic differentiation technology is used to calculate the predicted quality value y for each process parameter. partial derivatives The calculated partial derivatives are used as the model prediction gradients, with each process parameter corresponding to a model prediction gradient.

[0078] Understandably, the model prediction gradient is the basic data for judging whether the model's prediction behavior conforms to empirical rules. Only by determining the model prediction gradient can its direction be compared with the direction of empirical rules. Therefore, this step provides a calculation object for subsequent direction consistency judgment.

[0079] Step S32: Determine the empirical direction of the process parameters based on the aforementioned constraint relationship.

[0080] It should be noted that the direction of empirical rules refers to the direction of change that the quality indicators should exhibit when the process parameters change, as determined by experience in the processing technology.

[0081] Specifically, based on the correspondence between the changing directions of each process parameter and the changing directions of the quality index in the rule constraint relationship, the empirical rule direction corresponding to each process parameter is determined. The value for this direction is +1 or -1. If the rule constraint is that the quality index increases as the process parameter increases, then the empirical rule direction is... The value is +1; if the quality index decreases when this process parameter increases, then the empirical rule applies. The value is -1.

[0082] Understandably, empirical rules originate from long-term accumulated technological experience during processing and serve as a benchmark for judging whether the model's predicted behavior deviates from actual processing patterns. This step transforms qualitative empirical rules into quantitative directional symbols, enabling computers to recognize and calculate these rules.

[0083] Step S33: Based on the gradient predicted by the model, determine the predicted change direction of the process parameters.

[0084] It should be noted that the predicted direction of change is the actual direction of the quality prediction value as indicated by the model's predicted gradient as the process parameters change.

[0085] Specifically, the model prediction gradient for each process parameter The direction of predicted change is determined based on the sign of the gradient predicted by the model. If the model predicts a positive gradient, the predicted direction of change is positive, indicating that the predicted quality value increases as the process parameter increases; if the model predicts a negative gradient, the predicted direction of change is negative, indicating that the predicted quality value decreases as the process parameter increases.

[0086] Understandably, the model prediction gradient is calculated by automatic differentiation, and its sign directly reflects the direction of the model's prediction behavior under the current parameters. Therefore, this step decomposes the model prediction gradient into directional information, providing a model-side comparison object for subsequent directional comparison.

[0087] Step S34: Determine the empirical constraint loss term based on the predicted change direction and the empirical law direction.

[0088] In one feasible implementation, step S34 may include: determining the predicted change magnitude corresponding to the process parameter based on the model prediction gradient; determining an empirical penalty value based on the predicted change magnitude when the predicted change direction is inconsistent with the empirical law direction; and determining an empirical constraint loss term based on the empirical penalty value.

[0089] It should be noted that the predicted change magnitude is the absolute value of the model's predicted gradient, used to characterize how quickly the predicted quality value changes with process parameters. The empirical penalty value is a penalty amount determined based on the predicted change magnitude, used to penalize the model when the direction of the predicted change is inconsistent with the direction of empirical laws.

[0090] Specifically, the predicted direction of change for each process parameter. The empirical regularity direction corresponding to this process parameter Compare them. If the predicted direction of change is consistent with the direction of empirical patterns, that is... and If they are the same, then the empirical penalty value corresponding to that process parameter is set to zero. If the predicted direction of change is inconsistent with the direction of empirical law, i.e. and Conversely, the model prediction gradient corresponding to the process parameter is obtained. The absolute value of the model's predicted gradient | The predicted change amplitude is determined, and an empirical penalty value is calculated based on this amplitude; the larger the amplitude, the larger the empirical penalty value. The empirical penalty values ​​corresponding to all process parameters are then summarized, and the sum of these empirical penalty values ​​is used as the empirical constraint loss term. The formula for calculating the empirical constraint loss term is:

[0091] in, Represents the mathematical expectation on the training samples. This represents the empirical gradient direction corresponding to the i-th process parameter, and its value is set to 1 or -1 according to empirical rules. This represents the i-th input process parameter.

[0092] when When, it indicates that the predicted direction of change is consistent with the direction of empirical laws, max(0, A value of 0 results in no penalty; when... When, it indicates that the predicted direction of change is inconsistent with the direction of empirical laws, max(0, The value can be... It itself generates a penalty, the penalty value of which varies. It increases as the amplitude increases.

[0093] Taking laser processing as an example, the process parameters include laser power P, scanning speed V, defocusing amount D, and scanning spacing H, while the quality index is the molten pool width W. Based on experience in laser processing, the following constraints apply: As P increases, W increases, that is... When V increases, W decreases, that is... When D increases, W decreases, that is... When H increases, W decreases, that is... .

[0094] Therefore, the empirically constrained loss term for:

[0095] Understandably, this step determines whether to trigger a penalty based on directional consistency and the intensity of the penalty based on the magnitude of the magnitude; together, these two factors constitute a complete constraint judgment logic. If there were only a direction judgment without magnitude quantification, all directional errors would be treated equally, making it impossible to distinguish the degree of deviation; if there were only magnitude without direction judgment, it would be impossible to determine whether the model prediction violates empirical rules. Therefore, this step, by combining the two dimensions of direction and magnitude, enables empirical constraints to accurately identify and penalize predictive behaviors that violate empirical rules.

[0096] This embodiment provides a method for constructing a process prediction model that integrates empirical constraints. By splitting the model's predicted gradient into two dimensions—the predicted direction of change and the predicted magnitude of change—an empirical penalty value is determined based on the predicted magnitude of change when the predicted direction of change is inconsistent with the empirical law direction. This solves the problem that the direction consistency judgment cannot distinguish the degree of deviation, and realizes an empirical constraint mechanism that applies differentiated penalties based on the degree of deviation, further improving the model's ability to follow empirical laws.

[0097] Based on the first and second embodiments of this application, in this third embodiment, the content that is the same as or similar to that in embodiments one and two above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 5 After step S40, steps S50 to S51 may also be included: Step S50: Construct a bi-objective optimization problem with the quality prediction value as the first optimization objective and the empirical constraint loss term as the second optimization objective.

[0098] It should be noted that the bi-objective optimization problem is an optimization problem that simultaneously optimizes two objectives. The first optimization objective is the quality prediction value output by the target process prediction model, and the second optimization objective is the empirical constraint loss term. The first optimization objective is used to optimize the processing quality, and the second optimization objective is used to optimize the degree to which the model's predictive behavior follows empirical rules.

[0099] Specifically, the first optimization objective is to optimize the quality prediction value, setting the optimization direction to minimize or maximize the quality prediction value, determined according to actual processing requirements. The second optimization objective is to optimize the empirical constraint loss term, setting the optimization direction to minimize the empirical constraint loss term to reduce the degree to which the model's predicted behavior deviates from empirical patterns. Based on the first and second optimization objectives, a bi-objective optimization problem is constructed.

[0100] Understandably, since existing optimization methods usually only take the processing quality index as the single optimization objective and ignore the dimension of whether the model's predicted behavior conforms to empirical rules, this step takes the quality prediction value and the empirical constraint loss term as optimization objectives at the same time. This makes the optimization process not only focus on improving the processing quality, but also on the model's predicted behavior following empirical rules, providing an optimization framework for obtaining a combination of process parameters that takes into account both objectives in the subsequent solution process.

[0101] Step S51: Solve the bi-objective optimization problem according to the preset search strategy and the target process prediction model to obtain the target process parameters.

[0102] It should be noted that the preset search strategy is an enhanced elite-preserving genetic algorithm, which searches for the optimal solution in the parameter space by simulating selection, crossover, and mutation operations in biological evolution. The target process prediction model is used to evaluate the quality prediction values ​​and empirical constraint loss terms corresponding to the candidate process parameter combinations, serving as the basis for fitness evaluation during the search process.

[0103] Specifically, the population is initialized, generating multiple candidate process parameter combinations, each candidate combination representing an individual. Each candidate process parameter combination is input into the target process prediction model, which outputs the corresponding quality prediction value and empirical constraint loss term. The quality prediction value and empirical constraint loss term are used together as the fitness value. During iteration, individuals with higher fitness are selected as parents. Crossover operations are performed on the parent individuals to generate offspring individuals, and mutation operations are performed on the offspring individuals to introduce new genetic information. The individuals with the highest fitness are directly retained in the next generation to prevent the optimal solution from being lost during iteration. The diversity index of the current population is calculated, and the crossover and mutation probabilities are dynamically adjusted based on the diversity index. The adjustment formulas for the crossover and mutation probabilities are:

[0104] in, and Let be the crossover probability and mutation probability of generation t, respectively. The population diversity index for generation t. and The sensitivity coefficient is adjusted for probability. A new generation of population is generated through the above selection, crossover, mutation, and elite retention operations. The above iterative process is repeated until the preset termination condition is met, at which point the iteration stops, and the individual with the best fitness is output as the target process parameter.

[0105] Understandably, since the solution space of a bi-objective optimization problem is usually large and contains multiple local optima, this step employs an enhanced elite-preserving genetic algorithm for global search. The elite-preserving strategy prevents the loss of optimal solutions, and the population diversity is maintained by dynamically adjusting crossover and mutation probabilities to avoid the algorithm getting trapped in local optima prematurely. The target process prediction model serves as a fitness evaluation model during the solution process, providing quantitative assessments of quality predictions and empirical constraint loss terms. This allows the search process to simultaneously consider both optimization objectives, ultimately obtaining the optimal combination of process parameters that satisfies both processing quality requirements and empirical constraints.

[0106] This embodiment provides a method for constructing a process prediction model that integrates empirical constraints. By constructing a bi-objective optimization problem with quality prediction as the first optimization objective and empirical constraint loss as the second optimization objective, and using an enhanced elite-preserving genetic algorithm to solve the problem with the objective process prediction model as the fitness evaluation model, this method solves the problem that existing optimization methods only focus on a single quality objective and ignore empirical consistency. It achieves synergistic optimization between improving processing quality and maintaining empirical patterns, and further improves the reliability of process parameter optimization results.

[0107] For example, to help understand the implementation process of the process prediction model construction method that integrates empirical constraints obtained in this embodiment in combination with the above embodiments, please refer to... Figure 6 , Figure 6 A flowchart for process parameter optimization is provided for a method of constructing a process prediction model that integrates empirical constraints. Specifically: First, initialize the population by randomly generating multiple candidate process parameter combinations within the range of process parameter values. Each candidate combination is treated as an individual, and each individual contains a complete set of process parameters.

[0108] Secondly, each candidate process parameter combination is input into the trained processing quality prediction surrogate model. The surrogate model outputs the corresponding quality prediction value and empirical constraint loss term, which are then used together as the fitness value of that individual. The fitness value is used to evaluate the quality of each individual.

[0109] Then, an enhanced elite retention strategy is adopted to directly retain the individuals with the highest fitness values ​​to the next generation of the population, preventing the current optimal solution from being lost in subsequent iterations.

[0110] Next, the diversity index of the current population is calculated. This index measures the degree of difference between individuals in the population, and the crossover and mutation probabilities are dynamically adjusted based on the diversity index. When population diversity decreases, the crossover probability decreases and the mutation probability increases to prevent the population from falling into a local optimum prematurely. When population diversity is high, the crossover probability increases and the mutation probability decreases to promote the combination and preservation of superior genes.

[0111] Next, based on the adjusted crossover and mutation probabilities, selection, crossover, and mutation operations are performed on individuals in the population. The selection operation is used to select individuals with higher fitness from the current population as parents, the crossover operation is used to exchange partial genetic information between two parent individuals to generate offspring individuals, and the mutation operation is used to randomly perturb some genes of offspring individuals to introduce new genetic information.

[0112] Afterwards, the offspring individuals generated through selection, crossover, and mutation operations will be combined with the elite individuals to form a new generation of population.

[0113] Finally, determine whether the current iteration meets the preset termination condition. If the termination condition is not met, return to the "Generate Process Parameter Combination Individual" step to continue the next iteration; if the termination condition is met, stop the iteration, select the individual with the highest fitness value from the current population, and output the process parameter combination contained in that individual as the target process parameter.

[0114] For example, to help understand the implementation process of the process prediction model construction method that integrates empirical constraints obtained in this embodiment in combination with the above embodiments, please refer to... Figure 7 , Figure 7 A modeling optimization explanation flowchart is provided for a method of constructing a process prediction model that integrates empirical constraints. Specifically: First, process parameter samples, corresponding processing quality indicators, and empirical constraints are obtained and used as input data. Based on the above input data, a data-experience hybrid-driven prediction model construction process is executed to obtain a trained data-experience hybrid-driven prediction model, which serves as a surrogate model.

[0115] Secondly, based on the surrogate model, a bi-objective integral gradient interpretable analysis is performed to calculate the contribution of each process parameter to the predicted quality value and the empirical constraint loss term, thus obtaining the evaluation results of the process parameter contribution. Simultaneously, based on the surrogate model, an enhanced elite-preserving genetic algorithm is executed for optimization to obtain the recommended combination of process parameters.

[0116] Next, it is determined whether the processing quality corresponding to the recommended combination of process parameters meets the requirements. If the quality is unsatisfactory, the process parameter contribution evaluation step is returned to, and the optimization solution is re-solved based on the contribution evaluation results, adjusting the combination of process parameters. If the quality meets the requirements, verification and repeatability experiments are conducted on the recommended combination of process parameters.

[0117] Finally, the processing quality of this combination of process parameters was verified through experiments. The results remained stable in multiple repeated experiments, and the recommended combination of process parameters was output as the final target process parameters.

[0118] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the method for constructing a process prediction model that incorporates empirical constraints. Any simple modifications based on this technical concept are within the scope of protection of this application.

[0119] This application also provides a process prediction model construction device that integrates empirical constraints; please refer to [reference needed]. Figure 8 The apparatus for constructing a process prediction model that integrates empirical constraints includes: Data acquisition module 10 is used to acquire process parameters in the manufacturing process, quality indicators corresponding to the process parameters, and the regular constraint relationship between the process parameters and quality indicators; A prediction module 20 is used to construct an initial process prediction model and obtain a quality prediction value based on the process parameters and the initial process prediction model. The empirical constraint module 30 is used to determine the model prediction gradient based on the quality prediction value and the process parameters, and to determine the empirical constraint loss term based on the relationship between the model prediction gradient and the rule constraint. The constraint training module 40 is used to train the initial process prediction model based on the empirical constraint loss term, the process parameters, and the quality index to obtain the target process prediction model.

[0120] The process prediction model construction apparatus integrating empirical constraints provided in this application, employing the process prediction model construction method integrating empirical constraints in the above embodiments, can solve the technical problem that the prediction results of data-driven models easily deviate from the actual processing rules in manufacturing processes. Compared with the prior art, the beneficial effects of the process prediction model construction apparatus integrating empirical constraints provided in this application are the same as the beneficial effects of the process prediction model construction method integrating empirical constraints provided in the above embodiments, and other technical features in the process prediction model construction apparatus integrating empirical constraints are the same as the features disclosed in the methods of the above embodiments, and will not be repeated here.

[0121] In one embodiment, the empirical constraint module 30 is further configured to determine the model prediction gradient based on the predicted quality value and the process parameters; Based on the aforementioned constraint relationships, determine the empirical trend direction corresponding to the process parameters; Based on the gradient predicted by the model, the predicted change direction of the process parameters is determined. Based on the predicted direction of change and the empirical law direction, the empirical constraint loss term is determined.

[0122] In one embodiment, the empirical constraint module 30 is further configured to determine the predicted change magnitude of the process parameters based on the model prediction gradient; When the predicted change direction is inconsistent with the empirical law direction, an empirical penalty value is determined based on the predicted change amplitude. Based on the empirical penalty value, determine the empirical constraint loss term.

[0123] In one embodiment, the constraint training module 40 is further configured to determine a data prediction loss term based on the quality prediction value and the quality index; The total loss term is determined based on the empirical constraint loss term and the data prediction loss term. Using the total loss term as a constraint, the initial process prediction model is trained using the process parameters and the quality index to obtain the target process prediction model.

[0124] In one embodiment, the constraint training module 40 is further configured to construct a bi-objective optimization problem with the quality prediction value as the first optimization objective and the empirical constraint loss term as the second optimization objective; Based on the preset search strategy and the target process prediction model, the bi-objective optimization problem is solved to obtain the target process parameters.

[0125] In one embodiment, the constraint training module 40 is further configured to use the quality prediction value as a first explanatory target and the empirical constraint loss term as a second explanatory target; Based on a preset attribution method, determine the first degree of contribution of each process parameter to the first explanatory objective and the second degree of contribution to the second explanatory objective; Based on the first contribution level and the second contribution level, the comprehensive contribution level of each process parameter is determined, and the process parameters are adjusted according to the comprehensive contribution level.

[0126] In one embodiment, the prediction construction module 20 is further configured to obtain an initial neural network structure and a preset activation function, wherein the initial neural network structure includes an input layer, at least one hidden layer, and an output layer; Based on the initial neural network structure and the activation function, an initial process prediction model is constructed; The process parameters are input into the input layer, the hidden layer, and the output layer of the initial process prediction model for forward propagation to obtain the quality prediction value.

[0127] This application provides a process prediction model construction device that integrates empirical constraints. The process prediction model construction device that integrates empirical constraints includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the process prediction model construction method that integrates empirical constraints in the above embodiment 1.

[0128] The following is for reference. Figure 9 This document illustrates a structural schematic diagram of a process prediction model construction device suitable for implementing the fusion of empirical constraints in the embodiments of this application. The fusion of empirical constraints in the process prediction model construction device in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 9 The illustrated process prediction model building device that incorporates empirical constraints is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0129] like Figure 9As shown, the process prediction model building apparatus incorporating empirical constraints may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.) that can perform various appropriate actions and processes based on a program stored in ROM (Read Only Memory) 1002 or a program loaded from storage device 1003 into RAM (Random Access Memory) 1004. RAM 1004 also stores various programs and data required for the operation of the process prediction model building apparatus incorporating empirical constraints. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via bus 1005. Input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, LCDs (Liquid Crystal Displays), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the process prediction model building device incorporating empirical constraints to communicate wirelessly or wiredly with other devices to exchange data. Although a process prediction model building device incorporating empirical constraints with various systems is shown in the figure, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems may be implemented alternatively.

[0130] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0131] The process prediction model construction device integrating empirical constraints provided in this application, employing the process prediction model construction method integrating empirical constraints described in the above embodiments, can solve the technical problem that the prediction results of data-driven models easily deviate from actual processing rules in manufacturing processes. Compared with the prior art, the beneficial effects of the process prediction model construction device integrating empirical constraints provided in this application are the same as the beneficial effects of the process prediction model construction method integrating empirical constraints provided in the above embodiments, and other technical features in this process prediction model construction device integrating empirical constraints are the same as those disclosed in the method of the previous embodiment, and will not be repeated here.

[0132] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0133] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0134] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the process prediction model construction method incorporating empirical constraints in the above embodiments.

[0135] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, RAM (Random Access Memory), ROM (Read Only Memory), EPROM (Erasable Programmable Read Only Memory or Flash Memory), optical fibers, CD-ROM (CD-Read Only Memory), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0136] The aforementioned computer-readable storage medium may be included in the process prediction model building device that incorporates empirical constraints; or it may exist independently and not be assembled into the process prediction model building device that incorporates empirical constraints.

[0137] The aforementioned computer-readable storage medium carries one or more programs. When these programs are executed by a process prediction model construction device that integrates empirical constraints, the process prediction model construction device integrates empirical constraints to: acquire process parameters in the manufacturing process, quality indicators corresponding to the process parameters, and regular constraint relationships between the process parameters and quality indicators; construct an initial process prediction model; obtain quality prediction values ​​based on the process parameters and the initial process prediction model; determine the model prediction gradient based on the quality prediction values ​​and the process parameters; and determine an empirical constraint loss term based on the model prediction gradient and the regular constraint relationships; and train the initial process prediction model based on the empirical constraint loss term, the process parameters, and the quality indicators to obtain a target process prediction model.

[0138] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including LAN (Local Area Network) or WAN (Wide Area Network)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0139] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0140] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0141] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described method for constructing a process prediction model that incorporates empirical constraints. This addresses the technical problem that, in manufacturing processes, the prediction results of data-driven models easily deviate from actual processing patterns. Compared to the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the process prediction model construction method that incorporates empirical constraints provided in the above embodiments, and will not be elaborated upon here.

[0142] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described method for constructing a process prediction model that incorporates empirical constraints.

[0143] The computer program product provided in this application can solve the technical problem that the prediction results of data-driven models tend to deviate from the actual processing rules in manufacturing processes. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the process prediction model construction method that integrates empirical constraints provided in the above embodiments, and will not be repeated here.

[0144] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A method for constructing a process prediction model that integrates empirical constraints, characterized in that, The method includes: The process parameters in the manufacturing process, the quality indicators corresponding to the process parameters, and the regular constraint relationship between the process parameters and the quality indicators are obtained. An initial process prediction model is constructed, and the quality prediction value is obtained based on the process parameters and the initial process prediction model. The model prediction gradient is determined based on the predicted quality value and the process parameters, and the empirical constraint loss term is determined based on the relationship between the model prediction gradient and the rule constraint. The initial process prediction model is trained based on the empirical constraint loss term, the process parameters, and the quality index to obtain the target process prediction model.

2. The method as described in claim 1, characterized in that, The step of determining the model prediction gradient based on the predicted quality value and the process parameters, and determining the empirical constraint loss term based on the relationship between the model prediction gradient and the regular constraint, includes: The model prediction gradient is determined based on the predicted quality value and the process parameters. Based on the aforementioned constraint relationships, determine the empirical trend direction corresponding to the process parameters; Based on the gradient predicted by the model, the predicted change direction of the process parameters is determined. Based on the predicted direction of change and the empirical law direction, the empirical constraint loss term is determined.

3. The method as described in claim 2, characterized in that, The step of determining the empirical constraint loss term based on the predicted change direction and the empirical law direction includes: Based on the gradient predicted by the model, the predicted change magnitude of the process parameters is determined. When the predicted change direction is inconsistent with the empirical law direction, an empirical penalty value is determined based on the predicted change amplitude. Based on the empirical penalty value, determine the empirical constraint loss term.

4. The method as described in claim 1, characterized in that, The step of training the initial process prediction model based on the empirical constraint loss term, the process parameters, and the quality index to obtain the target process prediction model includes: Based on the predicted quality value and the quality index, determine the data prediction loss item; The total loss term is determined based on the empirical constraint loss term and the data prediction loss term. Using the total loss term as a constraint, the initial process prediction model is trained using the process parameters and the quality index to obtain the target process prediction model.

5. The method as described in claim 1, characterized in that, After the step of training the initial process prediction model based on the empirical constraint loss term, the process parameters, and the quality index to obtain the target process prediction model, the method further includes: Construct a bi-objective optimization problem with the predicted quality value as the first optimization objective and the empirical constraint loss term as the second optimization objective; Based on the preset search strategy and the target process prediction model, the bi-objective optimization problem is solved to obtain the target process parameters.

6. The method as described in claim 1, characterized in that, After the step of training the initial process prediction model based on the empirical constraint loss term, the process parameters, and the quality index to obtain the target process prediction model, the method further includes: The quality prediction value is used as the first explanatory objective, and the empirical constraint loss term is used as the second explanatory objective. Based on a preset attribution method, determine the first degree of contribution of each process parameter to the first explanatory objective and the second degree of contribution to the second explanatory objective; Based on the first contribution level and the second contribution level, the comprehensive contribution level of each process parameter is determined, and the process parameters are adjusted according to the comprehensive contribution level.

7. The method as described in claim 1, characterized in that, The construction of the initial process prediction model, and the obtaining of quality prediction values ​​based on the process parameters and the initial process prediction model, includes: Obtain an initial neural network structure and a preset activation function, wherein the initial neural network structure includes an input layer, at least one hidden layer, and an output layer; Based on the initial neural network structure and the activation function, an initial process prediction model is constructed; The process parameters are forward-propagated through the input layer, the hidden layer, and the output layer of the initial process prediction model to obtain the quality prediction value.

8. A device for constructing a process prediction model that integrates empirical constraints, characterized in that, The device includes: The data acquisition module is used to acquire process parameters during the manufacturing process, quality indicators corresponding to the process parameters, and the regular constraint relationship between the process parameters and quality indicators. A prediction module is constructed to build an initial process prediction model and obtain the quality prediction value based on the process parameters and the initial process prediction model. The empirical constraint module is used to determine the model prediction gradient based on the predicted quality value and the process parameters, and to determine the empirical constraint loss term based on the relationship between the model prediction gradient and the rule constraint. The constraint training module is used to train the initial process prediction model based on the empirical constraint loss term, the process parameters, and the quality index to obtain the target process prediction model.

9. A process prediction model construction device that integrates empirical constraints, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the method for constructing a process prediction model with fused empirical constraints as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the process prediction model construction method with fused empirical constraints as described in any one of claims 1 to 7.

Images