Intelligent control method and device for defects in thermosetting resin forming and manufacturing process

Abstract

The invention relates to the field of artificial intelligence, and provides an intelligent defect management and control method and device in a thermosetting resin forming and manufacturing process. The method comprises the following steps: acquiring multi-source heterogeneous data in a thermosetting resin forming and manufacturing process, and preprocessing the multi-source heterogeneous data; constructing a defect prediction AI model fusing a trend estimation module, a self-supervised model updating mechanism and a quantum-reinforcement learning optimization framework; the preprocessed multi-source heterogeneous data are input into the defect prediction AI model, and the predicted defect occurrence probability and severity in the thermosetting resin forming and manufacturing process are output; formulating a process parameter optimization or formula structure adjustment strategy based on the predicted defect occurrence probability and severity; and performing real-time regulation and control on the thermosetting resin forming and manufacturing process according to the technological parameter optimization or formula structure adjustment strategy. In this way, efficient defect prediction and intelligent regulation and control in the thermosetting resin forming and manufacturing process are supported.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the field of artificial intelligence, and in particular, to a method and device for intelligent control of defects in a thermosetting resin forming manufacturing process. BACKGROUND

[0002] Urea-formaldehyde resin, as a common thermosetting resin, is widely used in the fields of artificial board and adhesive. However, its production process highly depends on the accurate control of multiple process variables such as raw material ratio, reaction temperature, pH value, etc. A slight deviation may cause product quality defects such as insufficient bonding strength, excessive free formaldehyde, incomplete curing, etc. Traditional quality control methods mainly rely on manual experience and post-detection, which have problems such as response lag, untimely control, and difficulty in adapting to complex manufacturing environments, and are difficult to meet the requirements of real-time, accuracy and self-adaptive ability in modern manufacturing.

[0003] With the development of artificial intelligence (AI) technology, using machine learning, deep learning and other methods to model and analyze manufacturing process data has become an important means to improve the intelligent level of manufacturing. However, manufacturing data often has non-stationary, trend drift and lack of labeling problems, and direct application of conventional models often leads to unstable prediction performance. In addition, in the face of dynamic and complex production environment, relying only on static models cannot achieve efficient and sustainable process control. Therefore, there is an urgent need for a comprehensive method that integrates trend modeling, self-supervised learning and intelligent optimization mechanism to support efficient prediction and intelligent control of urea-formaldehyde resin forming manufacturing process. SUMMARY

[0004] The present disclosure provides a method and device for intelligent control of defects in a thermosetting resin forming manufacturing process, aiming to break through the bottlenecks of traditional AI models in non-stationary data processing, abnormal interference suppression, global parameter optimization, etc., and realize efficient prediction and intelligent control of defects in thermosetting resin forming manufacturing process.

[0005] According to a first aspect of the present disclosure, a method for intelligent control of defects in a thermosetting resin forming manufacturing process is provided, comprising:

[0006] Obtaining multi-source heterogeneous data in a thermosetting resin forming manufacturing process, and preprocessing the multi-source heterogeneous data;

[0007] Constructing a defect prediction AI model that integrates a trend estimation module, a self-supervised model updating mechanism and a quantum-reinforcement learning optimization framework, wherein the trend estimation module separates the trend component and the dynamic component in the data, the self-supervised mechanism optimizes the model parameters in the unlabeled scenario, and the quantum neural network and reinforcement learning are combined to realize efficient search of global parameters;

[0008] Input the pre-processed multi-source heterogeneous data into the defect prediction AI model, and output the predicted defect occurrence probability and its severity in the thermosetting resin molding manufacturing process;

[0009] Based on the predicted defect occurrence probability and its severity, develop a process parameter optimization or formulation structure adjustment strategy;

[0010] According to the process parameter optimization or formulation structure adjustment strategy, real-time control is performed on the thermosetting resin molding manufacturing process.

[0011] As a preferred embodiment, the pre-processing of the multi-source heterogeneous data comprises:

[0012] Anomaly data is identified and removed or corrected using statistical methods and machine learning methods;

[0013] According to the characteristics of missing data, linear interpolation, time series backfilling or similar sample based filling strategy is used to backfill the missing data samples;

[0014] The data samples after backfilling are standardized and normalized by Z-score standardization or Min-Max normalization method;

[0015] Align the data of different frequencies or different time stamps, construct a time sliding window according to the prediction task requirements, and reconstruct the input feature sequence.

[0016] As a preferred embodiment, the construction method of the trend estimation module is:

[0017] Calculate the mean value in the current window according to the sliding window method;

[0018] Based on the trend estimation value of the previous window and the mean value in the current window, update the trend estimation value at the current time according to the exponential moving average method;

[0019] Subtract the updated trend estimation value at the current time from the original feature corresponding to the current time to obtain the detrended feature sequence.

[0020] As a preferred embodiment, the self-supervised model update mechanism comprises:

[0021] Filter the data samples predicted as normal by the defect prediction AI model through the mask vector mechanism;

[0022] Calculate the reconstruction loss of the normal sample data using the loss function, and optimize the model parameters through online gradient descent, wherein the loss function is:

[0023] ;

[0024] wherein, is a mask vector for activating normal samples; is a reconstruction error for measuring the expressive ability of the model on the current feature; is the input de-trended feature; is the reconstructed feature output by the model; is the model predicted label, = 0 indicates normal, = 1 indicates abnormal.

[0025] As a preferred embodiment, the quantum-reinforcement learning optimization framework construction method is:

[0026] A multi-layer quantum gate circuit is used to construct a quantum neural network to generate an exponential level of quantum measurement probability;

[0027] A parameter mapping model is introduced to map the quantum measurement probability to the policy network parameters;

[0028] A standard policy gradient loss function is used to optimize the behavior selection of the policy network;

[0029] The mapping process is:

[0030] ;

[0031] Wherein, is a quantum state defined by the quantum parameters ; is the i-th parameter of the policy network (continuing from the parameters trained in step 2.1 ); is the total number of parameters; is the ground state bit string encoding; is the probability of measuring state ; is a mapping function, is a learnable mapping model parameter.

[0032] As a preferred embodiment, after the behavior selection of the policy network is optimized using the standard policy gradient loss function, it further includes:

[0033] The reward signal contained in the policy gradient loss function is back-propagated to the quantum neural network, and the quantum circuit parameters are updated through the chain rule, and the formula is:

[0034] ;

[0035] Wherein, is the parameter of the quantum circuit in the QNN, is the i-th policy network parameter, For the policy gradient loss function.

[0036] As a preferred embodiment, the predicted defect occurrence probability and its severity in the thermosetting resin molding manufacturing process output by the defect prediction AI model are:

[0037] ;

[0038] Wherein, : the predicted probability or defect degree of the first possible defect at time t.

[0039] As a preferred embodiment, the process parameter optimization control strategy includes increasing the reaction temperature, prolonging the reaction time, adjusting the stirring rate, and dynamically adjusting the PH value:

[0040] The formulation structure adjustment control strategy includes adjusting the molar ratio of formaldehyde to urea, adding modifiers, and controlling the filler ratio.

[0041] As a preferred embodiment, after real-time control of the thermosetting resin molding manufacturing process according to the control strategy, the method further comprises:

[0042] Feedback the controlled production data and quality detection results to the defect prediction AI model;

[0043] Dynamically adjust the model parameters through the self-supervised model update mechanism to form a closed-loop control of prediction-control-feedback-optimization.

[0044] According to a second aspect of the present disclosure, a thermosetting resin molding manufacturing process defect intelligent control device is provided, comprising:

[0045] A data acquisition module is configured to acquire multi-source heterogeneous data in a thermosetting resin molding manufacturing process and pre-process the multi-source heterogeneous data;

[0046] A model training module is configured to construct a defect prediction AI model that integrates a trend estimation module, a self-supervised model update mechanism, and a quantum-reinforcement learning optimization framework. The trend estimation module separates the trend component and the dynamic component in the data, the self-supervised mechanism optimizes the model parameters in a label-free scenario, and the quantum neural network and reinforcement learning are combined to achieve efficient search of global parameters.

[0047] A defect prediction module is configured to input the pre-processed multi-source heterogeneous data into the defect prediction AI model and output the predicted defect occurrence probability and its severity in the thermosetting resin molding manufacturing process;

[0048] A control strategy generation module is configured to develop a process parameter optimization or formulation structure adjustment strategy based on the predicted defect occurrence probability and its severity.

[0049] A production regulation module is used to regulate the thermosetting resin forming manufacturing process in real time according to the process parameter optimization or recipe structure adjustment strategy.

[0050] According to a third aspect of the present disclosure, an electronic device is provided, at least one processor, and a memory connected in communication with the at least one processor; wherein the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to perform the method according to any one of the above.

[0051] According to a fourth aspect of the present disclosure, a computer readable storage medium is provided, and computer instructions are used to make the computer execute the method according to any one of the above.

[0052] Compared with the prior art, the present disclosure has the following beneficial effects:

[0053] (1) The present disclosure proposes a trend estimation module for separating the long-term trend component and the instantaneous dynamic component in the feature, estimating the current feature trend by combining the exponential moving average with the sliding window strategy, so that the defect prediction AI model input can eliminate the interference of trend deviation on defect prediction, thereby enhancing the sensitivity and robustness of the model to process state changes.

[0054] (2) The present disclosure proposes a mask vector mechanism to construct a reconstruction loss function, and only the error of the de-trended sample predicted as normal is calculated back, thereby effectively avoiding the misleading of the model training by abnormal samples, solving the interference problem of unlabeled samples in the test phase of the model, and improving the stability and generalization ability of the AI model in the actual production environment.

[0055] (3) The present disclosure proposes an optimization framework based on quantum-reinforcement learning, which combines quantum neural network (QNN) and classical reinforcement learning strategy, realizes more efficient expression and search of model parameters in the optimization process of policy network through the modeling path of "quantum participation-classical execution", so that the model can dynamically adapt to the change of manufacturing conditions, and realize global optimization of model parameters.

[0056] (4) The present disclosure constructs a local and global collaborative optimization mechanism, realizes continuous adjustment of local parameters through self-supervised training of the AI model, and forms a closed-loop optimization loop by combining regulation strategy implementation and feedback, so that the system realizes integrated collaborative evolution of parameters in the aspects of model prediction and production regulation, and improves the adaptability and intelligent decision-making level of the overall manufacturing system.

[0057] It is to be understood that the description in the summary is not intended to identify key or essential features of embodiments of the disclosure or to limit the scope of the disclosure. Other features of the disclosure will be apparent from review of the description below and the associated drawings. BRIEF DESCRIPTION OF DRAWINGS

[0058] The above and other features, aspects, and advantages of embodiments of the present disclosure will become more apparent from the following detailed description when taken in conjunction with the accompanying drawings. The drawings are intended to provide a better description of the scheme and not to limit the disclosure. In the drawings, the same or similar reference numerals refer to the same or similar elements, in which:

[0059] Figure 1 A flowchart of a thermosetting resin molding manufacturing process defect intelligent management and control method according to an embodiment of the present disclosure is shown;

[0060] Figure 2 A block diagram of a thermosetting resin molding manufacturing process defect intelligent management and control device according to an embodiment of the present disclosure is shown;

[0061] Figure 3 A schematic diagram of an exemplary electronic device according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0062] To make the objects, technical solutions, and advantages of embodiments of the present disclosure clearer, the technical solutions in the embodiments of the present disclosure will be described clearly and completely below with reference to the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are some, but not all, of the embodiments of the present disclosure. Based on the embodiments in the present disclosure, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the scope of protection of the present disclosure.

[0063] In addition, the term "and / or" herein merely describes an association relationship of associated objects, which means that there can be three relationships, for example, A and / or B can mean that there are three cases of A alone, A and B together, and B alone. In addition, the character " / " herein generally represents an "or" relationship between the front and rear associated objects.

[0064] As Figure 1 A flowchart of a thermosetting resin molding manufacturing process defect intelligent management and control method according to an embodiment of the present disclosure is shown, the method 100 includes:

[0065] S110: acquiring multi-source heterogeneous data in a thermosetting resin molding manufacturing process, and preprocessing the multi-source heterogeneous data.

[0066] In some embodiments, the data acquisition range in the thermosetting resin (for example, urea-formaldehyde resin) molding manufacturing process should cover the entire production chain, mainly including the following four categories:

[0067] (1) Raw material parameters: such as the mass, purity, molecular ratio, supply batch number, and storage conditions of formaldehyde and urea, which directly affect the reactant ratio and the stability of resin structure, and are important predisposing factors for defects.

[0068] (2) Process parameters: including the temperature curve of the reaction kettle, pressure value, pH value trend, reaction time, stirring speed, and cooling rate, which have a direct impact on the bonding performance, free formaldehyde release, and curing degree of the formed resin.

[0069] (3) Equipment operating state data: including heating power, sensor readings, flowmeter data, valve opening and closing status, and alarm records, which can be collected at high frequency by combining PLC control systems and edge devices to ensure real-time and integrity of production.

[0070] (4) Product quality data: such as bonding strength, solid content, free formaldehyde content, and appearance defects (such as blistering, cracking, and poor bonding), which include both laboratory test results and online detection instrument feedback data, and are the core reference for model output verification and control strategy development.

[0071] The above data collection can be completed through multiple channels such as sensor networks, DCS systems, and MES systems to ensure time consistency and data traceability.

[0072] Furthermore, to ensure the accuracy and stability of subsequent modeling, the collected multi-source heterogeneous data should be systematically cleaned and preprocessed, mainly including the following steps:

[0073] (a) Outlier detection and elimination: statistical methods (such as Z-score and boxplot method) and machine learning methods (such as Isolation Forest and Local Outlier Factor (LOF)) are used to identify data anomalies caused by sensor faults, human errors, or environmental mutations, and to eliminate or correct them.

[0074] (b) Missing value processing: different data imputation strategies are selected based on the characteristics of missing data, such as linear interpolation, time series imputation, and similar sample-based imputation, to maximize the integrity and continuity of the data.

[0075] (c) Feature standardization and normalization: considering the physical dimension differences of each feature, Z-score standardization and Min-Max normalization methods are used to unify the feature scale, avoiding feature weight imbalance during model training.

[0076] (d) Time synchronization and sliding window reconstruction: Align data with different frequencies or different timestamps, construct a time sliding window according to the needs of the prediction task, reconstruct the input feature sequence, and provide a data structure foundation for trend modeling and dynamic analysis.

[0077] S120: Construct a defect prediction AI model that integrates a trend estimation module, a self-supervised model update mechanism, and a quantum-reinforcement learning optimization framework. The trend estimation module separates the trend component and dynamic component in the data, the self-supervised mechanism optimizes the model parameters in unlabeled scenarios, and the quantum neural network and reinforcement learning are combined to achieve efficient search of global parameters.

[0078] To achieve accurate prediction of defects in the molding and manufacturing process of thermosetting resins (e.g., urea-formaldehyde resin), this disclosure presents a defect prediction AI model that includes a trend estimation module, a self-supervised model update mechanism, and a quantum-reinforcement learning optimization framework. By training the defect prediction AI model through the trend estimation module and the self-supervised model update mechanism, the problems of trend drift and insufficient sample labels in manufacturing data can be effectively addressed, thereby improving the model's generalization ability and adaptability.

[0079] In some embodiments, the trend estimation module aims to capture the long-term trend component in manufacturing data using the exponential moving average method, thereby stripping away the dynamic fluctuations in the manufacturing data and separating the trend component from the dynamic component in the features, providing a more stable input for subsequent anomaly detection or defect prediction. The construction of the trend estimation module specifically includes the following steps:

[0080] First, the mean within the current window is calculated using a sliding window approach. Used to depict local feature trends:

[0081] ;

[0082] in, : Feature observations at time i; Size of the sliding window; The average value within the current window, used as the initial input for trend estimation.

[0083] Secondly, based on the trend estimate from the previous window... and the current window average Update the trend estimate at the current time t using the exponential moving average method. :

[0084] ;

[0085] in, The current trend estimate; Smoothing coefficient This is used to balance the importance of historical trends with current information; The trend estimate for the previous window.

[0086] Finally, to eliminate the interference of trend shift on model predictions, the corresponding trend estimate is subtracted from the original features to obtain the detrended feature sequence. :

[0087] ;

[0088] in, : The original feature input within the sliding window at time t; : The feature sequence after detrending; : The corresponding trend estimate.

[0089] The detrended feature sequence Using these defect prediction AI models as input for training and prediction helps enhance their adaptability to non-stationary manufacturing processes.

[0090] In some embodiments, the self-supervised model update mechanism aims to dynamically and adaptively update the model using unlabeled data during the testing phase after the defect prediction AI model is deployed, capturing possible distribution drift or changes in new operating conditions during the manufacturing process. Specifically, it includes the following steps:

[0091] First, samples predicted as normal are screened by the abnormal scores (such as reconstruction loss, mean square error, etc.) output by the defect prediction AI model, eliminating potential abnormal data and retaining only low-risk samples for updating the model.

[0092] Next, after selecting normal samples, the reconstruction loss is calculated using an autoencoder structure, and the model parameters are optimized using online gradient descent. :

[0093] ;

[0094] in, The learning rate (hyperparameter) during testing controls the step size for parameter updates; Loss function on model parameters The gradient; Model parameters control the step size for each update; Detrending features; : Reconstructed features output by the model; Model predicts labels. =0 indicates normal. =1 indicates an anomaly; loss function A masking mechanism is used to calculate the reconstruction error only for samples predicted as normal, thereby avoiding abnormal samples from interfering with model training.

[0095] ;

[0096] in, : Mask vector, activated only for normal samples. Reconstruction error is used to measure the model's ability to represent the current features.

[0097] In some embodiments, in order to further improve the defect prediction capability of the above-mentioned defect prediction AI model in the urea-formaldehyde resin molding manufacturing process, this disclosure proposes an optimization framework based on quantum-reinforcement learning. This framework integrates quantum neural networks (QNN) and reinforcement learning mechanisms, realizes a collaborative modeling strategy of "quantum participation-classical execution", and has stronger policy expression capability and more efficient parameter optimization capability.

[0098] The core of the quantum-reinforcement learning optimization framework lies in using QNN and reinforcement learning to optimize model parameters. Optimization and adjustment are performed to achieve global optimization updates, thereby improving the model's generalization ability and reducing parameter redundancy. The construction process includes the following steps:

[0099] (a) Construction of quantum neural networks:

[0100] QNN is modeled using multi-layer quantum gate circuits, and its basic components include single qubits. Doors and Controlled Door( (gate), its expression is as follows:

[0101] ;

[0102] in, Real-valued parameters, which control the phase modulation parameters of the quantum state; : Real-valued parameter, controlling the rotation angle of the quantum state; Imaginary unit.

[0103] The gate, as a two-qubit gate, is expressed as follows:

[0104] ;

[0105] in, : Single-qubit identity matrix; The ground state of a qubit; :and The parameters in the gate have the same meaning.

[0106] The matrix is ​​represented as:

[0107] ;

[0108] The circuits composed of these quantum gates can generate exponential levels of measurement probabilities, thereby enhancing the expressive power of the network parameter space.

[0109] (b) Parameter mapping model:

[0110] For the above model parameters Further optimization is achieved through a quantum-assisted mechanism. Specifically, a QNN is introduced to generate quantum states. And through the mapping function Compare quantum measurement probability with existing parameters Establish a connection to achieve quantum-guided parameter tuning. The mapping process is defined as follows:

[0111] ;

[0112] in, : By quantum parameters Defined quantum state; The i-th parameter of the policy network (continuing from the parameters obtained during training in step 2.1) ); Total number of parameters; Ground state Bit string encoding; : The state is obtained by measurement The probability of; : Mapping function, These are the parameters of the learnable mapping model.

[0113] This quantum-guided optimization path combines the expressive power of quantum states with the training stability of classical neural networks, realizing a collaborative modeling strategy of "quantum participation-classical execution".

[0114] (c) Policy gradient optimization:

[0115] The reinforcement learning part uses the standard policy gradient loss function to optimize the behavior selection of the policy network:

[0116] ;

[0117] in, : Policy loss function; Policy network parameters; Actions at time t;

[0118] The state at time t; : When the parameter is Under the strategy, execute actions The probability of;

[0119] Standardized return, defined as:

[0120] ;

[0121] in, and Rewards respectively The mean and standard deviation.

[0122] By maximizing the probability of high-reward actions, network strategies can be optimized, enhancing the intelligence of defect prediction and control.

[0123] Furthermore, to achieve closed-loop optimization of the joint quantum-classical architecture, the aforementioned strategy loss function needs to be further optimized. The included reward signal is backpropagated to the QNN to update its quantum circuit parameters. This process is accomplished through the chain rule:

[0124] ;

[0125] in, Parameters of quantum circuits in QNN; : The network parameters of the i-th policy.

[0126] Through this reverse optimization process, QNN can automatically adjust its internal quantum parameters based on the reward feedback signal in reinforcement learning, thereby achieving the co-evolution of the quantum mapping model and the policy learning process, and improving the generalization and adaptability of the overall policy network.

[0127] S130: Input the preprocessed multi-source heterogeneous data into the trained defect prediction AI model to predict the probability and severity of defects in the thermosetting resin molding process.

[0128] In some embodiments, the output of the defect prediction AI model is:

[0129] ;

[0130] in, Historical states or latent variables (such as the historical memory of RNN models, or prediction error information from previous moments) are used to capture time dependence and reflect the model's ability to remember the evolution trend of the production process.

[0131] The input feature vector at the current time t includes:

[0132] ;

[0133] : The i-th raw material ratio, process parameters (temperature, pressure, reaction time, stirring rate, etc.) or equipment status information at the current time t;

[0134] The prediction function is expressed as:

[0135] ;

[0136] : Reconstructed features output by the model;

[0137] , The weights and biases of the model's output layer;

[0138] Activation functions (such as sigmoid or softmax) are used to output defect probabilities or confidence levels.

[0139] The prediction output for the current time t includes multiple defect indicators, such as:

[0140] ;

[0141] : No. The predicted probability or degree of a possible defect (such as insufficient adhesive strength, incomplete curing, loose structure, etc.) at time t.

[0142] The predicted output It can be used to determine which quality defects have a higher risk; based on For defects with high values, the system can match the corresponding control rules for processing.

[0143] S140: Develop control strategies for the manufacturing process based on the predicted probability of defects and their severity.

[0144] In some embodiments, based on the predicted probability of defect occurrence and its severity, this step focuses on proactive intervention in defect risk. The formulation of control strategies integrates expert knowledge bases and data-driven analysis, and has the following two types of control methods:

[0145] (a) Optimization of process parameters: such as increasing the reaction temperature to accelerate the resin crosslinking reaction, extending the reaction time to reduce free formaldehyde residue, appropriately adjusting the stirring rate to improve the reaction uniformity, and dynamically adjusting the pH to control the reaction rate and product stability.

[0146] (b) Formula structure adjustment: Adjust the raw material ratio according to the prediction results, such as adjusting the molar ratio of formaldehyde and urea, adding modifiers to enhance performance, and controlling the filler ratio to reduce appearance defects such as particle aggregation and bubbles.

[0147] This strategy formulation module can combine the model's sensitivity analysis and backpropagation contribution to generate precise control recommendations.

[0148] S150: The thermosetting resin molding process is controlled in real time according to the control strategy.

[0149] In some embodiments, after the above-mentioned control strategy is integrated into the production control system, real-time process adjustments are made through an automated execution module, forming a model-driven active control closed loop. Specific implementation methods include:

[0150] (1) Connect to the DCS / PLC control system and automatically adjust key actuators such as heaters, pressure valves, stirring systems, and pH adjustment devices by setting control commands.

[0151] (2) Combine the real-time prediction results to set dynamic control thresholds, such as temperature adjustment range and reaction time extension step, to realize a graded response strategy.

[0152] (3) The system supports linkage control, that is, when multiple defect indicators trigger early warning at the same time, a multi-objective coordinated control strategy is adopted to ensure the overall optimization of product performance.

[0153] Furthermore, after the control measures are implemented, the new round of production process data and final quality inspection results need to be transmitted back to the AI ​​model in real time to achieve model adaptation and strategy re-optimization. The main processes include:

[0154] Use the complete data before and after regulation as new training samples to expand the model dataset;

[0155] The self-supervised learning mechanism in step S120 above is used to update the model parameters online;

[0156] Dynamically evaluate the effectiveness and stability of the control strategies;

[0157] The strategy library is updated by weeding out the inferior ones, strengthening effective strategies and eliminating ineffective interventions.

[0158] This closed-loop control model of "prediction-regulation-feedback-optimization" enables the co-evolution of the model and the production system.

[0159] In some embodiments, the defect prediction model and control scheme of this solution can be seamlessly embedded into the urea-formaldehyde resin molding manufacturing production line. Through an interface program, it can interface with subsystems such as PLC, SCADA, and MES to achieve end-to-end data communication and control signal interaction, specifically including:

[0160] Data acquisition, prediction result distribution, and control command push are achieved through industrial protocols such as OPCUA; lightweight middleware is built to support multi-level deployment modes such as local / edge / cloud, ensuring system stability and response speed; and standardized interfaces are provided for subsequent expansion, upgrades, and cross-device integration.

[0161] Additionally, the following functions can be achieved through a visual interface on a web or industrial configuration platform:

[0162] Develop a visual interface based on a web-based or industrial configuration platform to achieve the following functions:

[0163] Real-time display of key production parameters, predicted defect types, and their risk levels;

[0164] Visual tracking of control strategy recommendations and their implementation status;

[0165] System status monitoring, alarm notification and emergency response mechanism;

[0166] It supports historical data query and report export to assist in quality analysis and decision-making.

[0167] Finally, it is worth emphasizing that, with the accumulation of data and changes in technology, the system supports continuous iterative upgrades of models and strategies, including:

[0168] Automatically triggers the model retraining process, integrating the latest data to update parameters; reconstructs feature importance ranking and adjustment priorities based on feedback data; introduces mechanisms such as federated learning and transfer learning to support multi-production line model sharing and generalization; combines operation and maintenance data to evaluate the indirect impact of equipment status on product quality and improves the control logic, etc.

[0169] Through continuous optimization and intelligent upgrades, the system can always maintain its adaptability and control over complex production conditions.

[0170] According to the above embodiments of this disclosure, the following technical effects are achieved:

[0171] (1) This disclosure proposes a trend estimation module to separate the long-term trend component and the instantaneous dynamic component in the feature. The current feature trend is estimated by combining exponential moving average with a sliding window strategy, so that the model input can eliminate the interference of trend offset on defect prediction, thereby enhancing the model's sensitivity and robustness to changes in process status.

[0172] (2) This disclosure proposes a mask vector mechanism to construct the reconstruction loss function, and only performs error backpropagation calculation on samples predicted as normal, thereby effectively avoiding the misleading effect of abnormal samples on model training, solving the problem of interference from unlabeled samples in the testing phase of the model, and improving the stability and generalization ability of AI models in actual production environments.

[0173] (3) This disclosure proposes an optimization framework based on quantum reinforcement learning, which integrates quantum neural network (QNN) and classical reinforcement learning strategy. Through the modeling path of "quantum participation - classical execution", it realizes more efficient expression and search of model parameters in the policy network optimization process, so that the model can dynamically adapt to changes in manufacturing conditions and achieve global optimization.

[0174] (4) This disclosure constructs a local and global collaborative optimization mechanism. Through the self-supervised training of the AI ​​model, the local parameters are continuously adjusted. At the same time, the implementation and feedback of the control strategy form a closed-loop optimization circuit, enabling the system to achieve integrated collaborative evolution of parameters at both the model prediction and production control levels, thereby improving the adaptability and intelligent decision-making level of the overall manufacturing system.

[0175] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this disclosure is not limited to the described order of actions, because according to this disclosure, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to this disclosure.

[0176] The above is an introduction to the method embodiments. The following describes the solution described in this disclosure further through device embodiments.

[0177] Figure 2 A block diagram of an intelligent defect control device for thermosetting resin molding manufacturing process according to an embodiment of the present disclosure is shown. Figure 2 As shown, the device 200 includes:

[0178] The data acquisition module 210 is used to acquire multi-source heterogeneous data during the thermosetting resin molding process and to preprocess the multi-source heterogeneous data.

[0179] The model training module 220 is used to construct a defect prediction AI model that includes a trend estimation module and a self-supervised model update mechanism, and to perform global optimization of the parameters of the defect prediction AI model by combining a quantum-reinforcement learning optimization framework to obtain a trained defect prediction AI model.

[0180] The defect prediction module 230 is used to input the preprocessed multi-source heterogeneous data into the trained defect prediction AI model to predict the probability and severity of defects in the thermosetting resin molding process.

[0181] The control strategy generation module 240 is used to formulate control strategies for the production and manufacturing process based on the predicted probability of defect occurrence and its severity.

[0182] The production control module 250 is used to control the thermosetting resin molding process in real time according to the control strategy.

[0183] The acquisition, storage, and application of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0184] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the described module can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0185] The acquisition, storage, and application of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0186] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0187] Figure 3 A schematic block diagram of an electronic device 300 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0188] Electronic device 300 includes a computing unit 301, which can perform various appropriate actions and processes according to a computer program stored in ROM 302 or a computer program loaded into RAM 303 from storage unit 308. RAM 303 can also store various programs and data required for the operation of electronic device 300. The computing unit 301, ROM 302, and RAM 303 are interconnected via bus 304. I / O interface 305 is also connected to bus 304.

[0189] Multiple components in electronic device 300 are connected to I / O interface 305, including: input unit 306, such as keyboard, mouse, etc.; output unit 307, such as various types of displays, speakers, etc.; storage unit 308, such as disk, optical disk, etc.; and communication unit 309, such as network card, modem, wireless transceiver, etc. Communication unit 309 allows electronic device 300 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0190] The computing unit 301 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 301 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 301 performs the various methods and processes described above, such as method 100. For example, in some embodiments, method 300 may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 308. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 300 via ROM 302 and / or communication unit 309. When the computer program is loaded into RAM 303 and executed by the computing unit 301, one or more steps of method 100 described above may be performed. Alternatively, in other embodiments, the computing unit 301 may be configured to perform method 100 by any other suitable means (e.g., by means of firmware).

[0191] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0192] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0193] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0194] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including voice input, speech input, or tactile input).

[0195] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0196] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0197] The acquisition, storage, and application of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0198] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this disclosure can be achieved, and this is not limited herein.

[0199] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for intelligent control of defects in the thermosetting resin molding manufacturing process, characterized in that, include: Acquire multi-source heterogeneous data during the thermosetting resin molding process, and preprocess the multi-source heterogeneous data. A defect prediction AI model is constructed that integrates a trend estimation module, a self-supervised model update mechanism, and a quantum-reinforcement learning optimization framework. The trend estimation module separates the trend component and the dynamic component in the data, the self-supervised mechanism optimizes the model parameters in the unlabeled scenario, and the quantum neural network and reinforcement learning are combined to achieve efficient search of global parameters. The preprocessed multi-source heterogeneous data is input into the defect prediction AI model, and the predicted probability and severity of defects in the thermosetting resin molding process are output. Based on the predicted probability of defect occurrence and its severity, formulate strategies for optimizing process parameters or adjusting the formulation structure; The thermosetting resin molding process is controlled in real time based on the optimization of process parameters or the adjustment of formulation structure.

2. The method according to claim 1, characterized in that, The preprocessing of the multi-source heterogeneous data includes: Statistical and machine learning methods are used to identify outlier data, and outlier data is removed or corrected. Based on the characteristics of the missing data, linear interpolation, time series backfilling, or similar sample imputation strategies are used to backfill the missing data samples. The backfilled data samples were processed by feature standardization and normalization using Z-score standardization or Min-Max normalization methods. Align data from different frequencies or timestamps, construct a time sliding window according to the prediction task requirements, and reconstruct the input feature sequence.

3. The method according to claim 1, characterized in that, The method for constructing the trend estimation module is as follows: Calculate the mean value within the current window using a sliding window method; Based on the trend estimate from the previous window and the mean within the current window, the trend estimate for the current moment is updated using the exponential moving average method. Subtract the updated trend estimate for the current time from the original feature corresponding to the current time to obtain the detrended feature sequence.

4. The method according to claim 1, characterized in that, The self-supervised model update mechanism includes: The defect prediction AI model filters out data samples that are predicted to be normal by the mask vector mechanism. The reconstruction loss is calculated using a loss function on the normal sample data, and the model parameters are optimized using online gradient descent. The loss function is: ； in, This is a mask vector used to activate normal samples; The reconstruction error is used to measure the model's ability to represent the current features. The input is detrended features; The reconstructed features output by the model; Predict labels for the model, =0 indicates normal. =1 indicates an exception.

5. The method according to claim 1, characterized in that, The method for constructing the quantum-reinforcement learning optimization framework is as follows: A quantum neural network is constructed using multi-layer quantum gate circuits to generate exponential quantum measurement probabilities. A parameter mapping model is introduced to map the quantum measurement probability to policy network parameters; The behavior selection of the policy network is optimized using the standard policy gradient loss function; The mapping process is as follows: ； in, : By quantum parameters Defined quantum state; The i-th parameter of the policy network (continuing from the parameters obtained during training in step 2.1) ); Total number of parameters; Ground state Bit string encoding; : The state is obtained by measurement The probability of; : Mapping function, These are the parameters of the learnable mapping model.

6. The method according to claim 5, characterized in that, After optimizing the behavior selection of the policy network using the standard policy gradient loss function, the following is also included: The reward signal contained in the policy gradient loss function is passed in reverse to the quantum neural network, and the quantum circuit parameters are updated using the chain rule, as shown in the formula: ； in, For the parameters of the quantum circuit in QNN, For the i-th policy network parameters, Let be the policy gradient loss function.

7. The method according to claim 1, characterized in that, The defect prediction AI model outputs the predicted probability and severity of defects during the thermosetting resin molding process: ； in, : No. The predicted probability or degree of a possible defect at time t.

8. The method according to claim 1, characterized in that, The optimization strategies for process parameters include increasing the reaction temperature, extending the reaction time, adjusting the stirring rate, and dynamically adjusting the pH value. The control strategies for adjusting the formulation structure include adjusting the molar ratio of formaldehyde to urea, adding modifiers, and controlling the proportion of fillers.

9. The method according to claim 1, characterized in that, After the real-time control of the thermosetting resin molding process according to the control strategy, the method further includes: The adjusted production data and quality inspection results are fed back as extended data to the defect prediction AI model; The self-supervised model update mechanism dynamically adjusts the model parameters, forming a closed-loop control of prediction, regulation, feedback, and optimization.

10. An intelligent defect control device for thermosetting resin molding manufacturing process, characterized in that, include: The data acquisition module is used to acquire multi-source heterogeneous data during the thermosetting resin molding process and to preprocess the multi-source heterogeneous data. The model training module is used to build a defect prediction AI model that integrates a trend estimation module, a self-supervised model update mechanism, and a quantum-reinforcement learning optimization framework. The trend estimation module separates the trend component and dynamic component in the data, the self-supervised mechanism optimizes the model parameters in unlabeled scenarios, and the quantum neural network and reinforcement learning are combined to achieve efficient search of global parameters. The defect prediction module is used to input the preprocessed multi-source heterogeneous data into the defect prediction AI model and output the predicted probability and severity of defects in the thermosetting resin molding process. The control strategy generation module is used to formulate process parameter optimization or formulation structure adjustment strategies based on the predicted probability of defect occurrence and its severity. The production control module is used to control the thermosetting resin molding process in real time according to the optimization strategy of the process parameters or the formula structure adjustment strategy.

Images