CN-121806513-B - Optical crystal production dynamic optimization control system based on defect monitoring

CN121806513BCN 121806513 BCN121806513 BCN 121806513BCN-121806513-B

Abstract

The invention discloses an optical crystal production dynamic optimization control system based on defect monitoring, which comprises a defect monitoring module, a reinforcement learning decision module, an in-situ repair execution module, a process optimization module, a data feedback module, an accuracy correction module and a federal learning cooperative module. The dual-layer framework combining the inner layer real-time closed loop and the outer layer cross-domain collaborative closed loop is adopted, wherein the inner layer closed loop realizes millisecond real-time optimization by defect monitoring, real-time decision making, restoration and process optimization and according to correction factors generated by a data feedback and precision correction module, and the outer layer closed loop generates global optimization parameters by aggregating cross-batch, cross-procedure and supply chain data by a federal learning collaborative module to carry out collaborative enhancement on decisions. The reinforcement learning decision module synchronously receives and fuses the precision correction factors and the global optimization parameters, dynamically corrects the decision instructions of the reinforcement learning decision module, and realizes the dynamic optimization control of the whole production chain.

Inventors

YANG JIANCHUN
ZHANG FURU
LI YUANSHENG
WEI WENXIANG
HUANG HANGUANG

Assignees

宁波翌波光电科技有限公司

Dates

Publication Date: 20260508
Application Date: 20260306

Claims (10)

1. An optical crystal production dynamic optimization control system based on defect monitoring, which is characterized by comprising: The defect monitoring module is used for collecting production data, extracting characteristics and outputting defect characteristic vectors; The reinforcement learning decision module is connected with the defect monitoring module and is used for receiving the defect characteristic vector and the production state parameter and outputting a decision instruction comprising an in-situ repair action and/or a process optimization action through a reinforcement learning model; the in-situ repair execution module is connected with the reinforcement learning decision module and is used for executing the in-situ repair action and collecting repair data; the process optimization module is connected with the reinforcement learning decision module and is used for executing the process optimization action and collecting process data; the data feedback module is respectively connected with the in-situ repair execution module and the process optimization module, and is used for storing the repair data and the process data and feeding back the repair data and the process data to the reinforcement learning decision module; the precision correction module is respectively connected with the defect monitoring module and the reinforcement learning decision module and comprises: the variable acquisition unit is used for acquiring raw material purity, equipment vibration, environmental data and growth dynamic data to form acquisition data; the correction factor unit is connected with the variable acquisition unit and used for generating correction factors according to the acquired data and the historical data and sending the correction factors to the reinforcement learning decision module; The federal learning coordination module is respectively connected with the reinforcement learning decision module and the precision correction module, and comprises: a data acquisition unit for acquiring cross-domain data from an external system and a supply chain; The collaborative optimization unit is connected with the data acquisition unit and is used for carrying out privacy aggregation on the cross-domain data through federal learning, generating global optimization parameters and sending the global optimization parameters to the reinforcement learning decision module; and the reinforcement learning decision module dynamically corrects the decision instruction according to the correction factor and the global optimization parameter.
2. The defect-monitoring-based optical crystal production dynamic optimization control system of claim 1, wherein the reinforcement learning decision module comprises: The system comprises a state space construction unit, a state space generation unit and a control unit, wherein the state space construction unit is used for constructing a state space containing defect characteristic vectors and production system state parameters, and the production system state parameters at least comprise crystal growth rate, furnace body temperature distribution, raw material proportion, energy consumption and production efficiency; the system comprises an action space construction unit, a process optimization unit and a control unit, wherein the action space construction unit is used for constructing an action space comprising an in-situ repair action and a process optimization action, the in-situ repair action at least comprises laser power, laser action time and local temperature field gradient, and the process optimization action at least comprises a growth rate adjustment amount, a temperature adjustment amount and a raw material proportion adjustment amount; The rewarding function construction unit is used for constructing a composite rewarding function, and the composite rewarding function is used for carrying out weighted calculation on defect repair quality, production efficiency and energy consumption indexes based on a preset quality weight coefficient, an efficiency weight coefficient and an energy consumption weight coefficient to obtain a comprehensive rewarding value; The training and decision unit is respectively connected with the state space construction unit, the action space construction unit and the rewarding function construction unit and is used for executing training and real-time decision tasks of the reinforcement learning model.
3. The defect-monitoring-based optical crystal production dynamic optimization control system according to claim 2, wherein the training and decision unit comprises: the model initialization subunit is used for loading a history training sample set to pretrain the reinforcement learning model; the real-time decision sub-unit is used for receiving the real-time state parameters and outputting an optimal decision instruction based on the model after the pre-training; A parameter updating subunit, configured to iteratively update the network parameter of the reinforcement learning model according to the feedback data from the data feedback module and the correction factor from the precision correction module; And when the real-time decision sub-unit outputs a decision instruction, synchronously receiving and applying the correction factors sent by the precision correction module and the global optimization parameters sent by the federal learning coordination module to dynamically correct the decision instruction.
4. The defect-monitoring-based optical crystal production dynamic optimization control system of claim 1, wherein the defect monitoring module comprises: the multi-mode sensing unit is used for collecting multi-source monitoring data of the optical crystal production process, and the multi-source monitoring data at least comprise spectrum data, ultrasonic data and infrared thermal imaging data; The noise filtering unit is connected with the multi-mode sensing unit and is used for processing the multi-source monitoring data through a wavelet packet transformation algorithm so as to filter electromagnetic interference and equipment noise in the multi-source monitoring data; the feature extraction unit is connected with the noise filtering unit and is used for extracting core feature parameters of defects based on the filtered multi-source monitoring data and generating defect feature vectors containing defect types, defect positions, defect sizes and defect growth rates; The defect feature vector is output to the reinforcement learning decision module at a sampling frequency not lower than a first preset threshold.
5. The defect-monitoring-based optical crystal production dynamic optimization control system of claim 1, wherein the in-situ repair execution module comprises: The positioning control unit is used for controlling the three-dimensional motion platform to position the focus of the repairing device to the defect center according to the defect position information in the decision instruction, and the positioning precision is not lower than a first positioning precision threshold; The laser execution unit is connected with the positioning control unit and used for controlling the high-precision pulse laser to execute pulse laser repairing operation according to the laser power and the laser action time parameter in the decision instruction; the temperature field execution unit is connected with the positioning control unit and is used for controlling the partition temperature control device to perform local temperature field adjustment on the repair area according to the local temperature field gradient parameters in the decision instruction; The repairing device comprises the high-precision pulse laser and the partition temperature control device, wherein the laser executing unit and the temperature field executing unit synchronously acquire temperature data and stress data of a repairing area when repairing operation is executed, and the temperature data and the stress data are used as repairing process data to be sent to the data feedback module.
6. The defect-monitoring-based optical crystal production dynamic optimization control system of claim 5, wherein the process optimization module comprises: the parameter receiving unit is used for receiving process optimization actions contained in the decision instruction, and the process optimization actions at least comprise a crystal growth rate adjustment quantity, a temperature adjustment quantity and a raw material proportion adjustment quantity; The control signal generation unit is connected with the parameter receiving unit and is used for converting the process optimization action into a corresponding control signal based on an adaptive PID compensation algorithm; The execution adjusting unit is connected with the control signal generating unit and is used for driving the crystal growth furnace control system, the raw material supply adjusting device and the energy consumption monitoring instrument to execute process parameter adjustment according to the control signal; the process state acquisition unit is connected with the execution regulation unit and used for acquiring the adjusted growth rate, temperature distribution, raw material proportion and energy consumption data in real time and sending the data to the data feedback module as process state data; The adjusting precision of the raw material supply adjusting device is not lower than a first precision threshold, and the temperature control precision of the crystal growth furnace control system is not lower than a second precision threshold.
7. The defect-monitoring-based optical crystal production dynamic optimization control system of claim 6, wherein the data feedback module comprises: a data receiving unit for receiving and integrating the repair process data and the process state data; the data cleaning unit is connected with the data receiving unit and is used for carrying out outlier detection and filtering treatment on the integrated data through a data cleaning algorithm; The data storage unit is connected with the data cleaning unit and used for storing cleaned data in the distributed database; the sample packaging unit is connected with the data storage unit, and is used for extracting data from the distributed database according to a preset period and packaging the data into a training sample set containing state parameters, action instructions and rewarding values; Wherein the training sample set is fed back to the reinforcement learning decision module for iterative updating of its model parameters.
8. The defect-monitoring-based optical crystal production dynamic optimization control system of claim 1, wherein the correction factor unit in the precision correction module comprises: The data preprocessing subunit is used for preprocessing raw material purity data, equipment vibration data, environment data and growth dynamic data from the variable acquisition unit; The weight distribution subunit is connected with the data preprocessing subunit and is used for distributing real-time weights for various dynamic data based on a self-adaptive mutual information entropy-random forest algorithm; the exception handling subunit is connected with the weight allocation subunit and is used for identifying and shielding exception data based on a preset statistical criterion and correspondingly adjusting the weight of the shielded data; The correction factor calculation subunit is respectively connected with the weight distribution subunit, the abnormality processing subunit and the data feedback module and is used for calculating a repair parameter correction factor and a process parameter correction factor through an online gradient descent algorithm according to the weighted dynamic data and the historical training sample set provided by the data feedback module; And the correction factor calculation subunit sends the generated correction factor to the reinforcement learning decision module with response time not higher than a preset delay threshold.
9. The defect monitoring-based optical crystal production dynamic optimization control system of claim 3, wherein the data acquisition unit in the federal learning coordination module comprises: A process data acquisition subunit for acquiring raw material granularity data from the raw material pretreatment process, surface roughness data from the processing process, and optical performance data from the detection process; A supply chain data interface subunit, configured to obtain raw material batch quality data and equipment maintenance record data from the supply chain database and the equipment maintenance management system through the API interface; The data acquisition unit integrates the acquired raw material granularity data, surface roughness data, optical performance data, raw material batch quality data and equipment maintenance record data into cross-domain associated data, and sends the cross-domain associated data to the collaborative optimization unit.
10. The defect-monitoring-based optical crystal production dynamic optimization control system of claim 9, wherein the collaborative optimization unit in the federal learning collaborative module comprises: The parameter aggregation subunit is used for carrying out homomorphic encryption processing on the cross-domain associated data based on a hierarchical federal learning architecture, and aggregating model parameters from different production batches and procedures to generate global optimization parameters; A reward function adjustment subunit, configured to extract supply chain features from the cross-domain association data and generate an association weight vector, and dynamically adjust a composite reward function of the reinforcement learning decision module by using the association weight vector; The cross-domain processing subunit is used for identifying abnormal characteristics in the cross-domain associated data through an isolated forest algorithm, positioning an abnormal source based on a blockchain tracing node and triggering a substitute optimization scheme; The parameter aggregation subunit sends the global optimization parameter to the reinforcement learning decision module, and the reward function adjustment subunit sends the dynamically adjusted composite reward function to the reinforcement learning decision module.

Description

Optical crystal production dynamic optimization control system based on defect monitoring Technical Field The invention relates to the technical field of industrial process control and intelligent manufacturing, in particular to a dynamic optimization control system for optical crystal production based on defect monitoring. Background The optical crystal is widely applied to the high-end technical fields of lasers, nonlinear optical devices, precise optoelectronic systems, semiconductor manufacturing and the like, and the internal structural integrity and the growth consistency of the optical crystal directly influence the stability, the reliability and the long-term service life of the optical performance of the device. In the crystal growth and subsequent processing process, the defects are easily generated in the crystal and on the surface due to the influence of multiple factors such as raw material fluctuation, equipment running state change, environmental factor disturbance and the like, the defects often have evolutionary and cumulative characteristics, and higher requirements are put on the dynamic control capability of the production process. With the increase of production scale and the continuous improvement of crystal performance indexes, optical crystal manufacturing has gradually progressed to the data driving and intelligent control direction by traditional empirical process adjustment. However, from the perspective of the whole production flow, the existing production control system still faces various technical bottlenecks in terms of defect sensing, decision response and cross-link coordination, and the requirements of high precision, continuity and self-adaptive optimization are difficult to meet. On the one hand, the generation and evolution of defects in the crystal production process are often highly coupled with technological parameters, equipment states and environmental conditions, if a defect monitoring result cannot form effective linkage with a subsequent control decision, control response delay or single processing mode is easy to cause, and local defect inhibition and overall process stability are difficult to be considered. Under a complex production scene, only static rules or manually set parameters are relied on, so that differentiated response aiming at different defect characteristics is difficult to realize, and a closed-loop optimization mechanism for continuous feedback is not easy to form. On the other hand, the multi-source uncertain factors in the production process have obvious staged and hidden characteristics, such as raw material quality fluctuation, equipment performance attenuation, environmental perturbation and the like, and the factors have obvious differences in the influence degree of the factors on the crystal quality in different production stages. If the control model lacks dynamic correction capability, decision offset and control accuracy drop easily occur in long-term operation, and production consistency and yield are affected. In addition, under the condition of multi-batch or large-scale production, the data samples available by a single production unit are limited, and model training and optimization are easily limited by insufficient samples. Meanwhile, data generated by different production units, upstream and downstream processes and supply chain links have important cooperative values, but related information is difficult to directly share due to the requirements of data safety and privacy protection, so that cross-domain data cannot fully participate in overall optimization decision, and global cooperative capacity of a production system is limited. In summary, the existing optical crystal production control system still has obvious defects in aspects of defect response closed loop, autonomous decision adaptability, long-term operation stability, cross-domain data collaborative capability and the like, and a production dynamic optimization control mechanism capable of integrating defect monitoring, intelligent decision, dynamic correction and multi-source data collaborative is needed to realize high-precision, stabilization and continuous optimization of the optical crystal production process. Disclosure of Invention Aiming at the defects existing in the prior art, the invention aims to provide an optical crystal production dynamic optimization control system based on defect monitoring, which is used for synchronously solving the problems of defect response lag, model precision drift and data island through fusion of an inner layer real-time closed loop and an outer layer cross-domain collaborative closed loop and realizing self-adaptive optimization and global collaborative control of a production process. In order to achieve the above purpose, the invention provides a dynamic optimization control system for optical crystal production based on defect monitoring, which comprises the following technical scheme: The defect monitoring module i