In the manufacturing of Czochralski-grown silicon single crystals, production control mainly relies on operator experience and conventional feedback strategies, which are often insufficient for addressing the nonlinear, time-varying nature and sensitivity to disturbances of the process. Consequently, real-time tracking of key parameter dynamics remains challenging, resulting in adjustment delays and limiting improvements in product quality and production efficiency. To overcome these challenges, this study proposes a hybrid framework that integrates a Particle Swarm Optimization (PSO)-based Long Short-Term Memory (LSTM) prediction model with a Model-Free Adaptive Control (MFAC) method. PSO is employed to optimize the hyperparameters of LSTM and MFAC, thereby improving prediction accuracy and enhancing the responsiveness and stability of diameter regulation. Experimental results demonstrate that the PSO-LSTM prediction model significantly improves prediction accuracy, enabling more effective decision support during crystal growth. Moreover, the combined PSO-MFAC strategy achieves faster diameter control response than traditional PID control. This work integrates predictive modeling and advanced control techniques into the Czochralski silicon crystal growth process, offering a practical approach to enhancing production reliability and efficiency.
扫码关注我们
求助内容:
应助结果提醒方式:
