Optimization of process manufacturing parameters through integration of CNN-mLSTM and whale algorithm

In the field of process manufacturing, the quality evaluation indicators for product quality play a crucial role in determining superiority or inferiority. Addressing the challenges associated with constructing a coupling model between process parameters, analyzing multi-objective conflicts, and ensuring real-time and accurate performance, this paper proposes an optimization algorithm for process manufacturing parameters. This algorithm integrates Convolutional Neural Networks (CNNs) with Matrix Long Short-Term Memory (mLSTM) and the Whale Optimization Algorithm. Initially, a continuous feature matrix is constructed from extensive process data, operational data, and temporal features using a time-series sliding window approach, which serves as the input for the algorithm. During the model construction phase, the CNN is utilized for dimensionality expansion and feature extraction, while the mLSTM network capitalizes on its temporal extraction capabilities and parallel computing performance to efficiently train and predict product quality. Furthermore, the predicted quality indicators are employed as the fitness function for the Whale Optimization Algorithm, which performs global optimization of the current production state to identify the optimal combination of process parameters across different conditions. This approach facilitates real-time prediction and dynamic optimization of processing quality across various steps in the production process. Finally, the validity and effectiveness of the proposed algorithm are demonstrated through a case study on a specific tobacco shredding line.