Supervised functional state-space modeling for monitoring multigrade batch processes with irregular data using meta-learning

The diverse nature of product grades in industrial batch processes often leads to limited data availability for each grade. Effective process monitoring requires the development of a robust statistical model based on historical data, capable of capturing the intricate relationships between input and output process variables. Furthermore, industrial batch data are susceptible to contamination from external disturbances and noise, resulting in irregular sampling intervals if contaminated samples are eliminated. This irregularity presents challenges in constructing monitoring models for these processes. This study presents an innovative approach that synergizes meta-learning techniques with functional state space modeling to address the complexities of multi-grade batch process dynamics. The methodology exploits wavelet function to mitigate irregular sampling challenges. Furthermore, it implements a meta-learning strategy that extracts insights from varied historical batch records across multiple grades, enhancing model robustness and adaptability. This methodology enables swift development of process model for new grade products, maintaining efficacy even when confronted with limit dataset scenarios. The efficacy of the proposed method is validated through both numerical simulations and application to a real-world batch process case.