LSTM-based framework with metaheuristic optimizer for manufacturing process monitoring

Quick process shift detection and lower out-of-control run length are essential for monitoring the production process, especially in modern smart manufacturing. Specifically, the out-of-control run length is one of the most critical performance measures to evaluate the manufacturing process monitoring (MPM) model. The sooner the out-of-control is detected, the better the model is. However, developing a monitoring model which can provide quick shift detection for various data dimensions and volumes is challenging. In this research, single (1_LSTM) and stacked (S_LSTM) long-short-term memory (LSTM) based models with metaheuristic optimizer were proposed to detect process shifts quickly in the manufacturing domain. Based on the literature, three metaheuristic methods: Clustering-based organism search (CSOS), Particle Swarm Optimization (PSO), and Simulated Annealing (SA) that are suitable for high-dimensional optimization were utilized in the proposed method to optimize weights in the LSTM-based network. The proposed models were evaluated based on average out-of-control run length (ARL1) against benchmark methods on various synthesized multivariate normal and real-world datasets. Also, the performances of CSOS, PSO, and SA were compared. The results show that CSOS_S_LSTM outperforms other methods with lower ARL1. The result also confirmed the effectiveness and applicability of the proposed models for real-world problems. The experimental results showed that the response time of detection can be improved by 33.19% and 38.77% on average using the proposed 1_LSTM and CSOS LSTM-based metaheuristics models, respectively.

Automated summary

Quick process shift detection is crucial in modern smart manufacturing. This research proposes LSTM-based models with metaheuristics optimizer for fast shift detection. The models were evaluated on various datasets and outperformed benchmark methods, confirming their effectiveness for real-world problems.