Two-stage coevolution method for deep CNN: A case study in smart manufacturing

Smart manufacturing system is very complex and there are a lot of different types of data to deal with, which lead to the difficulty of usage. Frequently manually tuning hyperparameters and modifying the architecture of the network have become a major problem for participants, and it has seriously affected the application and promotion of Deep learning (DL) in industry. In order to solve this problem, a novel self-evolving deep CNN method: two-stage coevolution method (TSC) is proposed in this paper to automatically optimize the hyperparameters and effectively evolve the most suitable network by summarizing the characteristics of the excellent artificial architectures. The first stage is mainly to optimize the hyperparameters with Orthogonal experimental algorithm. The second stage is to produce the best deep CNN with the optimized hyperparameters through self-evolving computation. In the second stage, three well-known deep-CNN architectures are used as the initialization seeds and each seed is presented by a particle and a gene to coevolve the necessary factors for a deep CNN driven by particle swarm optimization (PSO) and genetic algorithm (GA). At last, a case study for smart manufacturing systems was carried out to demonstrate the effectiveness and convenience of the proposed method. And the TSC method was also compared with other two self-evolving methods. The experiment results show that TSC method is superior over other well-known algorithms.(c) 2023 Elsevier B.V. All rights reserved.

Automated summary

This paper proposes a novel self-evolving deep CNN method to address the challenges of using DL in smart manufacturing systems. By optimizing hyperparameters and evolving network architectures, the proposed method demonstrates superior performance compared to other self-evolving methods in the experiment.