Industrial Artificial Intelligence from Data Analytics to Simulation

Abstract

Intelligent industrial systems are key to the future industry. Intelligent industrial systems have continued to develop along with the advancement of digital technologies, including the Internet of Things, big data analytics, and cloud computing. Intelligent industrial systems are rapidly transforming a wide range of industries, such as manufacturing, healthcare, and energy, and helping improve product quality or solve potential problems. However, for the wide adaptation of intelligent industrial systems, major challenges need to be addressed: data processing and analysis, real-time simulation, and system-level decision-making. Artificial intelligence (AI) technology has made significant achievements in recent years and has been adapted to intelligent industrial systems, but it has only achieved limited success. This dissertation presents researches on industry specialized AI. The first study was on AI-integrated probabilistic data analytics. A deep learning-integrated Bayesian health indicator is proposed for evaluating machine degradation, and the proposed indicator showed consistent evaluation even in cross-machine applications. In addition, an unsupervised fault clustering and domain adaptation framework is proposed for accurate fault identification in condition-varying systems, and the proposed framework showed accurate fault identification performance in various conditions without re-training. The second study was on physics-integrated AI for fast numerical simulation. A physics knowledge-integrated AI method is proposed, and the proposed method provided accurate numerical solutions for various simulation conditions in real-time. The last study was on multi-agent reinforcement learning for system-level decision- making. A pilot study on a traffic signal control system is designed and the effectiveness of multi-agent reinforcement learning in solving system-level decision- making problems is demonstrated by the pilot study. These researches in data analytics, real-time simulation, and system-level decision-making will greatly contribute to the AI-powered intelligent industrial systems.