AI-Based Perception and Adaptive Decision-Support in Industrial and Environmental Applications

The digitalization promoted by Industry~4.0 has enabled Smart Factories to generate large volumes of heterogeneous data through interconnected cyber-physical systems. However, this sensing capability has not been matched by equivalent advances in transforming data into actionable knowledge, resulting in a persistent gap between data acquisition and operational decision-making. In industrial environments, this gap is further aggravated by epistemic uncertainty, related to incomplete or noisy system state information, and stochastic uncertainty, arising from unpredictable future events.

This research addresses this challenge by framing industrial scheduling as a continuous perception-informed decision process rather than a static optimization task. The research proposes the systematic integration of AI-based perception with adaptive optimization to enable context-aware and resilient decision-making.

First, by developing advanced Deep Learning perception mechanisms, reliable state information is extracted from noisy and heterogeneous data. Second, robust mathematical scheduling models are extended to incorporate realistic industrial constraints and to analyze the trade-offs between optimality, scalability, and adaptability under uncertainty. Third, these capabilities are integrated into closed-loop architectures where perceptual outputs inform decision-making processes.

The proposed approach is validated through multiple real-world industrial case studies, including petrochemical furnace monitoring, photovoltaic cell manufacturing quality inspection, automotive manufacturing scheduling, and integrated water resource management using a Digital Twin framework. The results demonstrate that connecting perception and decision-making improves robustness, adaptability, and operational relevance compared to conventional disjointed approaches.

Overall, this research provides methodological and empirical evidence that formalizing perception as a key component of scheduling systems is essential to address uncertainty in dynamic environments