Augmentor: a cognitive-inspired approach for context-aware adaptive learning with human guidance

Abstract

Automation in complex industrial processes faces significant challenges due to its heavy reliance on human knowledge and tacit expertise. This knowledge is typically acquired through years of on-site training, making the transition to automation difficult. While data-driven approaches offer potential solutions by building automation models from expert observations, they often fall short of capturing deeper human understanding and contextual reasoning. In this paper, we introduce Augmentor, a cognitive-inspired framework for context-aware adaptive learning. The framework bridges reactive learning, which continuously updates models based on incoming data, with context-aware adaptive learning, which incorporates external context and expert knowledge for higher-order reasoning and decision-making. Inspired by theories of consciousness, Augmentor leverages autonomic computing principles through the MAPE-K model and semantic web technologies to create a global workspace for managing knowledge and guiding adaptive learning processes. The approach addresses two key objectives: continuously refining models through reactive learning and enabling experts to inject contextual knowledge to guide the learning process. Human-Gated Data Aggregation is used for data-driven updates, while context-aware adaptive learning ensures the system dynamically adjusts to complex scenarios and evolving industrial conditions. We validate Augmentor through a real-world use case in collaboration with our industrial partner, Mekanisk Service Halden, which specialises in various mechanical services. Experimental results show significant performance improvements across three strategies–static model without relearning, reactive learning, and context-aware adaptive learning–demonstrating the advantages of incorporating context-aware strategies to achieve more robust and adaptive control.