Human control mode enables accurate real-time risk warning in human–machine systems

Abstract

Data-driven risk analysis serves as an essential approach to risk mitigation in human–machine systems. Presently, risk management rooted in data often depends on labels extracted from risk outcomes, accentuating a causative risk management paradigm. However, these labels frequently fall short in capturing the dynamic evolution of risks in real-time, especially accounting for the impact of human intervention on risk dissemination. In striving for greater precision in real-time risk prediction within human–machine systems, human control is identified as a pivotal factor in shaping risk progression. A precise warning model is devised based on human control patterns, discerned through clustering control data focusing on “timeliness,” “stability,” and “coordination.” This methodology facilitates the development of machine learning-driven warning models. The viability of the proposed approach is substantiated through a case study involving aircraft landing mishaps. This research furnishes a conceptual framework and procedural guidelines to propel risk analysis within human–machine systems, with an emphasis on human-centric risk warnings across diverse industrial contexts.