AI-Driven Assistant for Critical Infrastructures Human Operators: an Online Cognitive Performance and Stress Monitoring Platform Based on ECG Morphology and Heart Rate Variability

Stress can severely impair cognitive performance in safety critical occupations, motivating the use of artificial intelligence (AI) to support human operators. However, current systems seldom infer operators’ internal states implicitly, limiting the quality of human–machine interaction (HCI). This work introduces the Human Assessment Model (HAM), a human-in-the-loop framework that continuously estimates workers’ psychophysiological condition to enable AI-driven adaptive HCI. HAM was evaluated with 11 air traffic controllers. A laboratory protocol was designed to elicit stress and quantify its impact on individual cognitive performance (CP) while electrocardiogram (ECG) signals were recorded. Personalized models were then trained using ECG wave morphology and heart rate variability features to predict CP and detect stress in each participant. The AI outputs were mapped onto interpretable descriptors of human states, and feature importance was analysed to enhance explainability. Finally, real-time deployment of HAM was demonstrated. CP estimation yielded moderate yet promising accuracy (avgRMSE = 0.15), and stress was reliably identified (avgF1 = 0.95). Notably, an ECG morphology feature—the QRS complex duration—emerged as the most influential predictor. These findings support HAM as a foundation for monitoring tools that promote occupational health and inform adaptive HCI and decision support systems in high-risk work environments.