Real-time cognitive load monitoring of fusion remote maintenance system operators by electroencephalogram

Abstract

In tokamak maintenance, components near the plasma are exposed to radiation and hazardous materials, necessitating remote handling to protect operators from direct exposure and ensure operational integrity. Efficient and reliable human–machine collaboration in these high-stakes environments is essential, as operator errors are intolerable and environmental feedback is limited. This study proposes an electroencephalography (EEG)-based workload classification algorithm for real-time monitoring of remote handling (RH) operators, aiming to enhance operational safety and efficiency by assessing operator workload and facilitating timely adjustments. EEG signals undergo denoising through independent component analysis (ICA) and bandpass filtering to remove artifacts and improve signal quality. Feature extraction and selection identify the top EEG features most indicative of workload, including power spectral density (PSD) metrics and fractal dimensions, which serve as inputs to an optimized artificial neural network (ANN) model. The ANN model, trained on both the STEW dataset and custom data, achieves 82% classification accuracy, generalizing well across varying cognitive load levels in multi-task settings. The model’s performance metrics, including validation accuracy and confusion matrix results, confirm its reliability in discerning workload levels. The findings demonstrate the utility f EEG-based workload monitoring in RH environments, providing a foundation for future improvements in human–machine interaction and operational safety. This research contributes to the advancement of RH systems by enabling real-time workload adjustments based on operator status.