Real -Time EEG-Based Detection of Cognitive Fatigue in Human–Machine Interaction Systems: A Biomedical Engineering Approach

Ayeoribe, Olarewaju Peter; Ayeoribe, Atinuola Elizabeth · 2025 · Crossref

DOI: 10.21203/rs.3.rs-8059037/v1

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study addresses the critical safety challenge of cognitive fatigue in Human–Machine Interaction (HMI) systems, such as aviation, driving, and industrial control. Traditional detection methods relying on behavioral metrics or self-reports are subjective, non-continuous, and often lag behind physiological changes. The authors propose a real-time, EEG-based detection system that integrates biomedical signal processing with deep learning to provide objective, continuous monitoring of mental fatigue. The research aims to bridge the gap between laboratory-based EEG analysis and practical, embedded deployment by developing a low-latency, artifact-robust pipeline capable of operating in dynamic environments. The methodology involved a controlled experiment with ten healthy participants performing a 60-minute sustained-attention driving simulation to induce fatigue. EEG signals were acquired using a 14-channel wireless headset (Emotiv EPOC X) at 256 Hz. Data preprocessing included bandpass filtering (1–40 Hz), notch filtering (50 Hz), and Independent Component Analysis (ICA) to remove ocular and muscular artifacts. Time–frequency features, including Power Spectral Density (PSD), Hjorth parameters, and Theta/Alpha-to-Beta ratios, were extracted from 2-second windows. These features were classified using a hybrid Convolutional Neural Network–Long Short-Term Memory (CNN–LSTM) model, which captures both spatial electrode correlations and temporal fatigue progression. The system was deployed on an NVIDIA Jetson Nano embedded platform to evaluate real-time performance. The CNN–LSTM model achieved a classification accuracy of 94.2%, significantly outperforming traditional Support Vector Machine (87.6%) and Random Forest (89.1%) classifiers. Statistical analysis confirmed significant differences in Theta/Alpha ratios and Beta power between alert and fatigued states. The embedded implementation demonstrated an average processing latency of 420 ± 25 ms, well within the <500 ms threshold required for real-time operational feasibility. The system maintained high data retention (>95%) after artifact removal and received high usability ratings from participants. The findings demonstrate that hybrid deep learning architectures can effectively model the non-stationary nature of EEG signals for accurate fatigue detection. By integrating signal acquisition, preprocessing, and classification into a unified, low-latency biomedical framework, this study validates the potential for continuous cognitive state assessment in safety-critical HMI applications. The system’s ability to operate on embedded hardware with minimal delay supports its use in adaptive automation, where real-time feedback can trigger interventions to mitigate human error caused by cognitive decline.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-nvidia on 2026-08-10; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• drowsiness detection algorithms
• neuro workload indices

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Empirical Findings: physiological data