Predicting mental workload of using exoskeletons for construction work: a deep learning approach

Afolabi, Adedeji; Yusuf, Anthony; Akanmu, Abiola · 2025 · Crossref

DOI: 10.36680/j.itcon.2025.001

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 unintended cognitive consequences of using active back-support exoskeletons in construction, specifically the increase in mental workload. While exoskeletons reduce physical strain and musculoskeletal disorders, they can impose cognitive burdens through device awareness, adjustment difficulties, and restricted movement, potentially leading to distraction, stress, and reduced safety. The research aims to predict this mental workload in real-time using electroencephalography (EEG) data and deep learning models, offering a solution to the limitations of subjective, time-consuming monitoring methods. The researchers conducted laboratory experiments with eight male participants performing simulated flooring tasks while wearing an active back-support exoskeleton (Cray X). EEG data was collected using a wireless Emotiv EPOC+ headset, capturing signals from frontal, temporal, parietal, and occipital brain regions. The data underwent rigorous preprocessing, including bandpass and notch filtering to remove extrinsic noise, and independent component analysis to eliminate intrinsic artifacts like muscle movement. Two deep learning frameworks were trained to forecast future EEG time steps: a regression-based Long Short-Term Memory (LSTM) network and a hybrid Convolutional Neural Network-LSTM (CNN-LSTM) model. Mental workload was quantified by calculating the ratio of frontal theta band power to parietal-occipital alpha band power derived from the predicted and actual EEG signals. The results demonstrated that the LSTM network significantly outperformed the CNN-LSTM model across all EEG channels. The LSTM achieved an average root mean square error (RMSE) of 0.162 and an R-squared value of 0.939, indicating high predictive accuracy. Specifically, the lowest RMSE values for the LSTM were observed in channels P7 (0.115) and AF3 (0.127), whereas the CNN-LSTM exhibited substantially higher errors, such as an RMSE of 1.069 in channel P8. The comparison between actual and predicted mental workload revealed that the LSTM model captured approximately 75% of the variance in the actual mental workload data. These findings highlight the effectiveness of LSTM networks in modeling the temporal dependencies of EEG data for mental workload prediction during exoskeleton use. The study identifies specific brain channels, particularly in the frontal and parietal regions, as most suitable for assessing cognitive load, which can guide the development of more cost-effective and minimal-channel EEG devices. By enabling real-time monitoring of mental workload, this approach provides construction stakeholders with actionable insights to mitigate cognitive triggers associated with exoskeleton use, thereby enhancing worker well-being, productivity, and safety.

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.

• neuro workload indices