Visual Mental Workload Assessment from EEG in Manual Assembly Task

Pušica, Miloš; Caiazzo, Carlo; Djapan, Marko; Savkovi\'c, Marija; Leva, Maria Chiara · 2023 · Crossref

DOI: 10.3850/978-981-18-8071-1_p667-cd

archive: archived pipeline: cataloged

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

Summary

This paper addresses the challenge of developing task-independent mental workload (MWL) estimators using electroencephalography (EEG). While machine learning (ML) models have shown promise in MWL assessment, they typically perform well only on the specific task used for training, limiting their utility in real-world environments where tasks vary. The authors argue that tasks engaging similar cognitive processes, such as visual cognition, may allow for cross-task generalization. To advance this goal, the study designs an experiment to collect EEG data during a manual assembly task that specifically isolates visual cognitive demands, aiming to create a dataset for training a task-domain-specific ML estimator. The experimental design involved 32 participants performing a manual assembly task in a controlled laboratory environment simulating an assembly line. Participants were required to connect holes on a plexiglass plate with wires and toggle switches according to visual instructions displayed on a screen. The task included two levels of visual complexity: "easy" schemes, which were straightforward diagrams, and "hard" schemes, which were photos of assembled items taken from non-ideal angles with tangled wires, sometimes rotated to increase difficulty. Time limits were set at 60 seconds for easy schemes and 90 seconds for hard schemes. EEG data was recorded using a wireless 24-channel cap with a sampling rate of 250 Hz, alongside video recordings and manual logs of completion status. The experiment included a baseline idle period and a total of 150 schemes (90 easy, 60 hard) per participant, with a mid-task break. Results from the initial analysis of ten subjects indicate that the experimental design successfully induced different levels of MWL. Participants completed 36.4% of easy schemes within the time limit, compared to only 3.6% of hard schemes, confirming that the harder tasks required more time and effort. EEG signals were pre-processed using band-pass filtering (1-40 Hz) and artifact removal via Independent Component Analysis. The Mental Workload Index (MWI), calculated as the ratio of frontal theta to parietal alpha power, was significantly higher for hard schemes in the majority of subjects, supporting the hypothesis that increased visual complexity leads to higher cognitive burden. Subjective questionnaires confirmed that participants found the high-complexity schemes more challenging. The significance of this work lies in its contribution to the development of cross-task MWL estimation within the domain of visual cognition. By isolating visual complexity as the primary variable, the study provides a foundation for training neural networks that may generalize across similar visual tasks. Future work involves training a neural network on this dataset to classify EEG segments by complexity level and evaluating its performance on an independent external dataset. This approach aims to overcome the limitations of current task-specific models, potentially enabling more flexible and practical MWL monitoring in diverse industrial and safety-critical settings.

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 qwen3.8-27b-gittensor on 2026-08-23; verification: pending re-verification.

Topics

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

• neuro workload indices
• workload measurement
• mental demand

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