I bring you a weekly bite-sized chunk of the science behind helicopter human factors and CRM in practice, simplifying the complex and distilling a helicopter related study into a summary of less than 500 words.
TITLE:
Machine learning methods for cognitive load analysis and classification in aviation.
WHAT?
A review paper investigating how artificial intelligence (AI), (specifically machine learning), is being used to measure and detect cognitive workload in aviation. It looks at what types of data are most useful (e.g. heart rate, brain activity, eye tracking), which AI techniques work best, and how close the industry is to being able to monitor workload reliably in real time.
WHERE?
University of New South Wales (Australia) and CAE, examining international studies across aviation domains including pilot training, flight simulation, air traffic control, and uncrewed systems.
WHEN?
Published 2026, covering research from 2010 onwards.
WHY?
High cognitive workload is a known contributor to errors, degraded decision-making, reduced situational awareness, and task shedding. Traditionally, workload assessment has relied on subjective self-ratings, simulator observation, or performance metrics after the fact.
Machine learning offers the possibility of objective, real-time workload detection, which could support adaptive cockpits, improved training design, better fatigue management, and more targeted safety interventions.
HOW?
The authors reviewed 43 relevant studies from an initial pool of nearly 2,000 papers. These studies measured workload using:
- Physiological signals (heart rate variability, ECG, EEG brain activity, skin conductance)
- Eye-tracking data (pupil size, gaze patterns, fixation behaviour)
- Or a combination of both
Workload levels were usually linked to task difficulty, simulator phases, or structured workload tasks, sometimes supported by NASA-TLX ratings (a scale for participants to report subjectively experienced workload).
Various AI models were tested to help classify workload. Some were relatively simple classification tools, while others were more advanced deep learning systems capable of identifying complex patterns over time.
FINDINGS:
Key findings include:
- Very few studies validated their systems in real flight conditions.
- Systems using a single data source (e.g., heart rate alone) showed wide accuracy ranges (around 40%–96%).
- Systems combining multiple data sources (e.g., heart rate + EEG + eye tracking) achieved much higher accuracy—up to 98–99%.
- More advanced AI systems that analyse time-based patterns performed better than simpler classification tools.
- Brain activity (EEG) and heart rate variability were strong predictors of workload, but combining them with eye tracking produced the most reliable results.
- Most studies were conducted in simulator environments with small participant numbers.
SO WHAT?
This review shows that AI-based workload monitoring is technically feasible and increasingly accurate, particularly when multiple physiological and behavioural signals are combined.
However, most evidence comes from controlled environments, not operational flight. Before these systems can be used in cockpits, challenges remain around reliability in real-world conditions, data privacy, integration into safety systems, and ensuring the outputs are interpretable in a safety-critical environment.
The study suggests that with AI assistance, real-time workload monitoring is moving from research toward operational possibility. In the future, this could support adaptive training, identify high-risk flight phases, assist fatigue management, and enhance safety management systems. But further validation in live aviation settings is still needed before we’re likely to see widespread implementation.
REFERENCE:
Molloy, O., Eves, G., Vahidnia, S., & Shahin, M. (2026). Machine learning methods for cognitive load analysis and classification in aviation: A systematic review. International Journal of Human–Computer Interaction. https://doi.org/10.1080/10447318.2026.2632151
competentaviators.com 