Evaluating Flight Crew Performance by a Bayesian Network Model.

Area of Science:

• Aviation Safety
• Human Factors Research
• Artificial Intelligence in Aviation

Background:

• Evaluating flight crew performance is crucial for aviation safety.
• Quantitative behavioral data is often unavailable for crew performance assessments.
• Existing methods may not fully utilize multidisciplinary information sources.

Purpose of the Study:

• To introduce a Bayesian Network (BN) for robust flight crew performance evaluation.
• To enable the use of both objective and subjective data, especially when behavioral data is sparse.
• To provide data support for human error management interventions in aviation.

Main Methods:

• Analysis of 484 aviation accidents caused by human factors to identify causal factors.
• Construction of a 'Flight Crew Performance Model' using Bayesian Network.
• Utilizing the Delphi technique for subjective data collection to supplement objective data.
• Eliciting conditional probabilities with the leaky noisy MAX model.
• Applying BN inference for probability prediction and probabilistic diagnosis.

Main Results:

• A functional Bayesian Network model for flight crew performance evaluation was developed.
• The model successfully integrates multidisciplinary data sources, including subjective inputs.
• Inference methods provided insights into performance prediction and diagnostic analysis.
• The study demonstrates the feasibility of using BN with sparse behavioral data.

Conclusions:

• Bayesian Networks offer a powerful tool for flight crew performance evaluation, especially with limited data.
• The developed model can support data-driven interventions for human error management in aviation.
• Integrating subjective and objective data via BN enhances the comprehensiveness of performance assessment.
• This approach contributes to improving overall aviation safety by addressing human factors.