Ship Helicopter Operations: A Machine Learning Approach for Workload Prediction Using Linear Discriminant Analysis

This paper demonstrates the training, optimisation, and predictive capabilities of Machine Learning (ML) for helicopter-ship certification. The work focuses on the development of a Linear Discriminant Analysis (LDA) model, trained specifically on pilot control activity data recorded during the hover phase of a recovery to a ship, to determine an operational boundary driven by pilot workload. The certification process currently relies heavily on embarked trials and the subjective workload assessment of test pilots. Modelling and Simulation (M&S), however, offers a potentially more efficient approach to addressing the high costs, resource-intensive nature, and inherent dangers associated with traditional clearance methods. By providing a relatively large amount of data for analysis, this approach creates an opportunity to bridge the gap between subjective and objective measures, enabling the prediction of workload limitations. An LDA model was trained using cross-validation on pilot control activity data and optimised through the inclusion of a penalty factor to reduce overfitting. Throughout the training process, the model demonstrated good performance, effectively distinguishing between high and low workload conditions based on pilot control activity data. When tested on unseen data, the model accurately predicted the Ship-Helicopter Operating Limit (SHOL) boundary for most cases. These results support the application of ML in the helicopter-ship certification process and demonstrate the model's ability to identify correlations within high-dimensional datasets, offering a more data-driven and objective approach to determining workload and clearance boundaries.