Improving Flight Safety with an Automated Analysis of Pilot Performance during Helicopter Approaches to Landing

Minimizing risks during approaches to landing can pose unique challenges to helicopter pilots. Effective training is key to ensuring pilots have the proficiency required to safely perform this maneuver under varying conditions, and timely, objective assessments of performance are required to ensure effective training. This paper presents an analytical technique to autonomously identify, characterize, and objectively evaluate visual approaches to landing at the CH-53E Fleet Replacement Squadron (FRS), where U. S. Marine Corps pilots first learn to fly the heavy-lift helicopter. Data from the aircraft's onboard Integrated Mechanical Diagnostics System (IMDS) are used for the analysis. The initial step in identifying an approach to landing maneuver is the detection of a landing, transition to hover, or "wave-off" (aborted landing) following a continuous descent below 300 feet above ground level. The type of approach to landing is then characterized based on altitude above ground level versus ground speed profiles statistically derived from multiple approaches to landing that have been assessed to represent properly flown maneuvers. There are three principal types of visual approaches to landing taught and practiced at the FRS - normal, obstacle (also known as precision), and running. Each type has a distinct data distribution for altitude above ground level versus ground speed that enables it to be distinguished from other types of approaches to landing. If the automated analysis characterizes the approach to landing as one of the three types of interest, select parameters (e.g., ground speed, engine torque, aircraft attitude, descent rate, and position of flight controls) versus altitude above ground level throughout the maneuver are compared with statistically derived baselines indicative of squadron norms for that type of approach. Results provide an objective assessment of pilot performance for individual maneuvers and are presented in a combined graphic and text report format. Initial validation of the accuracy of the algorithmic technique was conducted using a control dataset comprised of 346 approaches to landing identified and characterized by a subject matter expert, with 97% overall concurrence with maneuver identification and characterization by the automated technique. In addition to presenting details of the automated analytical technique and presentation format, this paper also discusses how the technique can be used for aggregate analyses to monitor or trend performance of individual pilots or squadrons, as well as compare performance among multiple pilots or squadrons.