Assessing Aircraft Trim Conditions for Anomaly Detection in Flight Test Data

Understanding and analyzing trim conditions is essential for extracting the relevant information required to develop accurate and reliable flight simulators. In this context, this paper proposes a method to identify improper pilot maneuvers by focusing on incorrect trim conditions. Trim conditions play a key role in replicating realistic flight dynamics and pilot control inputs, ensuring that flight simulators provide an authentic training and testing environment. The study assesses aircraft trim quality using a diverse dataset collected from various onboard sensors, including accelerometers, gyroscopes, and many others. The data, consisting of time series, requires the development of a statistical algorithm capable of transforming the time series into exploitable point datasets. This analysis leverages artificial intelligence tools, including classification algorithms such as Isolation Forest, which detects anomalies using binary trees. In addition, two statistical methodologies from the literature are used: a piecewise linear regression algorithm that identifies slope variations and trend changes using double sliding windows and the least squares method, and an association rule that merges and verifies small error-detecting segments. The primary objective is to compare various methodologies to determine the most effective approach for detecting anomalies and atypical flight test data. The results from this comparison will guide the development of an optimal model for training a neural network dedicated to identifying improper trim conditions. This research offers promising prospects for improving aircraft flight safety and performance.