Development of an Artificial Intelligence based Framework for Flight Data Analysis

Abstract

Modern aircraft are embedded with numerous sensors that record and store data from various systems of the aircraft. With the increase in the data volume, stakeholders have understood the need for advanced data analysis methods to obtain useful infor- mation from this data. The conventional mechanism involves exceedance detection that is based on OEM provided limits for selected important parameters. The task of ana- lyzing individual parameters manually is arduous and there has been several attempts in recent times to automate the process utilizing state of the art methods and algo- rithms.Recently, deep learning algorithms have become a go to approach for flight data mining tasks due to their ability to model complex high dimensional data. However, in most cases, the work focuses on a specific action where novel techniques are proposed for various tasks such as classification, clustering, and anomaly detection. Most of these works rely on domain experts’ interpretation after the final results. This work provides a two-pronged solution for flight data fault detection: (i) using an exceedance detec- tion software tool, and (ii) artificial intelligence based flight data anomaly detection framework. The anomaly detection framework provides an end to end solution for iden- tifying an anomalous landing where there is no need for domain experts’ interpretation at each step, rather the framework incorporates this step during its development . Thus, this work is unique and tries to combine several fields of studies in modern flight data analysis. It takes advantage of latest Transformer model for supervised classification of anomalous flights and then applies an integrated method of change point detection and historical behavior of normal data with an infusion of domain knowledge to provide a final interpret-able and actionable results. The results can be utilized for flight safety management as well as maintenance management systems in an efficient manner. This study is focused on providing an applicable solution for flight safety and maintenance management that can be fine-tuned to any sub-field of aviation industry.