Flight Data Acquisition Without Sensors: Futuristic Data Acquisition Systems Embedded with Machine Learning and Big Data Analytics

Wg Cdr RR Senthil Kumar, Flight Test Engineer, ASTE, Indian Air Force

Abstract

A typical Fighter Jet development cycle consists of production of Technology Demonstrator (TD) Aircraft, Prototype (PT) Aircraft, Limited Series Production (LSP) aircraft and Series Production (SP) Aircraft. The general time period from first flight to Final Operational Clearance (FOC) is around 8-10 years for fighter jet program+$ . The Light Combat Aircraft has flown almost 3200 hours towards D&D in 12 years from first flight in 2001 to IOC in 2013^. The PT and LSP aircraft are always extensively instrumented towards performance validation and certification requirements. The sensors exclusively required for Health and Usage Monitoring (HUM) of airframe structures such as strain gauges, vibration sensors, force sensors etc are generally provided only for Flight Test purposes on the PT and LSP aircraft. Such exclusive instrumentation are not part of the standard SP aircraft. An attempt has been made by the author to explore the feasibility to build Machine Learning (ML) models using the abundant PT & LSP flight test data. The ML models developed using PT aircraft data can be used for predicting the data of sensors which are not installed on the SP aircraft. The concept was implemented on a smaller scale on a jet Trainer aircraft at Aircraft and Systems Testing Establishment (ASTE). Two jet trainer aircraft were instrumented with Attitude and Heading Reference System (AHRS), Angle of Attack (AOA) & Angle of Sideslip (AOS) vanes. The flight test data from one aircraft (model aircraft) was used for builing a ML model for deriving the AOA & AOS data from AHRS data. The ML model built with data from the first aircraft was used to predict the AOA& AOS using only the AHRS data of second aircraft (test aircraft). The ML derived AOA & AOS data of test aircraft was validated against the vane data during multiple test flights. Thus the Proof of Concept (POC) was established wherein ML techniques could be used to derive data of sensors which were not fitted on the aircraft. The similar concept could be extended to much larger scale for development of intelligent Data Acquistion Systems with embedded Machine Learning applications for deriving data without sensors. The Prototype aircraft data can be used for developing the ML models. Such models can be provided as part of the Data Acquistion Systems such as Flight Data Recorders / Mission Plannig Debrief Systems. These systems can derive the parameters of interest based on the ML model application over the available sensor data, thereby paving way for acquiring data sans sensors.