Learning based model correction and data imputation algorithms for rotorcraft systems

Thumbnail Image
Makkar, Gaurav
Issue Date
Electronic thesis
Mechanical engineering
Research Projects
Organizational Units
Journal Issue
Alternative Title
The future of vertical takeoff and landing (VTOL) vehicles encompasses numerous thrilling and innovative designs, including electric multirotor vehicles and high-speed compound helicopters. Electric multirotor vehicles are expected to be utilized for various purposes such as human transportation ("air taxis"), package delivery, and surveillance, among others. On the other hand, the development of high-speed compound helicopters aims to cater to military missions, including aerial scout missions such as assault and reconnaissance, as well as troop transport. The lack of high-fidelity flight simulation models for rotorcraft poses a significant challenge in understanding complex systems and limits the efficiency and cost-effectiveness of rotorcraft development and evaluation processes. Secondly, the impact of sensor data loss on autonomous rotorcraft operations raises concerns about compromised behavior and safety mid-flight. To tackle these problems, this research presents a novel learning-based framework for model correction and data imputation tailored to different rotorcraft platforms. Leveraging machine learning approaches that have shown promise in model correction, multifidelity modeling, and data imputation, this framework aims to enhance the development of accurate simulation models for rotorcraft and provide solutions for sensor data loss. By employing machine learning techniques, the proposed solution seeks to improve the fidelity of simulation models, reduce the reliance on costly experiments or high-fidelity simulations, and ensure reliable decision-making and conditional assessment during autonomous flights. This body of work explores the development of a model correction algorithm for a low-order model of a compound helicopter with targeted high-fidelity data at specific areas of interest using machine learning approaches. The method is applied to a low-fidelity comprehensive trim analysis of a compound helicopter with three degrees of control redundancy: main rotor speed, auxiliary thrust, and stabilator setting. The final low-fidelity correction model applies small changes to the power requirement and main rotor trim control predictions to match the high-fidelity data. To reduce the computational time, and labor cost of querying the high-fidelity data, the algorithm prioritizes data acquisition by iteratively selecting the data where the error is expected to exceed the model tolerance by the greatest margin. Next, this work presents a novel methodology for identifying nonlinear corrections to improve the accuracy of a physics-based simulation model of a hexacopter using flight test data. The nonlinear corrections are identified by analyzing correlations between different flight variables and utilizing a filtered dataset with a high normalized correlation. The regularized version of partial least squares is applied for identifying the correction terms. For imputing missing sensor data in multicopters, which enables enhanced safety and reliability, a deep learning-based framework is developed by leveraging the multitude of sensors on these aerial vehicles. The proposed approach is based on two deep learning techniques, namely Autoencoders (AE) and Long Short-Term Memory (LSTM) networks. The effectiveness of this approach is evaluated using flight test data from a 2.5 kg hexacopter, and three different scenarios of missing data are considered.To validate the performance of the proposed approach, it is compared against two commonly used imputation techniques: k-Nearest Neighbor (KNN) imputation and Random Forest imputation.
School of Engineering
Full Citation
Rensselaer Polytechnic Institute, Troy, NY
Terms of Use
PubMed ID