An emerging technology for automating Unmanned aircraft is digitally twining the system, and employing AI-based data-driven solutions. Digital Twin (DT) enables real-time information flow between physical assets and a virtual model, creating a fully autonomous and resilient transport system. A key challenge in DT as a Service (DTaaS) is the lack of Real-world data for training algorithms and verifying DT functionality. This article focuses on data augmentation using Real-world Similar Synthetic Data Generation (RSSDG) to facilitate DT development in the absence of training data for Machine Learning (ML) algorithms. The main focus is on the noise generation step of the RSSDG for a common Hybrid turbo-shaft engine because there is a significant gap in transforming synthetic data to Real-world similar data. Therefore we generate noise through 6 different noise generation algorithms before Rolling Linear Regression and Filtering the noisy predictions through Kalman Filter. The primary objective is to investigate the sensitivity of the RSSDG process concerning the algorithm that is used for noise generation. The study’s results support the potential capacity of RSSDG for digitally twining the engine in a Real-world operational lifecycle. However, noise generation through Weibull and Von Mises distribution showed low efficiency in general. In the case of Normal Distribution, for both thermal and hybrid models, the corresponding DT model has shown high efficiency in noise filtration and a certain amount of predictions with a lower error rate on all engine parameters, except the engine torque; however, Students-T, Laplace, and log-normal show better performance for engine torque RSSDG.
Exploring Synthetic Noise Algorithms for Real-World Similar Data Generation: A Case Study on Digitally Twining Hybrid Turbo-Shaft Engines in UAV/UAS Applications
Aghazadeh Ardebili, Ali
Primo
;Longo, AntonellaSupervision
;Ficarella, AntonioProject Administration
;
2023-01-01
Abstract
An emerging technology for automating Unmanned aircraft is digitally twining the system, and employing AI-based data-driven solutions. Digital Twin (DT) enables real-time information flow between physical assets and a virtual model, creating a fully autonomous and resilient transport system. A key challenge in DT as a Service (DTaaS) is the lack of Real-world data for training algorithms and verifying DT functionality. This article focuses on data augmentation using Real-world Similar Synthetic Data Generation (RSSDG) to facilitate DT development in the absence of training data for Machine Learning (ML) algorithms. The main focus is on the noise generation step of the RSSDG for a common Hybrid turbo-shaft engine because there is a significant gap in transforming synthetic data to Real-world similar data. Therefore we generate noise through 6 different noise generation algorithms before Rolling Linear Regression and Filtering the noisy predictions through Kalman Filter. The primary objective is to investigate the sensitivity of the RSSDG process concerning the algorithm that is used for noise generation. The study’s results support the potential capacity of RSSDG for digitally twining the engine in a Real-world operational lifecycle. However, noise generation through Weibull and Von Mises distribution showed low efficiency in general. In the case of Normal Distribution, for both thermal and hybrid models, the corresponding DT model has shown high efficiency in noise filtration and a certain amount of predictions with a lower error rate on all engine parameters, except the engine torque; however, Students-T, Laplace, and log-normal show better performance for engine torque RSSDG.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
