Modeling of performance and thermodynamic metrics of a conceptual turboprop engine by comparing different machine learning approaches

Abstract

In this study, several performance and thermodynamics metrics of a conceptual turboprop engine (C-TPE) were computed at 50 (fifty) power settings. Based on these computations, these parameters were predicted by employing artificial neural networks (ANN) and long-short term memory (LSTM) approaches. The obtained parametrically data were subjected to preprocessing for normalization. After determining model inputs to estimate the engine outputs, these data were introduced to ANN and LSTM. Then, the findings by two methods were compared with each other. In this context, fuel flow, air mass flow, exhaust velocity, compressor pressure ratio, turbine outlet temperature, and revolutions per minute (RPM) were selected as model inputs to estimate several metrics such as net thrust, specific fuel consumption (SFC), overall efficiency, exergy efficiency, and environmental effect factor (EEF) regarding the C-TPE. For modeling of the mentioned metrics, 80% and 20% of the data were used for training and testing, respectively. According to the findings of energetic and exergetic computations, SFC value of the C-TPE increases from 16.205 to 17.737 g/ kNs whereas its overall efficiency varies from 23.63% to 22.31% owing to decrement in RPM. Moreover, exergy efficiency of C-TPE remains between 23.93% and 26.2% whilst its EFF value resides between 2.66 and 3.05 throughout power settings. Finally, the modeling results indicate that coefficient of determination (R 2 ) regarding overall efficiency of C-TPE was found as 0.951867 with respect to the ANN model whereas it was calculated as 0.999906 with the proposed LSTM. It could be deduced that since LSTM is capable of providing lower the model error than the ANN for the same set of data, this method could be implemented to predict thermodynamics metrics of different kinds of gas turbine engine where obtaining data is limited.

Data-Driven Exhaust Gas Temperature Baseline Predictions for Aeroengine Based on Machine Learning Algorithms

“…Yu Y et al [18] established the aeroengine EGT baseline model by combining the kernel principal component analysis (KPCA) with the deep trust network (DBN). Omer Osman Dursun et al [19] predicted EGT of the conceptual turboprop engine (C-TPE) by using artificial neural networks (ANN) and short and long-term memory (LSTM) methods. Although the neural network has a strong nonlinear fitting ability, it is prone to problems of divergence and low accuracy when the training sample set is small.…”