Session: 20-04: Optimization Through Digital Tools

Paper Number: 151808

Gas Turbine Diagnostics by Means of Convolutional Neural Networks Fed With Time Series Data Encoded As Images 

Since gas turbine performance degradation is inevitable, gas turbine diagnostics is commonly performed to identify anomalies in time series. However, anomalies are often not directly observable and thus they must be indirectly diagnosed by employing data-driven models.

This paper proposes to exploit the outstanding capabilities of Convolutional Neural Networks (CNNs), by feeding them with images obtained from multivariate time series data, transformed by means of Gramian Angular Summation Field (GASF) method and Markov Transition Field (MTF) method. Two CNNs are investigated and compared to a Temporal Convolution Network (TCN) fed with time series data.

The novel methodology is applied to 150 days of operation, taken from the ten GTs and spread over multiple years. Twenty measured variables are considered, namely compressor discharge temperature, compressor discharge pressure, fuel mass flow rate, power output, and sixteen exhaust gas temperatures.

The simulated fault is a spike, which is implanted by considering nine combinations of fault parameters, i.e., three values of fault magnitude and three numbers of implanted spikes in each time series. The spikes are supposed to occur in two out of the twenty available variables, i.e., compressor discharge temperature and compressor discharge pressure.

The analyses carried out in this paper unequivocally demonstrate that both CNNs fed with images achieve significantly higher classification accuracy than TCN models fed with raw time series data. Moreover, the MTF method always proves more robust than GASF method, and also allows CNNs to achieve higher accuracy values.

Presenting Author: Mauro Venturini Università degli Studi di Ferrara

Presenting Author Biography: Mauro Venturini graduated in Nuclear Engineering with honor at the University of Bologna (Italy) in 1997.
He is currently an Associate Professor of “Energy Systems and Power Generation” at the Department of Engineering of the University of Ferrara (Italy).
Mauro Venturini is an Associate Editor for the ASME Journal of Engineering for Gas Turbines and Power.
He served as OGA Committee Chair for the term 2011-2013 and once again for the term 2021-2023.

Authors:

Enzo Losi Università degli Studi di Ferrara
Mauro Venturini Università degli Studi di Ferrara
Lucrezia Manservigi Università degli Studi di Ferrara
Giovanni Bechini Siemens Energy