Data-Driven Approaches for Thermocouple Conduction and Inertia Correction to Improve Engine Digital Twin Modeling

Abstract Digital engine twins duplicate the real existing, complex system by a digitally created construct representing its time-dependent characteristics. Based on real-time measured data, they provide information about the system state, which are not accessible by a data analysis. This makes accurate transient temperature measurements critical to digital engine twins. Thermocouples represent an established means of engine temperature measurement. However, they are affected by thermal inertia and conductive heat transfer, which introduce a bias and a time-delay of the sensor response. There are different physical phenomena leading to bias and time delay. Discussion of these phenomena makes it obvious that the transient reading of a thermocouple highly depends on its installation in the engine and the associated thermal environment. It is demonstrated that these distortions obscure the time series of measurements and complicate the distinction between nominal and non-nominal engine state. The time delay can be corrected a priori via a first-order time lag system based on basic physical properties. The potential of calibrating for the complex gas path flow conditions affecting the time lag is estimated using a physics-informed neural network. Matching the temperature readings completely to the model output indicates the amount of transient calibration data required to capture all effects due to the installation of the thermocouple in the engine structure.