A digital twin-oriented lightweight modeling approach for gas turbine performance degradation monitoring

A gas turbine (GT) is usually operating under harsh environment such as high temperature, high pressure and high corrosion, resulting in the decreased efficiency and the increased failures. In the past decade predictive maintenance has become a promising tool to reduce maintenance cost, increase productivity and improve the efficiency and reliability of the GT system. However, this method usually requires the high-accuracy predictive analytics based on numerous multi-model data, which is quite challenging for a given machine. This paper presents a universal deep learning operator network as a lightweight modeling approach to integrate data-physics-AI for high-efficiency performance modeling of a heavy-duty gas turbine with limited data, towards the digital twin-oriented predictive maintenance. Data optimization is first conducted to produce the high-quality data for system modeling. The thermodynamic heat balance principle is then applied to model the performance of three subsystems and the entire system. A deep neural network operator model is developed as a lightweight approach to integrate the operating data and physics-based model for performance degradation simulation of the gas turbine in different scenarios. The model is integrated on a GT virtual reality platform to realize the DT-based performance degradation monitoring. The proposed methodology is illustrated with a real-world GT and event data.