Machine Learning-Based Digital Twin Model of Gas Turbines and its Application to Vibration Faults

Vibration issues in gas turbines can lead to miscellaneous failures i.e. Rotor imbalance, bearing damage, misalignment of the shaft, blade fatigue and cracking, seal failure, heat generation, coupling failure, and damage to auxiliary systems among others. All these issues can be addressed by utilizing Predictive maintenance to continuously monitor the health condition of gas turbines and by utilizing a Digital twins performance model. An emerging machine learning algorithm i.e. Temporal Multimodal Multivariate Learning (TMML) is utilized to perform Predictive Maintenance of the gas turbine. The algorithm handles massive volumes of complicated and diverse data, and it also captures the interdependencies between variables throughout various time periods and data streams. This research presents an early warning strategy for vibration based on a digital twin model that overcomes the aforementioned challenges from three angles. First, the measured data and mechanism model are merged for performance modeling and bring the real gas turbine closer to the simulation effects. Secondly, the relative precision of modeling parameters is controlled by developing a notion. The modeling cycle of the existing model is reduced by adding an error module. Third, Digital twin concept is based on early warning, and it uses kernel density estimation to let the warning feature's vibration parameters self-learn the alarm threshold. Using the proposed method, the actual measured data of the gas turbine is examined, and the findings validate that the newly built digital model is more precise and effective. Findings indicate that the digital twin model performs better.