Neural Network Application to Rotordynamic Predictions of an Overhung Flexible Rotor

This study focuses on pioneering regressive neural network models to predict lateral dynamics in a centrifugal compressor stage validation test bench. This tool is positioned as a critical resource to be used during the design of the centrifugal stage test, enabling test bench personnel to understand the rotordynamic behavior in terms of critical speeds and vibration amplitudes within the desired operating range. Addressing the demand for swift and manageable solutions for repeatable configurations within defined design parameters motivates this work. While Finite Element Method (FEM) analysis remains valuable, its time-intensive nature, particularly in model definition, drives the search for more time-efficient methods. Machine learning stands as an innovative, streamlined alternative for predicting lateral dynamics, differing from traditional approaches. The adoption of machine learning-based regressive neural networks marks a significant leap in analyzing centrifugal compressor lateral dynamics. This innovation not only boosts operational efficiency but also enhances accessibility and cost-effectiveness in the process.