Evaluation of Non-Stationary Signal Processing and Nonlinearity Effects in Hydrodynamic Bearings

• Grant number: 25/14412-6
• Support Opportunities: Scholarships abroad - Research Internship - Doctorate (Direct)
• Start date: March 01, 2026
• End date: February 28, 2027
• Field of knowledge: Engineering - Mechanical Engineering - Mechanics of Solids
• Principal Investigator: Gregory Bregion Daniel
• Grantee: Matheus Victor Inacio
• Supervisor: Athanasios Chasalevris
• Host Institution: Faculdade de Engenharia Mecânica (FEM). Universidade Estadual de Campinas (UNICAMP). Campinas , SP, Brazil
• Institution abroad: National Technical University Of Athens, Zografou Campus, Greece
• Associated to the scholarship: 22/12565-1 - Fault identification in hydrodynamic bearings using machine learning, BP.DD

Abstract

Hydrodynamic journal bearings play a key role in supporting rotors, dampingvibrations, and minimizing friction in rotating machinery, being that their performanceimpacts equipment stability and lifespan. Since faults can lead to serious issues duringthe operational condition, early fault detection is crucial to avoid unexpected downtimeand high repair costs. Thus, several researches focus on advanced monitoring anddiagnosis to improve system reliability. Within this context, this project BEPE (Bolsa deEstágio em Pesquisa no Exterior) aims to develop a new approach for identifyinghydrodynamic bearings that operate with faults related to non-circular profiles. Theproposed approach involves training a neural network model that is fed by the features ofthe dynamic response of the rotor, enabling it to map the relationship between thefeatures and the non-circular bearing profiles. At this moment, several features wereevaluated by applying the Fast Fourier Transform (FFT) to the time-domain signals,obtained promise results under steady-state rotational condition. However, to ensure itseffectiveness in practical applications, this approach must be made more robust andreliable, particularly in machines under transient conditions. Therefore, in this new phaseof the project, the use of new signal processing techniques is proposed to ensure theextraction of meaningful features that effectively capture the fault signatures of noncircular bearing profiles. Moreover, the study will address the effects of nonlinearity inbearing models, as developing accurate and representative models is essential forgenerating reliable datasets that reflect realistic fault conditions. In the originally proposedapproach, the neural network is trained using simulated data generated from HD andTHD bearing models, in which the hydrodynamic forces are linearized around the rotor'sequilibrium position. However, the rotor may operate under extreme conditions, such ashigh eccentricity, high vibrational orbits, or close critical speeds, what tends to increasethe nonlinear effects of the hydrodynamic bearing on the dynamic response of the rotor,justifying a better evaluation of these conditions and effects on the proposed approachfor identification of bearings with non-circular profiles. Therefore, this BEPE project aimsto deepen the development of the proposed approach for identifying bearings with noncircular profiles, in order to explore new relevant features through other processingtechniques and investigate the influence of nonlinear effects of hydrodynamic forces inthe identification process.