Project start: 1.10.2025.

Rotating machines are the most common type of machinery and are present in all major industrial sectors. The sudden downtime of rotating machines due to faults or malfunctions can lead to significant costs. Rolling bearings constitute a substantial portion of the global bearing market, accounting for approximately 45% of the market share in 2023. Therefore, this project aims to develop an intelligent method for diagnosing rolling bearing failures using machine learning.

The research is broadly divided into two parts. The first part focuses on developing simulation models of rolling bearings (RBs) with various types of damage to determine their characteristic vibration responses. Since RBs are components of rotating structures, it is essential to develop both analytical and numerical models of the rotor to analyze various nonlinear effects and errors. Additionally, the project aims to conduct simple and efficient simulations whose results can be compared with experimental measurements.

The second part of the research focuses on developing diagnostic methods for detecting the type and severity of RB damage using machine learning. Vibration response tests will be conducted on a newly acquired bearing failure simulator, utilizing contact acceleration measurements on the outer ring and non-contact displacement measurements on the inner ring. Initially, we plan to evaluate the effectiveness of feature extraction methods in the time-frequency domain for RBs with embedded localized faults on the inner ring, outer ring, and rolling elements. Empirical Mode Decomposition (EMD), Variational Mode Decomposition (VMD), and Wavelet Transform (WT) will be employed first, followed by their more advanced variants.

For fault diagnosis and classification, we initially plan to use the Random Forest (RF), Support Vector Machine (SVM), and Logistic Regression (LR) algorithms. By the end of the research, additional algorithms published during the course of this study will also be tested.