Embedded Files
The project focuses on developing a methodology for monitoring bearing conditions. It combines Poincare Maps, Fast Fourier Transform (FFT), and Convolutional Neural Networks (CNN) to diagnose faults and predict the Remaining Useful Life (RUL) of bearings.
During my Bachelors in Mechanical Engineering, I approached Dr. Singru to work on a research project, aiming to gain my first hands-on experience in research. I, along with Advait Mulay, Venugopalan Iyengar and Aniruddha Nayak worked on this project for more than a year. We initially took courses to understand how bearings and vibration work and understand the mathematics behind it. After that, we used datasets from the IMS (University of Cincinnati) and FEMTO-ST Institute, to develop a methodology for monitoring the bearing conditions. Our efforts culminated in the publication of a research paper based on our findings.
Key Techniques and Methods:
Poincare Maps: Used to analyze the non-linear, chaotic behavior of bearing vibrations. Poincare Maps are effective in detecting the non-periodic motion of spin elements, which is not easily identifiable through traditional frequency spectrum analysis. They provide a visual representation of vibration signatures, helping to pinpoint defect locations .
Fast Fourier Transform (FFT): Converts time-domain vibration signals into the frequency domain, identifying characteristic frequencies related to different bearing faults such as inner race defects, outer race defects, and roller element defects.
Convolutional Neural Networks (CNN): Utilizes Continuous Wavelet Transform (CWT) images of vibration signals to predict RUL. The CNN model processes these images, achieving high accuracy in life prediction.
Experimental Setup and Data:
Our study uses datasets from the IMS (University of Cincinnati) and FEMTO-ST Institute, available at the NASA Prognostics Center of Excellence. These datasets include detailed vibration data collected under controlled experimental conditions.
Research Highlights:
Poincare Maps: Poincare Maps generated from vibration accelerometer data reliably diagnose the approximate location and severity of defects. They provide a relative reference to the shaft position.
FFT Analysis: The FFT Analysis identifies characteristic frequencies and changes over time, correlating with the defects diagnosed through Poincare Maps.
CNNs: CNNs predict the fraction of RUL(Remaining Useful Life) with high accuracy. The study, using the FEMTO dataset, shows that the methodology achieves a 94.00% median accuracy and a 93.14% mean accuracy on test sets.
Select Images: Here are some selected image highlights from the project
FFT Analysis - Beginning of Test
FFT Analysis - End of Life
Poincare Map of one of the IMS Dataset Bearings
CNN Model architecture
Conclusion:
The integration of Poincare Maps, FFT, and CNNs constitutes a robust and precise method for bearing condition monitoring. This hybrid approach not only identifies faults with high precision but also provides reliable RUL predictions, enhancing maintenance practices and reducing operational downtime in rotating machinery.
Google Sites
Report abuse