Investigating bearing and gear vibrations with a Micro-Electro-Mechanical Systems (MEMS) and machine learning approach

• A MEMS setup comprised of a Raspberry Pi 4B+ module, a NuceloF401RET6 MCU, an OLED screen, and an Adxl1002z accelerometer is developed to acquire vibration data at the desired sampling frequency. • A Random Forest machine learning model is developed to classify the faults and to determine the MEMS performance using extracted vibrational features. • A detailed signal analysis evaluates MEMS performance and investigates the impact of bearing and gear vibration interactions. Bearings and gears are the pivotal components of mechanical systems and are prone to faults that can impact the system's overall performance. These components' condition monitoring and fault diagnosis are vital for maintaining system reliability and efficiency. In this research, a MEMS setup is initially developed, comprising a Raspberry Pi 4B+ CPU module, a NucleoF401RET6 MCU, an OLED screen, and an Adxl1002z accelerometer for acquiring vibration signals at the desired sampling frequency stored in the CPU memory. Further, an RF model is also developed to classify different types of faults based on features extracted from the acquired vibration data. The model evaluates the precision and reliability of the MEMS setup in capturing and classifying vibration signals. A detailed signal analysis is also conducted to determine the performance of the developed MEMS setup and to investigate the effect of bearing vibration signature due to gear fault and vice versa. The results indicate that bearing faults cause irregularities in the shaft's rotational speed, leading to modulation of the gear mesh frequency ( gmf ) of gears mounted on the affected shaft. Conversely, gear faults disrupt the shaft's rotational motion, imposing excessive loads on shaft-supported bearings. These disruptions result in distinct vibration patterns characterised by increased harmonics and side bands within the bearing frequency range. The RF model effectively identifies and classifies faults with high accuracy by leveraging its ability to prioritise the most significant vibrational features, resulting in improved predictive performance and robustness.