Closing the gap between research and development: condition monitoring in rotating machinery

SKF is a worldwide producer of bearings (in Dutch: kogellagers) which are present in many machinery types, such as wind turbines, cars and trains. During this talk we will give an introduction in vibration signal analysis to diagnose the state of this rotating equipment. By estimating the health condition of a machine using accelerometer signals, we can decrease the amount of unexpected failures and thus optimize maintenance costs. A key ingredient in these types of equipment are roller-element bearings which facilitate a rotating movement by having multiple rolling elements placed between two rings. By measuring the vibrations generated by these bearings at a sampling rate of many kilohertz we can monitor its condition and detect certain types of defects. An introduction to frequency domain analysis will be given, where key signal features related to certain bearing defects will be explained. Furthermore, we will look at how these processed signals are used to feed a machine learning model. To gather the data we need and to make sure we have enough computational power, we also built a scalable data platform on AWS, containing both prototyping and production capabilities. The platform is based on Airflow, Docker, and EMR. We will discuss the considerations we made in building this platform and show the final result. This tech talk is interesting for both data scientists and data engineers.

This lecture is made possible by bigdata republic.