A fault diagnosis methodology for rolling element bearings based on advanced signal pretreatment and autoregressive modelling

This study proposes a methodology for rolling element bearings fault diagnosis which gives a complete and highly accurate identification of the faults present. It has two main stages: signals pretreatment, which is based on several signal analysis procedures, and diagnosis, which uses a pattern-recognition process. The first stage is principally based on linear time invariant autoregressive modelling. One of the main contributions of this investigation is the development of a pretreatment signal analysis procedure which subjects the signal to noise cleaning by singular spectrum analysis and then stationarisation by differencing. So the signal is transformed to bring it close to a stationary one, rather than complicating the model to bring it closer to the signal. This type of pre-treatment allows the use of a linear time invariant auto-regressive model and improves its performance when the original signals are non-stationary. This contribution is at the heart of the proposed method, and the high accuracy of the diagnosis is a result of this procedure. The methodology emphasizes the importance of preliminary noise cleaning and stationarisation. And it demonstrates that the information needed for fault identification is contained in the stationary part of the measured signal.

The methodology is further validated using three different experimental setups, demonstrating very high accuracy for all of the applications. It is able to correctly classify nearly 100 percent of the faults with regard to their type and size. This high accuracy is the other important contribution of this methodology. Thus, this research suggests a highly accurate methodology for rolling element bearing fault diagnosis which is based on relatively simple procedures. This is also an advantage, as the simplicity of the individual processes ensures easy application and the possibility for automation of the entire process.