Prediction of bearing failures by the analysis of the time series

Abstract

As they are a part of the energy transmission chain, bearings are considered as important mechanical components in rotating machines. This importance requires a special attention in order to avoid expensive production shutdown due to the appearance of failures. It is therefore necessary to anticipate the appearance of faults by implementing an appropriate prediction model. There exist in the literature several examples of prediction models able to estimate the remaining useful life (RUL) of bearings. These models are based on the principle of the long-term prediction without considering the degradation state of bearings or by defining a degradation threshold arbitrarily. It should be noted that, in addition to a reliable prediction model, identifying the degradation states is an important parameter in the estimation of the RUL. The presented paper proposes a particular approach based on the artificial intelligence (AI) principle. The proposed approach is composed of two model-based AI inspired by the reasoning of the human for the RUL estimation. This reasoning is modeled via the neural networks for time-series prediction. These models are called the adaptive-neuro fuzzy inference system and the neo-fuzzy neuron. The proposed approach is also composed by a third model-based AI. This model is inspired by the behavior of ants to identify the different degradation states of bearings. The combination of these two types of AI provides reliable and robust prediction results.