The emergence of a defect in a mechanical system is always associated with a change in the vibration’s behavior in the spectral and temporal domains. Fault detection by vibration analysis is based on monitoring the behavior of a mechanical component by examining the evolution of fault indicators in real time. However, mere traditional bearing diagnosis is not sufficient to ensure effective and reliable assessment of the component’s health condition. Coupling several fault indicators extracted by different signal processing technique adds more reliability and accuracy to the diagnostic process.
Classifications methods are used to analyze the evolution of fault, yet only static methods are solicited which results in overlooking a great source of useful information.
In fact, fault indicators issued from turning machine are evolving; they change constantly over time, particularly when the defect is growing. In such situations, static classification methods are a poor choice that deprives the user of the information conveyed in the evolution of the indicators over time.
The dynamic classification of fault indicators in dynamic classes can provide useful information about the behavior of the damaged bearings. This information can also be used to predict the end of life of the components.
Unlike the static classification, the dynamic classification introduces a new dimension (time), which allows real-time detection of the fault and better visibility of the bearing behavior revealed by the motion and the temporal evolution of classes formed by the indicators.
This paper proposes a dynamic classification method that uses several fault detectors to assure the accuracy of the diagnostic and follow up any changing in the behavior of the bearing by analyzing the classes’ time evolution.
The proposed multi-features dynamic classification is a new method for fault detection and health condition monitoring for bearings; this technique utilizes multiple features, including traditional features extracted from the raw signal, two special features extracted by wavelet analysis, and the spectral kurtosis, coupled with a nonlinear principal component analysis.
This method of classification clusters the multi-features into several classes, the first class represents the healthy state of the bearing, and the other classes represent different damaging state. Monitoring the evolution of the” defective condition” class allows us to draw several useful information, such as the rate of degradation, the relationship between the cluster’ surface and density and the bearing state.
The chosen dynamic classification method will be validated by analyzing several degradation bearings from a fatigue bench of thrust ball bearings SNR51207.