An Active Learning approach for the condition monitoring of rotating machinery

Rotating machinery breakdowns are most commonly caused by failures in bearing subsystems. Consequently, condition monitoring of such subsystems could increase reliability of machines that are carrying out field operations. Recently, research has focused on the implementation of vibration signals analysis for health status diagnosis in bearings systems considering the use of acceleration measurements. Informative features sensitive to specific bearing faults and fault locations were constructed by using advanced signal processing techniques which enable the accurate discrimination of faults based on their location. In this paper, the architecture of a diagnostic system for detecting progressive faults in bearings is presented. Progressive appearance of faults in gear box bearings has been measured and different stages of faults corresponding to larger depths of cracks in inner and outer bearings have been identified. It is already known that novelty detection can be easily combined with machine learning techniques so as to detect abnormal events. For change detection, a healthy bearing state description was constructed. As a result, deviations in the vibrational behaviour were detected. Further, an active learning method in the form of one-class classifiers for the progressive detection of the advancement of faults based on vibrational features is proposed. This method learns to distinguish between different fault stages. The proposed active learning method detects different types of faults as outliers which are then augmented in final hybrid classifier which can learn by adding new types of faults. It was shown that the detection of the progressive fault stages was successful up to the level of 90% by utilizing One Class Classifiers (based on Self Organizing Maps and Support Vector Machines).