Within the HEMIS project, we are developing an in-vehiclePrognostic and Health Management System (PHMS) for monitoring the bearings of a Fully Electrical Vehicle. The objective of the system is to detect the onset of degradation, isolate the degrading bearing, classifying the type of defect, assess the degradation intensity and predict the failure time. The PHMS takes into account that automotive motors work in operational conditions characterized by variable load, rotational speed and other external condition which remarkably influence the degradation process.
The diagnostic system is based on a hierarchical structure of K-Nearest Neighbours classifiers (Figure 1).
Figure 1: Diagnostic system based on a hierarchical structure of K-Nearest Neighbours classifiers (DE=Drive End, NDE=Non Drive End)
Input to the PHMS are vibrational signals which contain information on the degradation state of the bearing system. In particular, within the HEMIS project we have select a set of features to be extracted from vibrational signals by resorting to a wrapper approach based on a Multi-Objective (MO) optimization that integrates a Binary Differential Evolution (BDE) algorithm with the K-Nearest Neighbour (KNN) classifiers. Objectives of the selection of the features have been to have high monitoring performances and to reduce the computational and deployment costs (low number of features, low number of vibrational sensors). The developed approach has been applied with success to an experimental dataset of literature (Case Western Reserve University Bearing Data,
http://csegroups.case.edu/bearingdatacenter/pages/welcome-case-western-reserve-university-bearing-data-center-website).The percentage of patterns for which the classification is correct in all the 4 stages of the diagnostic system is 97.61%.
Once the degrading bearing is identified, the type of defect classified and the intensity of the defect assessed, the prediction of the bearing remaining useful life (RUL) is performed by using a homogeneous 4-state continuous-time finite-state semi-Markov model (Figure 2).
Figure 2: Multi-state model of bearing degradation
The degradation process is hidden, as the information about the bearing degradation state cannot be directly known, but are estimated by the diagnostic system described above (level 1, for fault detection and level 4 for assessing the degradation state), as shown in Figure 3.
Figure 3: Hidden semi-Markov degradation modelling: the degradation state is not known a priori but is estimated by the diagnostic system.
The proposed prognostic method consists of two phases: first, an off-line phase where the parameters of the degradation model are estimated based on the gathered observations; then, an on-line phase, in which the inferred degradation model is used to assess the current degradation state of the component and, on this basis, estimate its Residual Useful Life (RUL) (Figure 4).
Figure 4: RUL estimation obtained at different time during the bearing life.