Abstract

Condition based maintenance (CBM) aims to schedule maintenance activities based on condition monitoring data in order to lower the overall maintenance costs and prevent unexpected failures. Effective CBM can lead to reduced downtime, less inventory, reduced maintenance costs, reliable operation and safety of entire system. The key challenge in achieving effective CBM is the accurate prediction of equipment future health condition and thus the remaining useful life. Existing prognostics methods mainly focus on constant loading conditions. However, in many applications, such as some wind turbine, transmission and engine applications, the load that the equipment is subject to changes over time. It is critical to incorporate the changing load in order to produce more accurate prognostics methods. This research is focused on the bearing prognostics, which are key mechanical components in rotary machines, supporting the entire load imposed on machines. Failure of these components can stop the operation due to machine down time, thus resulting in financial losses, which are much higher than the cost of bearing. In this thesis, an artificial neural network (ANN) based method is proposed for equipment health condition prediction under time varying conditions. The proposed method can be applied to bearing as well as other components under condition monitoring. In the proposed ANN model, in addition to using the age and condition monitoring measurement values as an inputs, a new input neuron is introduced to incorporate the varying loading condition. The output of the ANN model is accumulated life percentage, based on which the remaining useful life can be calculated once the ANN is trained. Two sets of simulated degradation data under time varying load are used to demonstrate the effectiveness of the proposed ANN method, and the results show that fairly accurate prediction can be achieved using the proposed method. The other key contribution of this thesis is the experiment validation of the proposed ANN prediction method. The Bearing Prognostics Simulator, after extensive adjustment and tuning, is used to perform bearing run-to-failure test under different loading conditions. Vibration signals are collected using the data acquisition system and the Labview software. The root mean square (RMS) measurement of the vibration signals is used as the condition monitoring input for the validation of the proposed ANN prediction method. Two bearing failure histories are used to train the ANN model and test its prediction performance. The results demonstrate the effectiveness of the proposed method in dealing with real-world condition monitoring data for health condition prediction. The proposed model can greatly benefit industry as well as academia in condition based maintenance of rotary machines.