Machine tool spindle bearing diagnostics under operating conditions
MetadataShow full item record
Bearing diagnostics provide valuable information related to a bearing’s health and facilitate Condition Based Maintenance (CBM) for rotary machines. This is an effective method to decrease unnecessary cost and downtime resulting from unanticipated machine spindle failure. Defective signatures can be extracted from the corresponding vibration signals through both the time and frequency domain signal processing procedures. However, techniques to effectively evaluate bearing damage severity from these extracted features are still a significant challenge, because the relationship between the bearing damage severity and the extracted feature is not well understood. Moreover, previous methods are mostly tested under constant loading conditions, and are not suitable for bearing diagnostics during machining operations. In this thesis, a time-domain-based bearing defect size estimation method is proposed for the inner and outer race defects. This new approach is built on the bearing system nonlinear dynamic model and the Hertzian contact defect size estimation model. The new defect size estimation model is independent of the contact force between ball and raceway, and all the required information for defect estimation can be obtained from the vibration signal. The signal processing method is developed to automatically extract the time information from the vibration signal for defect size estimation. Statistical analysis is performed on the time information and the results support the proposed bearing models. A test system designed for CNC-based bearing diagnostics was fabricated to validate the new method. Experiments in the speed range 500-3000 rpm were performed under both no-cutting and cutting conditions with different feed rates. Experiment results are consistent under different operational conditions (speeds/feeds), and they are agreeable to both the bearing system dynamic model and the defect estimation model. The estimation results are close to the true defect size with relative error of approximately 10%, validating the proposed method.