Wavelets for spindle fault diagnosis in high speed machining

Abstract

The spindle of machining centers must provide high rotational speed, transfer torque and power to the cutting tool during continuous periods of time. The constant forces generate faults in its components where the most important are the shaft and bearings. As the fault increases, it affects other components and may lead to a catastrophic damage and a production stoppage. The maintenance strategies have been evolving in order to prevent irreversible damages. Over the last years, great progress has been made in the condition-based maintenance, particularly in the vibration analysis, where the vibration signature can be associated with the fault. In recent years, several signal-processing techniques have been introduced to extract the features from vibration signals. The WT has caught the attention of the scientific community by its characteristics and its limitless number of wavelets. In this thesis a methodology based on the WT is proposed to detect faults in spindle. The approach is capable of extracting the bearing characteristic frequencies related to the fault from the resonance frequency and the low frequencies information associated with shaft faults. The implemented method contemplates the latest advances in the literature to detect robustly the type of the fault, it is focused on industrial environment were the faults are usually tainted by noise from other machines or by errors in the acquisition. The method is applied to different types of bearing faults to demonstrate its effectiveness and robustness when detecting faults at early stages. In the three studied cases the proposed methodology got several properties; for the CWRU signals the characteristic fault frequency peak got an increase from 6 to 32% compared with the traditional methods; when the signal is tainted by Gaussian noise, the method works more effectively, since in these cases the increase percentage reaches up to 57%. Similarly, in the IMS database the characteristic frequency peak increases from 6 to 70%. Finally, in the machining center database there was not an increment but the method acts as filter which eliminates the undesired frequencies. Experimental results indicate the proposed approach is reliable to detect bearing and shaft faults. It also has a superior diagnosis performance compared to traditional methods in extracting fault features. The method removes most of the noise and can be used in future works as preprocessor.