THE ROLE OF HOUGH TRANSFORM FOR AUTOMATIC INTERPRETATION OF SPECTRAL CORRELATION DIAGRAMS

摘要

Cyclostationarity analysis has been proved to be very effective to detect and classify rolling bearing faults. To detect 2nd order cyclostationarity properties in signals, spectral correlation density diagrams are computed. They are frequency- frequency diagram, where peaks are associated to the bearing fault frequencies.In spectral correlation diagrams peaks due to bearing faults present typical rhomboidal patterns. However spectral correlation diagrams can result pretty complex to interpret when bearing faults are at the initial stages. From a prognostics stand point, the detection of bearing defects at an early stage is fundamental to track the fault evolution. Hough transform is a feature extraction technique used in image processing. It allows identifying arbitrary shapes within an image (straight lines or circles) in presence of low SNR values. This paper presents a novel approach for the analysis of spectral correlation density diagrams based on Hough transform. The integration of Hough transform within the framework of the spectral correlation analysis aims at improving the detection and classification of bearing defects patterns at the initial stage of the fault. A fully automated method for detection and classification of bearing faults will be presented. In such strategy the coherent (complex) Hough transform has the role to improve the robustness of the detection and the classification stage.