Mechanical fault diagnostics for quality control of manufacturing appliances is often based on the analysis of machine vibrations. In the presence of mechanical faults, vibration signals comprise periodic impulses with a characteristic frequency corresponding to a particular defect.
Vibrometers based on LDV (Laser Doppler Vibrometry) are an alternative to the traditional use of piezo-electric accelerometers, devices that are the most common and popular vibration transducers. Laser vibrometry is now an established technique for vibration measurements in industrial applications where non-contact operations are essential. Despite the advantages of LDV, speckle noise occurs when rough surfaces are measured and the object is moving. Therefore, removal of speckle noise is a crucial point of a reliable system for diagnostics of mechanical faults.
This paper deals with the analysis of vibration signals measured by a Laser Doppler Vibrometer from different electromechanical components such as compressors and electrical motors. The goal is to suppress the speckle-noise effect, coming both from the surface of the electro-mechanical components and its movement, in order to perform an automatic test for mechanical fault detection in the production line. First, data acquisition and its problems are introduced. Then, the chosen solution is presented. In particular, a statistical approach based on kurtosis is used for detection of speckle noise and selection of an undistorted region within the signal. The algorithm is composed of band-pass filtering, segmentation of the signal and computation of a scalar indicator KR (Kurtosis Ratio) for each signal segment, in order to detect outlying samples caused by speckle noise. Finally, real examples are shown to test the proposed method, and a tool for validation of signal databases is briefly presented.