Higher order statistics for detection and classification of faulty fanbelts using acoustical analysis

Abstract

Higher order statistics (HOS) are well suited to solving detection and classification problems because they can suppress gaussian noise and preserve some of the non-gaussian information. This paper describes the use of these methods for acoustic quality control of manufactured goods on a production line, and specifically the detection of faulty fanbelts on the drying block of a washing machine. Two HOS based methods were used in this paper. The first is based on the properties of the bispectrum in the outer triangle and particularly of the normalized bispectrum also called the skewness function. The second uses the third order cumulant of a matched filter output. This combination has the advantages of matched filtering plus the properties of higher than second order statistics making this algorithm more robust than the conventional matched filter. The method was used to classify real signals from fanbelts suffering from specific known defects.