This paper describes the development of a failure diagnosis technique for V-belts through vibration monitoring. The V-belt vibration is monitored at a driven bearing body attached to a power transmission device. Seven basic causes of belt failure and their combinations are considered. Power spectra of the vibration data are calculated through noise reduction by a cross-spectnnn method. Six parameters characterizing the vibration data are extracted, and 16 typical combinations of the basic causes and a normal belt state are diagnosed successfully by a Bayes' discriminant function approach. Two types of incorrect diagnosis are examined: Type I leaves a failed belt not repaired, and type II causes overmaintenance. A risk ratio for the Bayes' discriminant function is determined to minimize the two types of incorrect diag-nosis. Moreover, the risk ratio is determined to minimize type I error.

No abstract