抄録
This paper presents an algorithm to judge multiple statuses of distribution equipment from continuous vibration measurement. The algorism finds the acceleration threshold of moving and staying statuses as cluster boundaries of acceleration level distributions. In the method, first the measured acceleration data are divided into multiple clusters using tentative boundaries. Then acceleration data are re-classified as one of clusters using k-nearest neighbor algorithm and Mahalanobis distances. Validity of the method is confirmed by the simulation with simply distributed samples and by an experiment with measured acceleration data of an automobile. For the case of overlapped boundaries, which is caused by inappropriate numbers of tentative boundaries, the algorism to find appropriate cluster number is proposed and confirmed by the simulation and measured acceleration data.
