F-1132 Machine Diagnosis by Neural Network Obtained by Category-Boundary Method
Hiroyuki TACHIBANA, Hiroshi TAKEDA, Keiichi TSUBOI · The proceedings of the JSME annual meeting · 2001
We investigate machine diagnostic systems by sound and vibration signals with neural networks (NN). In previous study, we introduced a new training method of NN by normal data and pseudo-normal data that is made near the normal data (Category-Boundary Method). In this paper, we propose how to make boundary data with statistical approach. In many cases, learning data have a certain statistical characteristics and we apply the characteristics to making boundary data. Next, we discuss an appropriate output function. A bounded output function has been used in our NN system, and we study an influence of the order of the bounded output function on a diagnosis by NN.