非線形識別手法による回転機音響監視の高度化
Improvement of Acoustic Monitoring of Rotating Machine by non-linear classification method
著者:
兼本 茂 Shigeru KANEMOTO
発刊日:
公開日:
Acoustic monitoringKernel-based PCAProbabilistic neural networkSVDD
概要
The acoustic data remotely measured by handy type microphones are investigated for monitoring and diagnosing the ball bearing type rotational machine integrity in nuclear power plants. The present study evaluates the state-of-the-art statistical signal processing and machine learning methods from the viewpoints of both sensitive and robust acoustic signal discrimination capability. The methods consist of feature extractions such as Fourier transformation or Cepstrum analysis, feature vector dimension compression by PCA or Kernel-based PCA, and, classification models by Probabilistic neural network (P NN) or support vector data description (SVDD). The performance of each algorithm is evaluated by experimentally measured acoustic data using mock-up test facility of roll bearing type rotating machine.
