A frequency domain feature based cascade classifier and its application to fault diagnosis

In this paper, a new method of frequency domain feature extraction based on real discrete Fourier transform (RDFT) is proposed. The feature dimension is greatly reduced and pretty good real-time performance is achieved. In addition, to solve the multiple signal classification problem and get a detailed result, a method which adaptively chooses signal combination with the best accuracy and a cascade classifier using adaboost for the case of insufficient signal sources are designed. Experimental results show that the proposed method is computationally efficient and can archive high diagnosis accuracy.