An SDP Characteristic Information Fusion-Based CNN Vibration Fault Diagnosis Method

Table 6

Comparison of classification accuracy with different algorithms from the literature.

Classifier	Feature	Fault detection test accuracy
SVM [24]	WP	92.5% (CWRU simulated data with small number of train/test samples)
SVM [25]	EMD	90.3% (CWRU simulated data with small number of train/test samples)
		92.4% (BRD simulated data with small number of train/test samples)
SVM	Locally linear embedding	88.9% (CWRU simulated data with small number of train/test samples)
		85.7% (BRD simulated data with small number of train/test samples)
DBN [26]	Original signal parameter	90.7% (CWRU simulated data with large number of train/test samples)
CNN [27]	Two-dimensional spectrum	92.6% (CWRU simulated data with large number of train/test samples)
CNN [28]	Time domain waveforms	91.2% (CWRU simulated data with large number of train/test samples)
CNN	SDP	93.7% (CWRU simulated data with large number of train/test samples)