Deep Learning Enabled Fault Diagnosis Using Time-Frequency Image Analysis of Rolling Element Bearings

Table 7

Prediction accuracies for 96 96-pixel image inputs.

Model	Spectrogram	Scalogram	HHT
MLP flat	80.1%	81.3%	56.8%
LSVM flat	77.1%	91.9%	52.8%
SVM flat	85.1%	93.3%	57.8%
MLP PCA	81.5%	96.4%	69.2%
LSVM PCA	74.1%	92.0%	51.4%
SVM PCA	49.6%	70.0%	68.8%
Architecture 2	81.5%	97.0%	74.2%
Architecture 1	86.2%	99.9%	91.8%
Proposed CNN architecture	91.7%	99.9%	95.5%