|
| Ref. | Main features | Database, bearing states | Classifier | Accuracy (%), state samples |
|
| [61] | EMD energy entropy of the first eight IMFs | NASA, seven | ANN | 93. Total 5394 records divided into five folds, training: Four folds, test: one fold |
| [62] | Time and frequency domain features | NASA, seven | Linear SVM and quadratic SVM | 99.4 of linear SVM, 99.3 of quadratic SVM. Training: 80%, test: 20% |
| [63] | Time domain features | NASA, two | Classical SVM (CSVM), incremental SVM (ISVM) | Outer: 91.1 CSVM and 98.7 ISVM, inner: 92.0 CSVM and 94.5 ISVM. Training: 70, test: 30 |
| [64] | FFT | NASA, two | 1D convolutional neural networks (1D CNN) | 97.1 of 1D CNN, 94.5 of FFT-SVM |
| [65] | LMD, sample entropy, and energy ratio | CWRU, four | SVM | 100. Training: 60, test: 20 |
| [66] | Time and frequency, EMD energy entropy | CWRU, even | Adaptive neuro-fuzzy inference system (ANFIS) | 94.7 (average). Training: 140, test: 70 |
| [67] | DWT (cluster-based feature extraction) | CWRU, ten | Probabilistic neural network | 98.2 (maximum). Training: 168, test: 60 |
| [68] | EMD sample entropy of the first ten IMFs | CWRU, six | Improved shuffled frog leaping algorithm (ISFLA) | 100 for H, others 95.4 (maximum). Training: 140, test: 70 |
| [69] | LMD-SVD | CWRU, ten | BPNN SVM, extreme learning machine (ELM) | 97.7 (average) for BP, 98.8 (average) for SVM, 99.3 (average) for ELM. Test: 228 |
| [70] | Continuous wavelet transform (CWT) | CWRU, ten | Convolutional neural network (CNN) | 99.7 of CNN, 99.7 of CNN, 85.1 of BPNN training: 200, Test: 200 |
| [71] | Transfer learning | CWRU, six | Neural networks | 91.8 (total). Total: 4832, training: 1208 |
| [72] | Time domain features | CWRU, ten | Hierarchical adaptive deep CNN (ADCNN) | 99.7 (average). Training: 500, test: 500 |
| [73] | SVD, the singular values | CWRU, four | ANN | 95.1, training: 336, test: 144 |
|
