Noise eliminated ensemble empirical mode decomposition scalogram analysis for rotating machinery fault diagnosis

Rotating machinery is one type of major industrial component that suffers from various faults and damage due to the constant workload to which it is subjected. Therefore, a fast and reliable fault diagnosis method is essential for machine condition monitoring. Artificial intelligence can be applied in fault feature extraction and classification. It is crucial to use an effective feature extraction method to obtain most of the fault information and a robust classifier to classify those features. In this study, an improved method, noise-eliminated ensemble empirical mode decomposition (NEEEMD), was proposed to reduce the white noise in the intrinsic functions and retain the optimum ensembles. A convolution neural network (CNN) classifier was applied for classification because of its feature-learning ability. A generalised CNN architecture was proposed to reduce the model training time. The classifier input used was 64×64 pixel RGB scalogram samples. However, CNN requires a large amount of training data to achieve high accuracy and robustness. Deep convolution generative adversarial network (DCGAN) was applied for data augmentation during the training phase. To evaluate the effectiveness of the proposed feature extraction method, scalograms from the related feature extraction methods such as ensemble empirical mode decomposition (EEMD), complementary EEMD (CEEMD) and continuous wavelet transform (CWT) were also classified. The effectiveness of the scalograms was also validated by comparing the classifier performance using greyscale samples from the raw vibration signals. The ability of CNN was compared with two traditional machine learning algorithms, k nearest neighbour (kNN) and the support vector machine (SVM), using statistical features from EEMD, CEEMD and NEEEMD. The proposed method was validated using bearing and blade datasets. The results show that the machine learning algorithms achieved comparatively lower accuracy than the proposed CNN model. All the outputs from the bearing and blade fault classifiers demonstrated that the scalogram samples from the proposed NEEEMD method obtained the highest accuracy, sensitivity and robustness using CNN. DCGAN was applied with the proposed NEEEMD scalograms to enhance the CNN classifier’s performance further and identify the optimal amount of training data. After training the classifier using the augmented samples, the results showed that the classifier obtained even higher validation and test accuracy with greater robustness. The test accuracies improved from 98%, 96.31% and 92.25% to 99.6%, 98.29% and 93.59%, respectively, for the different classifier models using NEEEMD. The proposed method can be used as a more generalised and robust method for rotating machinery fault diagnosis