A Robust Deep Neural Network for Rolling Element Fault Diagnosis under Various Operating and Noisy Conditions

Abstract

This study proposes a new intelligent diagnostic method for bearing faults in rotating machinery. The method uses a combination of nonlinear mode decomposition based on the improved fast kurtogram, gramian angular field, and convolutional neural network to detect the bearing state of rotating machinery. The nonlinear mode decomposition based on the improved fast kurtogram inherits the advantages of the original algorithm while improving the computational efficiency and signal-to-noise ratio. The gramian angular field can construct a two-dimensional image without destroying the time relationship of the signal. Therefore, the proposed method can perform fault diagnosis on rotating machinery under complex operating conditions. The proposed method is verified on the Paderborn dataset under heavy noise and multiple operating conditions to evaluate its effectiveness. Experimental results show that the proposed model outperforms wavelet denoising and the traditional adaptive decomposition method. The proposed model achieves over 99.6% accuracy in all four operating conditions provided by this dataset, and 93.8% accuracy in a strong noise environment with a signal-to-noise ratio of −4 dB.

Editorial for Special Issue: Machine Health Monitoring and Fault Diagnosis Techniques

“…The theme of this Special Issue focuses on machine health monitoring and fault diagnosis techniques, especially intelligent fault diagnosis. This Special Issue highlights 18 articles that can be divided into four categories: condition monitoring [ 1 , 2 , 3 , 4 ], degradation process prediction [ 5 , 6 , 7 , 8 ], intelligent diagnostic algorithms [ 9 , 10 , 11 , 12 , 13 , 14 , 15 ], and sensor fault detection [ 16 , 17 , 18 ]. In addition to the traditional bearing vibration signals, the research objects include the electrode signals, blade vibration signals, diesel engine vibration signals, and bearing heat signals.…”

“…In [ 11 ], a novel model was proposed for an intelligent bearing fault diagnosis in rotating machinery. The main contribution of this model is the construction of an effective image dataset using a combination of an improved fast kurtogram (IFK) that was based on nonlinear mode decomposition (NMD) and a gramian angular field (GAF).…”

Impact of Deep Learning Optimizers and Hyperparameter Tuning on the Performance of Bearing Fault Diagnosis

“…Recently, there have also been active studies on fault diagnosis based on 1D CNNs that employ structures such as residual connections and dilated convolutions, which have been proven to perform well in other application areas, including image processing and audio processing [15], [16], [17], [18], [19], [20]. On the other hand, 2D CNN-based algorithms incorporate various feature extraction techniques, such as signal-to-image mapping [21], [22], [23], STFT [24], [25], cyclic spectral coherence [26], and nonlinear mode decomposition [27].…”

“…In particular, deep learning models achieve higher performance than conventional data-based approaches even when raw inputs are used without feature extraction and selection [12]. Furthermore, in bearing fault diagnosis, many researchers have obtained approximately 100% accuracy without preprocessing using convolutional neural networks (CNN) [13], [14], [15], [16], [17], [18], [19], [20], [21], [22], [23], [24], [25], [26], [27], recurrent neural networks (RNN) [28], [29], [30], and Transformers [31], [32]. Studies have also been conducted to diagnose the failures of multi-domain data measured under different working loads or to develop a model that can classify bearing faults even in a noisy environment.…”

A Comparative Analysis of Deep Learning Convolutional Neural Network Architectures for Fault Diagnosis of Broken Rotor Bars in Induction Motors

“…Vibration analysis with classifiers like KNN has also been explored [24]. Advancements incorporate intelligent diagnostics [25] and real-time neural classifiers [26]. Hybrid statistical and machine learning approaches achieve high accuracy [23].…”