Transfer Relation Network for Fault Diagnosis of Rotating Machinery With Small Data
Abstract : Many deep-learning methods have been developed for fault diagnosis. However, due to the difficulty of collecting and labeling machine fault data, the datasets in some practical applications are relatively much smaller than the other big data benchmarks. In addition, the fault data come from different machines. Therefore, on some occasions, fault diagnosis is a multidomain problem with small data, where satisfactory transfer performance is difficult to obtain and has been rarely explored from the few-shot learning viewpoint. Different from the existing deep transfer learning solutions, a novel transfer relation network (TRN), combining a few-shot learning mechanism and transfer learning, is developed in this study. Specifically, the fault diagnosis problem has been treated as a similarity metric-learning problem instead of solely feature weighted classification. A feature net and a relation net have been, respectively, constructed for feature extraction and relation computation. The Siamese structure has been borrowed to extract the features of the source and the target domain samples with shared weights. Multikernel maximum mean discrepancy (MK-MMD) is employed on several higher layers with different tradeoff parameters to enable an efficient domain feature transfer considering different feature properties. To implement efficient diagnosis based on small data, an episode-based few-shot training strategy is adopted to train TRN. Average pooling has been adopted to suppress the noise influence from the vibration sequence which turns out to be important for the success of time sequence-based fault diagnosis. Transfer experiments on four datasets have verified the superior performance of TRN. A significant improvement of classification accuracy has been made compared with the state-of-the-art methods on the adopted datasets.
? In this survey paper, we systematically summarize existing literature on bearing fault diagnostics with deep learning (DL) algorithms.
? Designed to tackle this practical and widely existing issue in numerous applications, transfer learning has aroused extensive attention in the machine learning community, and various transfer learning frameworks are proposed based on classical ML algorithms.
? Most of the existing DL algorithms can achieve an excellent classification accuracy of over 95% using the CWRU dataset, even with the classical CNN without any add-on architectures, which indicates that this dataset contains relatively simple features that can be easily extracted by a variety of DL methods.
? Most of the existing work employing DL techniques for bearing fault diagnosis relies on vibration data collected from accelerometers in a laboratory environment.
? Adaptation methods are based on the adaptation of conventional classification methods to multi-label versions without problem transformation.
? However, practical engineering problems demanded the development of new automatic methods that are able to deal with multiple fault scenarios and noisy or uncertain fault features.
? The transfer of recent Machine Learning approaches to practical fault diagnosis problems of rotating electrical machines and drive systems may help to facilitate smart industrialization and intelligent modernization.
? Automated electrical signature analysis-based fault detection and diagnosis of electrical machines is still an especially important issue in modern industrial applications.
• The study demonstrated that the proposed PCA technique is effective in classifying bearing faults with a higher accuracy and a lower number of input features when compared to using all of the original feature.
• This proposed ADCNN is also able to predict the fault size (defect width) with a satisfactory accuracy.
• The proposed method has a better diagnosis accuracy than deep belief networks, particularly with the added noises, where an average improvement of 7% is achieved.
• In, an ensemble deep auto-encoder consisting of a series of auto-encoders (AE) based on different activation functions is proposed for unsupervised feature learning from the measured vibration signal.
? The proposed method also employs vibration and current signals and achieves nice performance; however, the method lacks the flexibility to easily fit with specific types of machines.
? In the presence of multiple faults, the single fault recognition techniques’ performance may be degraded.
? The performance of a classifier strongly depends on the effectiveness of the features representing the patterns of different fault events.
? It is worth noting that many attempts have been made with the purpose of improving performance through sophisticated feature extraction methods.
? We have also modified the maximum of tree depth in order to compare the performance using various maximum depths of the trees.
|