Search

Data-driven fault diagnostics of industrial systems suffer from class-imbalanced problems, which is a common challenge for machine learning algorithms as it is difficult to learn the features of the minority class samples. Synthetic oversampling methods are commonly used to tackle these problems by generating minority class samples to balance the majority and minority classes. Two major issues will influence the performance of oversampling methods which are how to choose the most appropriate existing minority seed samples, and how to synthesize new samples from seed samples effectively. However, many existing oversampling methods are not accurate and effective enough to generate new samples when dealing with high-dimensional faulty samples with different imbalanced ratios, since they do not take these two factors into consideration at the same time. This article develops a novel adaptive oversampling technique: expectation maximization (EM)-based local-weighted minority oversampling technique for industrial fault diagnostics. This method uses a local-weighted minority oversampling strategy to identify hard-to-learn informative minority fault samples and an EM-based imputation algorithm to generate fault samples based on the distribution of minority samples. To validate the performance of the developed method, experiments were conducted on two real-world datasets. The results show that the developed method can achieve better performances, in terms of F-measure, Matthews correlation coefficient (MCC), and Mean (average of F-measure and MCC) values, on multiclass imbalanced fault diagnostics in different imbalance ratios than state-of-arts’ baseline sampling techniques. [ABSTRACT FROM AUTHOR]

• Generalization of diagnosis model is improved on new hard disks without faults. • New faulty samples of hard disks could be generated from healthy samples. • Distribution discrepancy between different types of hard disks could be decreased. • End-end EM-based training strategy guarantees a good accuracy and convergence. Intelligent fault diagnosis of hard disks becomes significantly important to guarantee reliability of current cloud-based industrial systems. Most intelligent diagnostic methods are commonly based on assumptions that data from different disks are subject to the same distribution and there are sufficient faulty samples for training the models. However, in reality, there are types of hard disks from different manufacturers and their SMART encoding varies widely across manufacturers. It results in distribution discrepancy among disks and influences the generalization of machine learning methods. Moreover, hard disks usually work in healthy state that faulty events rarely happen on most of them, or especially never occur on new ones. Thus, this paper proposes a deep generative transfer learning network (DGTL-Net) for intelligent fault diagnostics on new hard disks. The DGTL-Net combines the deep generative network that generates fake faulty samples and the deep transfer network that solves the problem of distribution discrepancy between hard disks. An iterative end-end training strategy is also proposed for DGTL-Net to get the most optimal parameters of generative and transfer network simultaneously. Experiments have been conducted to prove that our method achieves better performance. [ABSTRACT FROM AUTHOR]