Diagnosing Rotating Machines With Weakly Supervised Data Using Deep Transfer Learning

Diagnosing Rotating Machines With Weakly Supervised Data Using Deep Transfer Learning

Table of Contents


Rotating machinery fault diagnosis problems have been well-addressed when sufficient supervised data of the tested machine are available using the latest data-driven methods. However, it is still challenging to develop an effective diagnostic method with insufficient training data, which is highly demanded in real-industrial scenarios, since high-quality data are usually difficult and expensive to collect. Considering the underlying similarities of rotating machines, data mining on different but related equipment potentially benefits the diagnostic performance on the target machine. Therefore, a novel transfer learning method for diagnostics based on deep learning is proposed in this article, where the diagnostic knowledge learned from sufficient supervised data of multiple rotating machines is transferred to the target equipment with domain adversarial training. Different from the existing studies, a more generalized transfer learning problem with different label spaces of domains is investigated, and different fault severities are also considered in fault diagnostics. The experimental results on four datasets validate the effectiveness of the proposed method, and show it is feasible and promising to explore different datasets to improve diagnostic performance.

  • Author Keywords

    • Adversarial training,
    • deep learning,
    • fault diagnosis,
    • rotating machinery,
    • transfer learning
  • IEEE Keywords

    • Fault diagnosis,
    • Training,
    • Task analysis,
    • Feature extraction,
    • Rotating machines,
    • Deep learning


Effective fault diagnosis of rotating machines is of great importance in real industries, leading to enhancement of machinery reliability, an increase in operating safety, and reduction of maintenance cost [1], [2]. In recent years, with the rapid development of data-driven algorithms, intelligent fault diagnosis methods have been attracting increasing attention, building straight-forward connections between measured signals and machine conditions with high accuracy.

In the current literature, a number of data-driven fault diagnosis approaches have been successfully proposed, including artificial neural networks (ANN) [3], support vector machines(SVM) [4], random forest (RF) [5], etc. Since the existing methods are mostly developed from statistics, it is generally assumed that sufficient high-quality data of the concerned machine are available for estimating the underlying data distributions in its different health conditions. While the data-driven methods generally outperform the conventional model-based approaches using large amounts of training data, they are less competitive if limited data can be obtained. In real industries, itis usually difficult and expensive to collect accurately labeledmachinery signals for training, which deteriorates the diagnostic performance and impedes the development of intelligent methods. Therefore, despite the promising fault diagnosis results obtained in the past years, further research efforts are supposed to be put on gaining useful and generalized knowledge from the small datasets. In order to address this challenging insufficient training data problem, one of the solutions lies in augmentations of the available data. The training dataset can be artificially expanded through transformations of samples, such as adding additional noise on the raw data, etc. Recent advances in generative models using deep neural networks are also promising in this perspective, creating fake samples to assist model training. Meanwhile, transfer learning offers an alternative method to effectively overcome the existing obstacles in data-driven fault diagnostics. In general, transfer learning aims to transfer the learned knowledge from the source domain to the target domain by relaxing the hypothesis that the data of the two domains must be from the same distribution [6]. Therefore, the insufficient training data problem can be potentially addressed by knowledge transformation from additional datasets with sufficient supervised information. In the existing studies, transfer learning approaches have already been employed in fault diagnosis tasks. However, the domain shift problem with respect to different operating conditions of the same machine is generally investigated inmost works [7]. This paper aims to improve the diagnostic performance on the target machine by knowledge transferationfrom multiple different but related source machines. A deep neural network is used in this transfer learning problem due to its high efficiency in feature extraction of high-dimensional machinery signals. As illustrated in Figure 1, discriminative features of machine conditions and generalized fault diagnostic knowledge are expected to be learned using sufficient high-quality data from different rotating machines, which poten 1551-3203 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information. This article has been accepted for publication in a future issue of this journal but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TII.2019.2927590, IEEE transactions on Industrial InformaticsIEEE TRANSACTIONS ON INDUSTRIAL INFORMATICSFig. 1. Proposed transfer learning method for diagnostics. The fault diagnosis knowledge learned from sufficient high-quality data of multiple rotating machines is transferred to diagnose the target machine.tially benefit the development of the data-driven diagnostic model of the target machine. The transfer learning task in this study is more challenging than most existing cross-domain problems in the literature, since more significant domain-biased features exist across multiple different machines, resulting from differences in mechanical structures, sampling frequencies. Nonetheless,it is still promising to utilize the shared features learned from different sources to improve the diagnostics of the target machine, where insufficient training data are available. The remainder of this paper starts with the related work in Section II. The proposed method is introduced in SectionIII and experimentally validated in Section IV. We close the paper with conclusions in Section V.


In this paper, a novel transfer learning method based on deep learning is proposed on machinery fault diagnosis. The diagnostic knowledge learned from sufficient supervised data in multiple sources rotating machines is transferred to the target machine, where the available data are insufficient to develop an effective fault diagnosis model independently. Generalizedfeatures are extracted from raw measured vibration signals through a deep neural network, which contains discriminative information with respect to different machine health conditions for different types of equipment. Domain adversarial training is adopted to obtain the shared features across machines facilitating knowledge transformation. Experiments are carried out using four different rotating machinery datasets, including three bearing fault and one shaft fault case. The results suggest when limited data of the tested machine are available, it is feasible and promising to explore data from different but related machines, and exploit the underlying shared features for diagnostics. Noticeable improvements in the diagnostic performance can be achieved using the proposed transfer learning method, even when the testing data are contaminated with additional noises. It should be pointed out that the main limitation of this study lies in the assumption that a small amount of supervised data is required in the target domain. Future research works will be concentrated on the investigation of transfer learning using unsupervised data, as well as the imbalanced data problem since data under machine faulty states, are most difficult to collect. Validations on real industrial data will also be focused on in the following studies.

About KSRA

The Kavian Scientific Research Association (KSRA) is a non-profit research organization to provide research / educational services in December 2013. The members of the community had formed a virtual group on the Viber social network. The core of the Kavian Scientific Association was formed with these members as founders. These individuals, led by Professor Siavosh Kaviani, decided to launch a scientific / research association with an emphasis on education.

KSRA research association, as a non-profit research firm, is committed to providing research services in the field of knowledge. The main beneficiaries of this association are public or private knowledge-based companies, students, researchers, researchers, professors, universities, and industrial and semi-industrial centers around the world.

Our main services Based on Education for all Spectrum people in the world. We want to make an integration between researches and educations. We believe education is the main right of Human beings. So our services should be concentrated on inclusive education.

The KSRA team partners with local under-served communities around the world to improve the access to and quality of knowledge based on education, amplify and augment learning programs where they exist, and create new opportunities for e-learning where traditional education systems are lacking or non-existent.

FULL Paper PDF file:

Diagnosing Rotating Machines with WeaklySupervised Data Using Deep Transfer Learning



X. Li, W. Zhang, Q. Ding and X. Li,




Diagnosing Rotating Machines With Weakly Supervised Data Using Deep Transfer Learning

Publish in

in IEEE Transactions on Industrial Informatics, vol. 16, no. 3, pp. 1688-1697, March 2020,



PDF reference and original file: Click here

+ posts

Somayeh Nosrati was born in 1982 in Tehran. She holds a Master's degree in artificial intelligence from Khatam University of Tehran.

Website | + posts

Professor Siavosh Kaviani was born in 1961 in Tehran. He had a professorship. He holds a Ph.D. in Software Engineering from the QL University of Software Development Methodology and an honorary Ph.D. from the University of Chelsea.

Website | + posts

Nasim Gazerani was born in 1983 in Arak. She holds a Master's degree in Software Engineering from UM University of Malaysia.