For the data-driven remaining useful life (RUL) prediction for rolling bearings, traditional machine learning-based methods generally provide insufficient feature representation and adaptive extraction. Although deep RUL can solve these problems to some extent, they still do not yield satisfactory predictive results due less degradation data inconsistent distribution among different bearings. To problems, a new method based on transfer learning is proposed in this paper. This includes an...

Due to the real working conditions and data acquisition equipment, collected of bearings are actually limited. Meanwhile, as rolling bearing works in normal state at most times, it is easy raise imbalance problem fault types which restricts diagnosis accuracy stability. To solve these problems, we present an imbalanced method based on generative adversarial network (GAN) provide a comparative study detail. The key idea utilizing GAN, kind deep learning technique, generate synthetic samples...

Intelligent fault-diagnosis methods using machine learning techniques like support vector machines and artificial neural networks have been widely used to distinguish bearings’ health condition. However, though these generally work well, they still two potential drawbacks when facing massive fault data: (1) the feature extraction process needs prior domain knowledge, therefore lacks a universal method for various diagnosis issues, (2) much training time is needed by traditional intelligent...

For bearing remaining useful life prediction problem, the traditional machine-learning-based methods are generally short of feature representation ability and incapable adaptive extraction. Although deep-learning-based proposed in recent years can effectively extract discriminative features for fault, these tend to less consider temporal information fault degradation process. To solve this a new approach based on deep long short-term memory neural network is article. First, criterion, named...

This paper presents a new online detection approach for rolling bearing's incipient fault based on self-adaptive deep feature matching (SDFM). includes offline and stages. At the stage, health state assessment algorithm is first proposed singular value decomposition (SVD) Kurtosis criterion. Based results, kind of learning algorithm, i.e., stacked denoising autoencoder (SDAE), introduced to extract common features normal early state. Support vector data description (SVDD) applied establish...

This article presents a new deep transfer learning method, named structured domain adversarial neural network (SDANN), for bearing fault diagnosis with the data collected under different working conditions. The key idea of this method is integrating strong adaptability (DANN) and relatedness information among multiple failure modes to improve effect learning. First, fine-grained alignment between from conditions, loss function discriminative regularizer designed DANN by using maximum...

This article tries to answer the two questions of bearings' remaining useful life (RUL) prediction with deep transfer learning: what bearing data in source domain contribute more learning and how quantify such contribution? From perspective sample-based interpretability, this proposes a new approach RUL prediction. First, time series clustering algorithm based on multiscale degradation similarity is proposed. Comprehensively considering geometry tendency characteristics sequences, can divide...

This article proposes a novel deep transfer learning-based online remaining useful life (RUL) approach for rolling bearings under unknown working condition. solves the following concerns: drift of condition would block data accumulation and raise bias in prediction model, bearing merely has early fault when activating RUL prediction, failing to conduct learning from offline data. First, new time series recursive forecasting model is constructed generate pseudovalues via fusing prior...

High-dimensional problems are ubiquitous in many fields, yet still remain challenging to be solved. To tackle such with high effectiveness and efficiency, this article proposes a simple efficient stochastic dominant learning swarm optimizer. Particularly, optimizer not only compromises diversity convergence speed properly, but also consumes as little computing time space possible locate the optima. In optimizer, particle is updated when its two exemplars randomly selected from current...

Despite the big success of transfer learning techniques in anomaly detection, it is still challenging to achieve good transition detection rules merely based on preferred data with one-class classification, especially for a large distribution difference. To address this challenge, novel deep algorithm domain-adversarial training proposed paper. First, by integrating hypersphere adaptation constraint into neural network, new adversarial mechanism designed. Second, an alternative optimization...

With deep transfer learning techniques, this paper focuses on the online remaining useful life (RUL) prediction problem across different machines, and tries to address following concerns: 1) The effect of decreases significantly due considerable divergence degradation characteristic; 2) A high computational cost is raised by re-training whole model with data; 3) Error accumulation occurs because lacking label information data. In paper, a self-supervised tensor domain-adversarial regression...

Online remaining useful life (RUL) prediction has been solved by deep transfer learning, but is still with challenges as follows: (1) Incomplete and unlabeled data under actual operation; (2) Condition monitoring streaming unknown distribution; (3) The distribution of degradation variable. To solve them, an unsupervised incremental learning approach knowledge distillation (KD) proposed. First, a time series recursive model built to generate pseudo labels. Second, online KD network...

In many applications, it is not easy to generate enough whole-life data for training a deep neural network, which may reduce the performance of health indicator (HI). To solve this problem, new HI construction method based on transfer learning proposed in article. First, multiscale domain-adversarial network extract representative features from collected under different working conditions. By introducing maximum mean discrepancy regularizer and Laplace regularizer, model can enhance...