Predictive maintenance plays an important role in modern Cyber-Physical Systems (CPSs) and data-driven methods have been a worthwhile direction for Prognostics Health Management (PHM). However, two main challenges have significant influences on the traditional fault diagnostic models: one is that extracting hand-crafted features from multi-dimensional sensors with internal dependencies depends too much on expertise knowledge; the other is that imbalance pervasively exists among faulty and normal samples. As deep learning models have proved to be good methods for automatic feature extraction, the objective of this paper is to study an optimized deep learning model for imbalanced fault diagnosis for CPSs. Thus, this paper proposes a weighted Long Recurrent Convolutional LSTM model with sampling policy (wLRCL-D) to deal with these challenges. The model consists of 2-layer CNNs, 2-layer inner LSTMs and 2-Layer outer LSTMs, with under-sampling policy and weighted cost-sensitive loss function. Experiments are conducted on PHM 2015 challenge datasets, and the results show that wLRCL-D outperforms other baseline methods.
Federated learning is a distributed learning framework that is communication efficient and provides protection over participating users' raw training data. One outstanding challenge of federate learning comes from the users' heterogeneity, and learning from such data may yield biased and unfair models for minority groups. While adversarial learning is commonly used in centralized learning for mitigating bias, there are significant barriers when extending it to the federated framework. In this work, we study these barriers and address them by proposing a novel approach Federated Adversarial DEbiasing (FADE). FADE does not require users' sensitive group information for debiasing and offers users the freedom to optout from the adversarial component when privacy or computational costs become a concern. We show that ideally, FADE can attain the same global optimality as the one by the centralized algorithm. We then analyze when its convergence may fail in practice and propose a simple yet effective method to address the problem. Finally, we demonstrate the effectiveness of the proposed framework through extensive empirical studies, including the problem settings of unsupervised domain adaptation and fair learning. Our codes and pretrained models are available at: https://github.com/illidanlab/FADE.
In modern industrial engineered systems, variant working conditions disturb the distributions of machines' operational data, which results in different feature distributions (DFD) problems for fault prognostics. Domain adaptation (DA) have been proved good at dealing DFD problems, and several deep DA-based methods have been also proposed in fault prognostics filed. However, existing methods refer to the DA tasks from one working condition to another, without considerations of transferring between datasets under complex working conditions. The prior distribution of working conditions will influence the distributions of machines' operational data, and few studies take prior distribution of working conditions into consideration of DA for fault prognostics. Thus, in this paper, a working-condition-based deep domain adaptation network (Deep wcDAN) is proposed to overcome the DFD problems caused by variant complex working conditions. In the proposed method, CNNs combines LSTMs with domain adaptive transfer technique to minimize the distribution discrepancy between training and testing datasets. Furthermore, a working-condition-based MMD (wcMMD) is proposed to optimize the DA process based on the prior distribution of each working condition. The performance of proposed model is evaluated and the negative transfer effects have been analyzed based on C-MAPSS datasets. The results show that the proposed method performs better than baseline methods on predicting remaining useful life (RUL) with DFD problems.INDEX TERMS Prognostics and health management, deep learning, transfer learning, domain adaptation, fault prognostics, remaining useful life.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.