We propose a novel method for weakly-supervised anomaly detection, where a limited number of labeled normal samples and a sufficient number of unlabeled samples are available for modeling. In particular, we seamlessly integrate label propagation with manifold graph learning into a support vector data description model. Consequently, the estimated manifold graph as well as its parameters will be adaptive to label propagation and benefit the anomaly detection performance. It is superior to most graph-based models that perform manifold graph learning separately by an independent step before label propagation. Theoretically, we derive a stability analysis based on the Rademacher complexity. Further, the effectiveness of the proposed method is demonstrated through several benchmark data sets and a real example of fault detection for high-speed train wheels. Note to Practitioners —This article provides a weakly-supervised anomaly detection method and addresses the challenge of insufficient normal samples for training. The proposed method integrates the adaptive embedded label propagation with adaptive manifold graph learning into a support vector data description model to additionally exploit the intrinsic data distribution information of the unlabeled data in the model formulation. It ensures that the results are jointly optimal for manifold representation and anomaly detection such that the detection accuracy is improved.
25 October 2022 in IEEE Transactions on Industrial Informatics
In practical industrial applications, the need for a large number of accurately labeled training samples is a significant challenge for fault detection tasks. However, labeling all training samples is expensive and prone to labeling errors, especially for early fault detection of wind turbine blades. This article proposes a labeling bias hypothesis. Assuming the labeler only needs to label parts of normal samples that are easy to judge, we design a probability ratio least-squares importance fitting (PRL-SIF) method based on variable homogeneity. Unlike other state-of-the-art positive unlabeled learning methods, PRL-SIF does not require knowledge of the class priors to achieve training. Furthermore, to better handle the multidimensional time-series data of wind turbines, we provide a data preprocessing method based on functional analysis to achieve time series feature extraction and dimensionality reduction. The effectiveness and robustness of the proposed method are verified on 23 real-world wind turbine datasets. Experimental results show that the proposed method can achieve nearly 90% accuracy while only labeling 20% of normal samples.
12 January 2022 in IEEE Transactions on Neural Networks and Learning Systems
We propose a probabilistic model for clustering spatially correlated functional data with multiple scalar covariates. The motivating application is to partition the 29 provinces of the Chinese mainland into a few groups characterized by the epidemic severity of COVID-19, while the spatial dependence and effects of risk factors are considered. It can be regarded as an extension of mixture models, which allows different subsets of covariates to influence the component weights and the component densities by modeling the parameters of the mixture as functions of the covariates. In this way, provinces with similar spatial factors are a priori more likely to be clustered together. Posterior predictive inference in this model formalizes the desired prediction. Further, the identifiability of the proposed model is analyzed, and sufficient conditions to guarantee ``generic'' identifiability are provided. An L₁-penalized estimator is developed to assist variable selection and robust estimation when the number of explanatory covariates is large. An efficient expectation-minimization algorithm is presented for parameter estimation. Simulation studies and real-data examples are presented to investigate the empirical performance of the proposed method. Finally, it is worth noting that the proposed model has a wide range of practical applications, e.g., health management, environmental science, ecological studies, and so on.
7 August 2024 in IEEE Transactions on Industrial Informatics
The challenges of class-imbalance and partially unknown training labels often arise in fault detection tasks. When these two problems occur simultaneously, existing imbalanced classification methods cannot be directly used due to the absence of the label, and the class-imbalance would lead to severe bias prediction. In this study, we proposed a novel deep imbalance separation network (deepImSN) framework that is capable of dealing with fault detection problems with both class imbalance and partially unknown labels. This framework integrates the one-class learning concept into the positive-unlabeled (PU) learning theory for the first time. It alleviates the bias of the class-imbalance while making full use of the limited label information in the PU set to optimize the feature space and guide model training. The proposed deepImSN is designed to be used in different scenarios. It can accurately complete the fault detection task whether only part of fault samples or normal samples are labeled, and the class-prior is known or unknown. Experimental results on real-world problems, such as high-speed rail wheels fault inspection and wafer map fault detection, demonstrate that deepImSN outperforms existing methods in various experimental conditions.
17 November 2025 in Measurement Science and Technology
Unmanned aerial vehicles (UAVs) are critical in modern industrial applications, yet their reliability is challenged by complex fault modes and limited labeled data for diagnosis. This paper proposes an interpretable weakly-supervised contrastive learning framework that simultaneously addresses accurate fault classification and precise component-level localization under partial annotations. Our method integrates multi-scale feature extraction using parallel one-dimensional convolutions with channel and temporal attention mechanisms to capture discriminative patterns while highlighting critical sensor channels and fault-relevant time steps. The framework incorporates contrastive learning with Gaussian noise injection and time warping augmentations to leverage unlabeled data, combined with a dynamic pseudo-label update mechanism that iteratively refines predictions. A specialized fine-grained localization module with fault-type-specific branches enables precise mapping of faults to specific components. Experimental results on the real UAV dataset demonstrate superior performance, achieving a macro -score of 0.89 with only 10% labeled data, outperforming state-of-the-art baselines. The framework also achieves near-perfect component-level localization under weak supervision. This work establishes a new paradigm for trustworthy industrial AI by integrating weak supervision efficiency with interpretable diagnostics, offering both high accuracy and operational transparency for UAV maintenance systems.