Abstract:Anomaly detection is a crucial means to detect unbalanced and multiclass abnormal data in industrial products. The existing anomaly detection model is usually suitable for products with a simple or regular background texture. However, there is a lack of a universal surface defect anomaly detection framework for railway products when facing the challenges: the complex background texture, huge intraclass variation, uneven distribution of interclass, and weak texture-featured defects. To fill this gap, an innovative generative adversarial network based on adaptive pyramid graph (APG) and variation residuals (APGVR-GAN) is proposed, aiming to improve the robustness of anomaly detection in railway products and other complex industrial supplies. First, the APG module is embedded in the encoder–decoder–encoder pattern, capturing the correlation description between neighbor regions, which is utilized to enhance the detection of abnormal defects with weak texture. Next, the variation residual module is employed to enhance the expression of various normal samples in the latent space and improve the identification ability for abnormal samples. Then, the dual-probability prototype loss is proposed to make different normal samples have more concentrated expression and more similar probability distribution centers in latent space. Finally, an adaptive focal-gate loss and a regularized log-likelihood loss are designed to overcome the imbalance problem in training samples with different background information. The effectiveness of the model is verified on three new railway datasets and three other industrial public datasets. The results show that in anomaly detection, APGVR-GAN on the three railway datasets, Northeastern University (NEU)-the rail surface defects dataset (RSDD)-1 dataset, NEU-RSDD-1 dataset, and RSDD area under curve (AUC) [that is the receiver operating characteristic curve (ROC)], can reach 0.89, 0.81, and 0.87, and its detection accuracy (ACC) reached 0.97, 0.96, and 0.92, which is much higher than other comparative anomaly detection algorithms. In the segmentation experiments on the RSDD dataset, compared with other known rail defect detection methods, the overall evaluation metric F-Score is 83.70 which is also better than other methods.

Real-Time Pantograph Anomaly Detection Using Unsupervised Deep Learning and K-Nearest Neighbor Classification

RTS-LCSS: A New Method for Real-Time Monitoring of Pantograph Structure

Semantic segmentation algorithm for pantograph based on multi-scale strip pooling attention mechanism and application research

Real-Time Accurate Deep Learning-Based Edge Detection for 3-D Pantograph Pose Status Inspection

On-line Detection of Pantograph Offset Based on Deep Learning

A Pantograph Horn Detection Method Based on Deep Learning Network

Pantograph Detection Algorithm with Complex Background and External Disturbances

Inception module and deep residual shrinkage network-based arc fault detection method for pantograph–catenary systems

Pantograph Pose Measurement Based on Keypoint Location and Edge Shape Optimization Under Complex Conditions

An Adaptive Pyramid Graph and Variation Residual-Based Anomaly Detection Network for Rail Surface Defects

Pantograph Slider Detection Architecture and Solution Based on Deep Learning

Data anomaly detection for structural health monitoring using a combination network of GANomaly and CNN

Dual-Path Reconstruction Guided Segmentation Network for Unsupervised Anomaly Detection and Localization.

Anomaly detection based on a deep graph convolutional neural network for reliability improvement

AnomalySeg: Deep Learning-Based Fast Anomaly Segmentation Approach for Surface Defect Detection

Electric Trucks Pantograph-Catenary Interaction Condition Monitoring Method Based on Semantic Segmentation Network and Linear Fitting

Design of Faster R-CNN-Based Fault Detection Method for Subway Vehicles

Intelligent Fault Diagnosis of Railway Pantograph using A Novel Graph Construction Methodology

Anomaly Detection for Data from Unmanned Systems via Improved Graph Neural Networks with Attention Mechanism

Online Intelligent Perception of Pantograph and Catenary System Status Based on Parameter Adaptation

Proactive Network Maintenance using Fast, Accurate Anomaly Localization and Classification on 1-D Data Series