Abstract:In modern industrial settings, the datasets available for mechanical fault diagnosis are often limited in size and exhibit significant imbalance posing a challenge to both the accuracy and stability of diagnostic processes. A particularly difficult task is the detection and localization of surface defects in textured materials. The complexity arises due to variable textures and the frequent scarcity of defective samples for preliminary testing. Conventional defect detection methods used in industry are typically complex, labor-intensive, and heavily dependent on expert knowledge due to the manual extraction of features and the complexity of the processing pipeline. To address these issues, the paper introduces an innovative imbalanced fault diagnosis approach leveraging the Fuzzy Clustering Generative Adversarial Network (FCGAN) model. The FCGAN is an advancement of the traditional Generative Adversarial Networks (GAN) which integrates fuzzy clustering into the discriminator's training process for unsupervised learning. The integration of fuzzy clustering enhances the generative network's performance. The model efficiency is further augmented by the feature extraction capabilities of FusionNet which undergoes structural optimization and conditional augmentation to generate diagnostic samples for machine faults. The effectiveness of the model is validated using two publicly accessible datasets i.e., DAGM 2007 synthetic and CCSD-NL Magnetic-Tile-Defect. The proposed method which is tested on two famous datasets, including DAGM 2007 synthetic and CCSD-NL Magnetic-Tile-Defect, has achieved prediction accuracy of 95.21% and 96.24% with a standard deviation of 65.78% and 57.44%, respectively. The results have been compared with other DL and traditional methods, including GAN, CNN, and adaptive deep CNN. The comparisons show that the proposed FCGAN data-driven fault diagnosis method has achieved significant improvements.

FD Technology for HSs based on Deep Convolutional Generative Adversarial Networks

ECNN: Intelligent Fault Diagnosis Method Using Efficient Convolutional Neural Network

Deep continuous convolutional networks for fault diagnosis

Multireceptive Field Denoising Residual Convolutional Networks for Fault Diagnosis

Wasserstein Generative Adversarial Network and Convolutional Neural Network (WG-CNN) for Bearing Fault Diagnosis

Application of Generative Adversarial Networks for Intelligent Fault Diagnosis

A New Convolutional Neural Network-Based Data-Driven Fault Diagnosis Method

FTT-NAS: Discovering Fault-Tolerant Convolutional Neural Architecture

An Intelligent Fault Diagnosis Based on Adversarial Generating Module and Semi-supervised Convolutional Neural Network

A transfer convolutional neural network for fault diagnosis based on ResNet-50

Fault identification of product design using fuzzy clustering generative adversarial network (FCGAN) model

DSMT-1DCNN: Densely supervised multitask 1DCNN for fault diagnosis

Global contextual feature aggregation networks with multiscale attention mechanism for mechanical fault diagnosis under non-stationary conditions

SKND-TSACNN: A novel time-scale adaptive CNN framework for fault diagnosis of rotating machinery

Fault Diagnosis of Nuclear Power Plants Based on 1D-CNN with Dual Attention Mechanism

A Hierarchical Training-Convolutional Neural Network for Imbalanced Fault Diagnosis in Complex Equipment

A New Deep Learning Model for Fault Diagnosis with Good Anti-Noise and Domain Adaptation Ability on Raw Vibration Signals

Multiattention-Based Feature Aggregation Convolutional Networks with Dual Focal Loss for Fault Diagnosis of Rotating Machinery under Data Imbalance Conditions

Fault diagnosis of reducers based on digital twins and deep learning