27 February 2024 in IEEE Transactions on Industrial Informatics
Imbalanced data constitute a significant challenge in intelligent fault diagnosis cases because they can result in degraded diagnosis accuracy, which can in turn jeopardize the safety and reliability of industrial equipment. Generative adversarial networks (GANs) have been effectively used as common data augmentation methods to address this issue. However, their training process is difficult to perform and prone to mode collapse. Therefore, this article proposes a novel data augmentation method grounded in a diffusion model. The proposed method generates samples through physical simulation rather than adversarial training, which avoids the instability and mode collapse issues faced by GANs, leading to a more stable training process. Moreover, the proposed method utilizes the characteristics of gradual diffusion and random sampling to enhance the authenticity and diversity of sample generation. In addition, in terms of evaluating generation models, most existing works do not have a unified and thorough evaluation framework. Therefore, a comprehensive evaluation framework is proposed to effectively and comprehensively evaluate the performance of data augmentation models. Finally, the proposed method is evaluated using an open-source dataset and two actual testbeds to validate its effectiveness. The experimental results show that our method can generate higher quality and more diverse pseudosamples, and achieve superior fault diagnosis performance under imbalanced data. Specifically, our approach achieves diagnosis accuracies of 97.00%, 96.48%, and 98.30% on the three different datasets, all of which are superior to those of the compared state-of-the-art data augmentation algorithms.
1 December 2023 in IEEE Transactions on Instrumentation and Measurement
The varying working conditions of rolling bearings lead to data distribution shifts, which presents an obstacle to intelligent fault diagnosis. To alleviate the performance degradation, unsupervised domain adaptation (DA) is used to address the issue where the test set has a different distribution from the training set. Most existing intelligent transfer fault diagnosis methods align features by using either a fixed metric or a domain discriminator to tackle data distribution divergence. However, the alignment of decision boundaries in fault classifiers has not been thoroughly studied, causing ambiguous predictions on unlabeled target domain samples and catastrophic performance degradation in fault diagnosis. To overcome the deficiencies, a novel diagnosis framework named prototype and stochastic neural network-based twice adversarial DA (PSNN-TADA) is proposed to extract domain-invariant fault features under varying working conditions. In particular, PSNN-TADA has two adversarial training processes at different levels. First, PSNN-TADA has a prototype-based discriminator using multimodal information to achieve conditional domain alignment. Second, PSNN-TADA learns reliable decision boundaries from an adversarial process between the extractor and the stochastic classifier, which further improves the accuracy of the classifier on target fault samples. The proposed method is comprehensively tested on both an open-access dataset and a practical fault diagnosis platform, and extensive experimental results illustrate that PSNN-TADA achieves superior adaptation performance compared with the state-of-the-art.
12 February 2025 in IEEE Internet of Things Journal
Wheeled robots play a crucial role in industrial internet of things (IIoT)-enabled manufacturing environments, and ensuring their reliable operation is essential for production efficiency and safety. However, their inherent complexity makes them prone to faults, while limited fault data results in imbalanced datasets, posing challenges for deep learning-based fault diagnosis models. Existing denoising diffusion probabilistic model (DDPM)-based fault diagnosis methods tend to address single-channel scenarios, ignoring the graph relationships inherent in multi-channel sensor data. Furthermore, current graph-based DDPM models also struggle in wheeled robot scenarios due to its multi-modal nature. To address these challenges, we propose an enhanced DDPM-based method for imbalanced fault diagnosis of wheeled robots. Our method integrates graph operations into the noise prediction network of the DDPM framework, enabling efficient modeling of the complex spatial-temporal relations in multi-modal graphical sequence data via a newly designed spatial-temporal graph U-Net (STGU-Net). Additionally, we present a dynamic degradation mechanism for the prior graph, simulating gradual structural changes during the diffusion process. Extensive experiments on a real-world wheeled robot platform demonstrate the superiority of the proposed model over state-of-the-art methods from multiple perspectives, showcasing its effectiveness in generating high-quality data and mitigating the imbalance problem in wheeled robot fault diagnosis.
11 February 2025 in IEEE Sensors Journal
In complex industrial environments, mechanical devices predominantly operate under normal conditions, resulting in a shortage of available fault samples. This imbalance significantly impedes the effectiveness of intelligent fault diagnosis approaches. To overcome the challenge, an innovative generative adversarial network (GAN) model is proposed in this study, termed the bidirectional local-global interaction guided GAN with discriminator gradient gap regularization (Bi-Interaction GAN). Firstly, the generator is designed to incorporate a bidirectional interaction mechanism that accounts for the interaction between local and global features of vibration signals, effectively capturing the deep relationships between local and global information within the bearing vibration signals with limited datasets. Subsequently, the stability of the model is enhanced by introducing a novel loss function that integrates discriminator gradient gap regularization with the Wasserstein distance. This approach aims to minimize the gradient discrepancy between real and generated samples within the discriminator, thereby facilitating a more stable training process. Extensive comparative experiments are carried out to validate the proposed approach using a widely utilized dataset and realistic experimental configuration. The results indicate that the Bi-Interaction GAN outperforms existing state-of-the-art GAN models in generating superior-quality samples while achieving high classification accuracy.
14 January 2026 in IEEE/ASME Transactions on Mechatronics
As robots are increasingly integrated into industrial applications, robust fault diagnosis is becoming increasingly crucial to ensure operational safety and minimize unplanned downtime. Graph neural networks have emerged as a promising solution for handling multisensor fault diagnosis tasks, but conventional approaches for constructing spatial–temporal graphs from robotic sensor data tend to rely on either static physical knowledge or purely statistical relationships, each with inherent limitations in changing operational contexts. To address these challenges, this article proposes a novel hybrid architecture driven by large language model (LLM), which integrates global physical knowledge with localized statistical patterns to guide the dynamic construction of spatial–temporal graphs. A concrete operational method is further proposed to structure textual inputs for the LLM, allowing it to integrate domain knowledge, task objectives, and fusion logic into graph representations that are physically grounded and responsive to real-time conditions. Building on the outputs of this LLM-driven process, we design a lightweight spatial–temporal graph network that incorporates the dynamically generated graphs into fault inference. Finally, we validate our approach on two real-world robot platforms, and demonstrating notable improvements in diagnostic accuracy, adaptability, and robustness compared with existing state-of-the-art methods.
2 September 2024 in IEEE Transactions on Instrumentation and Measurement
Fault diagnosis is of great significance for ensuring the reliability and safety of electromechanical equipment. However, in practical applications, the cost of labeling samples is high, which poses difficulties for most supervised fault diagnosis models that rely on a large number of labeled samples for training. Therefore, this paper proposes a novel semi-supervised fault diagnosis method for bearings under low labeled rates, which has the following characteristics: firstly, the first convolutional layer of the feature extraction network is reconstructed using the idea of wavelet transform, addresses the challenges of ineffective learning of the layer caused by long gradient propagation distance, consequently enhancing the feature extraction capability. Secondly, a semi-supervised loss is defined through pseudo-labeling, which makes the boundaries between different classes more explicit on the feature space. Thirdly, A new learning strategy is also designed for this process to overcome the confirmation bias problem. The experimental results based on a widely used public dataset and two real testbeds show that the proposed method achieve satisfactory diagnosis performance at labeled rates as low as 1.25% to 5%, and is superior to other classic and latest models.