MIM-Graph: A multi-sensor network approach for fault diagnosis of HSR Bogie bearings at the IoT edge via mutual information maximization

A pruning-aware dynamic slimmable network using meta-gradients for high-speed train bogie bearing fault diagnosis

Although intelligent fault diagnosis achieves remarkable achievements, computation efficiency is a commonly ignored problem in existing studies. Pruning network networks enable us to find compact models that not only retain the diagnosis accuracy, but also consume fewer computation resources for training and inference. However, current studies are inefficient in building a saliency criterion for parameter importance evaluation. In this paper, we identify a pruning-aware dynamic slimmable network which uses the meta-gradients to select unnecessary parameters to prune. The slimmable network is designed with two sub-networks, called the classifier and the evaluator to generate meta-gradients for parameter pruning. And an iterative pruning algorithm is proposed to improve computation efficiency while retaining diagnosis performance. Our method is verified on a high-precision bogie fault simulation experimental data set and achieves state-of-art performance in terms of accuracy and efficiency compared with existing studies.

Deep learning based approaches for intelligent industrial machinery health management and fault diagnosis in resource-constrained environments

Industry 4.0 represents the fourth industrial revolution, which is characterized by the incorporation of digital technologies, the Internet of Things (IoT), artificial intelligence, big data, and other advanced technologies into industrial processes. Industrial Machinery Health Management (IMHM) is a crucial element, based on the Industrial Internet of Things (IIoT), which focuses on monitoring the health and condition of industrial machinery. The academic community has focused on various aspects of IMHM, such as prognostic maintenance, condition monitoring, estimation of remaining useful life (RUL), intelligent fault diagnosis (IFD), and architectures based on edge computing. Each of these categories holds its own significance in the context of industrial processes. In this survey, we specifically examine the research on RUL prediction, edge-based architectures, and intelligent fault diagnosis, with a primary focus on the domain of intelligent fault diagnosis. The importance of IFD methods in ensuring the smooth execution of industrial processes has become increasingly evident. However, most methods are formulated under the assumption of complete, balanced, and abundant data, which often does not align with real-world engineering scenarios. The difficulties linked to these classifications of IMHM have received noteworthy attention from the research community, leading to a substantial number of published papers on the topic. While there are existing comprehensive reviews that address major challenges and limitations in this field, there is still a gap in thoroughly investigating research perspectives across RUL prediction, edge-based architectures, and complete intelligent fault diagnosis processes. To fill this gap, we undertake a comprehensive survey that reviews and discusses research achievements in this domain, specifically focusing on IFD. Initially, we classify the existing IFD methods into three distinct perspectives: the method of processing data, which aims to optimize inputs for the intelligent fault diagnosis model and mitigate limitations in the training sample set; the method of constructing the model, which involves designing the structure and features of the model to enhance its resilience to challenges; and the method of optimizing training, which focuses on refining the training process for intelligent fault diagnosis models and emphasizes the importance of ideal data in the training process. Subsequently, the survey covers techniques related to RUL prediction and edge-cloud architectures for resource-constrained environments. Finally, this survey consolidates the outlook on relevant issues in IMHM, explores potential solutions, and offers practical recommendations for further consideration.

Research on spatial-temporal synergistic sensor fault diagnosis method for top-blowing furnace

Top-blowing furnace systems, characterized by a large number of sensors and harsh working environments, are prone to sensor failures due to factors like component aging and external interference. These failures can significantly impact the system's safe and reliable operation. However, traditional sensor fault diagnosis methods often neglect the exploration of spatial-temporal characteristics and focus solely on learning temporal relationships between sensors, failing to effectively consider their spatial relationships. In this study, we propose a spatial correlation model based on the maximal information-based graph convolutional network (MI-GCN) by constructing a sensor network knowledge graph using maximal mutual information. The MI-GCN leverages the graph convolution mechanism to extract multi-scale spatial features and capture the spatial relationships between sensors. Additionally, we develop a spatial-temporal graph-level prediction model, known as the spatial-temporal graph transformer (STGT), to extract temporal features. By combining the spatial features extracted by the MI-GCN with the temporal features captured by the STGT, accurate predictions can be achieved. Sensor fault diagnosis is conducted by analysing the normalized residuals between the predicted values and the ground truth. Finally, the feasibility and effectiveness of the proposed method are validated using test data from a top-blowing furnace system in the nickel smelting process.

Unsupervised graph anomaly detection with discriminative embedding similarity for viscoelastic sandwich cylindrical structures

In the detection of slipping anomalies in viscoelastic sandwich cylindrical structures (VSCS), conventional methods may encounter challenges due to the extremely rare and weak nature of slipping signals. This study focuses on normal signals and introduces unsupervised graph representation learning (UGRL) with discriminative embedding similarity for VSCS's detection. UGRL involves data preprocessing, model embedding, and matrix reconstructing. Association graphs are constructed based on sample similarities for yielding adjacency and attribute matrices. Subsequently, the matrices undergo embedding and reconstruction via various network modules to enhance graph data characterization. Detection indicators are derived by calculating embedding similarities and reconstruction errors, and thresholds are constructed using these indicators to enable efficient anomaly detection. The experiments in VSCS slipping dataset effectively indicate the superiority of the proposed method.