In this survey paper, we systematically summarize existing literature on bearing fault diagnostics with deep learning (DL) algorithms. While conventional machine learning (ML) methods, including artificial neural network, principal component analysis, support vector machines, etc., have been successfully applied to the detection and categorization of bearing faults for decades, recent developments in DL algorithms in the last five years have sparked renewed interest in both industry and academia for intelligent machine health monitoring. In this paper, we first provide a brief review of conventional ML methods, before taking a deep dive into the state-of-the-art DL algorithms for bearing fault applications. Specifically, the superiority of DL based methods are analyzed in terms of fault feature extraction and classification performances; many new functionalities enabled by DL techniques are also summarized. In addition, to obtain a more intuitive insight, a comparative study is conducted on the classification accuracy of different algorithms utilizing the open source Case Western Reserve University (CWRU) bearing dataset. Finally, to facilitate the transition on applying various DL algorithms to bearing fault diagnostics, detailed recommendations and suggestions are provided for specific application conditions. Future research directions to further enhance the performance of DL algorithms on health monitoring are also discussed.

The rapid development of artificial intelligence and deep learning technology has provided many opportunities to further enhance the safety, stability, and accuracy of industrial Cyber-Physical Systems (CPS). As indispensable components to many mission-critical CPS assets and equipment, mechanical bearings need to be monitored to identify any trace of abnormal conditions. Most of the data-driven approaches applied to bearing fault diagnosis up-to-date are trained using a large amount of fault data collected a priori. In many practical applications, however, it can be unsafe and time-consuming to collect sufficient data samples for each fault category, making it challenging to train a robust classifier. In this paper, we propose a few-shot learning framework for bearing fault diagnosis based on model-agnostic meta-learning (MAML), which targets for training an effective fault classifier using limited data. In addition, it can leverage the training data and learn to identify new fault scenarios more efficiently. Case studies on the generalization to new artificial faults show that the proposed framework achieves an overall accuracy up to 25% higher than a Siamese-network-based benchmark study. Finally, the robustness of the generalization capability of MAML is further validated by applying the algorithm to identify real bearing damages using data from artificial damages, which compares favorably against 6 state-of-the-art few-shot learning algorithms using consistent test environments.

The detection of rolling-element bearing fault can be accomplished by monitoring and interpreting a variety of signals, including the vibration, the acoustic noise, and the stator current. The existence of a bearing fault as well as its specific fault type can be readily determined by performing frequency spectral analysis on the monitored signals with various signal processing techniques. However, this traditional approach, despite being simple and intuitive, is not able to identify the severity of a bearing fault in a quantitatively manner. Moreover, it is often-times tedious and time-consuming to apply this approach to electric machines with different power ratings, as the bearing fault threshold values need to be manually calibrated for each motor running at every possible speed and carrying any possible load. This paper thus proposes a quantitative approach to estimate the bearing fault severity based on the air gap displacement profile, which is reconstructed from the mutual inductance variation profile estimated from a quantitative electrical model that only takes the stator current as input. In addition, the accuracy of the developed electrical model and the estimated bearing fault severity are validated by the simulation and experimental results, and the explicit air gap variation profile is reconstructed with the superposition of multiple Fourier Series terms estimated from the stator current via the proposed scheme. The proposed method offers a quantitative and universal bearing fault indicator that can be used to monitor bearing faults in induction machines with any power ratings and operating under any speed and load conditions.

In this paper, multi-channel capacitive sensor arrays based on microstrip band-stop filters are studied. The sensor arrays can be used to detect the proximity of objects at different positions and directions. Each capacitive sensing structure in the array is connected to an inductive element to form resonance at different frequencies. The resonances are designed to be isolated in the frequency spectrum, such that the change in one channel does not affect resonances at other channels. The inductive element associated with each capacitive sensor can be surface-mounted inductors, integrated microstrip inductors or metamaterial-inspired structures. We show that by using metamaterial split-ring structures coupled to a microstrip line, the quality factor of each resonance can be greatly improved compared to conventional surface-mounted or microstrip meander inductors. With such a microstrip-coupled split-ring design, more sensing elements can be integrated in the same frequency spectrum, and the sensitivity can be greatly improved.

The rapid growth of artificial intelligence is revolutionizing classical engineering society, offering novel approaches to material and structural design and analysis. Among various scientific machine learning techniques, physics-informed neural network (PINN) has been one of the most researched subjects, for its ability to incorporate physics prior knowledge into model training. However, the intrinsic continuity requirement of PINN demands the adoption of domain decomposition when multiple materials with distinct properties exist. This greatly complicates the gradient computation of design features, restricting the application of PINN to structural shape optimization. To address this, we present a novel framework that employs neural network coordinate projection for shape optimization within PINN. This technique allows for direct mapping from a standard shape to its optimal counterpart, optimizing the design objective without the need for traditional transition functions or the definition of intermediate material properties. Our method demonstrates a high degree of adaptability, allowing the incorporation of diverse constraints and objectives directly as training penalties. The proposed approach is tested on magnetostatic problems for iron core shape optimization, a scenario typically plagued by the high permeability contrast between materials. Validation with finite-element analysis confirms the accuracy and efficiency of our approach. The results highlight the framework’s capability as a viable tool for shape optimization in complex material design tasks.

In this paper, we propose a topological data analysis (TDA) method for the processing of induction motor stator current data, and apply it to the detection and quantification of eccentricity faults. Traditionally, physics-based models and involved signal processing techniques are required to identify and extract the subtle frequency components in current data related to a particular fault. We show that TDA offers an alternative way to extract fault related features, and effectively distinguish data from different fault conditions. We will introduce TDA method and the procedure of extracting topological features from time-domain data, and apply it to induction motor current data measured under different eccentricity fault conditions. We show that while the raw time-domain data are very challenging to distinguish, the extracted topological features from these data are distinct and highly associated with eccentricity fault level. With TDA processed data, we can effectively train machine learning models to predict fault levels with good accuracy, even for new data from eccentricity levels that are not seen in the training data. The proposed method is model-free, and only requires a small segment of time-domain data to make prediction. These advantages make it attractive for a wide range of data-driven fault detection applications.