A Hybrid Bayesian ICA-LSTM Framework for Unsupervised-Like Anomaly Detection in Rolling Element Bearings

Rolling bearings play an important role in rotating machinery. According to statistics, rolling bearings cause one-third faults of rotating machinery. Once a rolling bearing malfunctions, it may induce maintenance, affect work efficiency, or even cause the entire equipment to malfunction. Therefore, accurately determining the operating status of bearings is of great significance for maintaining the health of the rotating machinery. Most current fault detections of rolling bearing works focus on traditional anomaly detection models which assume the training set to follow the same distribution of test set. This assumption does not hold in fault detection of rolling bearings across different conditions and traditional anomaly detection models may be invalid. This paper introduces domain adaptation anomaly detection (DAAD) in the fault detection of rolling bearings to address this issue. DAAD can adapt anomaly detection across different distributions. The experiments of rolling bearing fault detection under single condition or across different condition show that DAAD is superior to most of the traditional anomaly detection models.

Condition monitoring of roller element bearings is of considerable interest to many industries since faulty roller element bearings are known to cause the majority of problems in rotating machinery. The class of failure modes in roller bearings that has received the most attention include spalling or localized defect in one or more of the bearing components. This refers to dislodgement of a portion of the contact surface in one or more of the bearing components in the presence of various kinds of stresses like rotor imbalance, speed, bad lubrication, heavy radial and axial loads. Whereas substantial work has been done in the detection of spalls in bearing components, studies show that one of the primary root causes of spalls in a bearing tends to be ineffective lubrication resulting either from lack of lubrication conditions or the presence of contaminants in the lubricant. Leading bearing companies have indicated that incorrect lubrication can account for more than 90% of bearing failures because of which lubrication can be a key influence that can make or break bearing service and life. This emphasizes the need to develop techniques to sense lubrication anomalies as well as provide information needed to act upon them. In other words, monitoring the health and effectiveness of the bearing lubricant should be at the forefront of a condition-monitoring program for bearings. In this paper, we describe experiments, analyses and results obtained for monitoring and detecting anomalies in bearing lubrication for oil-lubricated bearings. More specifically, we consider two kinds of lubrication anomalies: lack of lubrication and contamination of lubrication. Our work involves the use of techniques in sensing and analyses of acoustic emissions from the bearing housing for detection of anomalous lubrication conditions of the above classes. We explore these techniques by conducting controlled experiments where, conditions equivalent to the appropriate fault condition are simulated in a test rig and AE readings are recorded; as a baseline, we also record vibration readings using traditional accelerometers. Our analyses will consist of extracting features from the AE signal that can suitably distinguish between normal and abnormal lubrication conditions. Additional analyses that can potentially be used to understand the degree of severity of the abnormality will also be presented.

The detection of faults and operational abnormalities in rotating machine elements like rolling element bearings and gears requires information about kinematic properties, such as ball-pass and gear mesh frequencies. Typically, condition-monitoring experts obtain such information from the manufacturers for diagnostics purposes. However, the reliability of such information can be compromised during installation and maintenance, for example, if components are replaced and do not match the documented specifications. Thus, methods enabling verification and online extraction of such kinematic properties are needed to improve diagnostic reliability. Unsupervised machine learning methods, like sparse coding with dictionary learning, enable automatic modeling and characterization of repeating signal structures in the time domain, which are naturally generated by rotating equipment. Sparse coding with dictionary learning represents a vibration signal as a linear superposition of noise and atomic waveforms. The activation rate of the atomic waveforms typically possesses a cyclic nature in rotating environments, similar to how bearing kinematic frequencies correlate with faults in a rolling element bearing. However, there is no explicit relationship between the activation rates of the atoms and the bearing kinematic frequencies. This motivates this investigation of the possibility to extract bearing kinematic frequencies from sparse representations. Former work describes the use of dictionary learning for the detection of anomalies in rolling element bearings. In this paper, we describe how a similar unsupervised machine learning method can be used to extract kinematic frequencies of bearings and gears, for example for anomaly detection purposes and comparisons with an expected signature. We study the activation rates and changes of atoms learned from vibration signals in two case studies. The first case is based on data from a well-known controlled experiment with faults seeded in the bearings. The second case is based on a public dataset recorded from the high-speed shaft of a wind turbine with a bearing failure. Furthermore, we compare the activation rates and weights of the atoms to the bearing kinematic frequencies and harmonics. Sparse coding with dictionary learning offers a possibility for self-learning of the kinematic frequencies of a bearing, which can be useful for the further improvement of automated anomaly detection methods in condition monitoring.

Various remaining useful life (RUL) prediction methods, encompassing model-based, data-driven, and hybrid methods, have been developed and successfully applied to prognostics and health management for diverse rolling bearing. Hybrid methods that integrate the merits of model-based and data-driven methods have garnered significant attention. However, the effective integration of the two methods to address the randomness in rolling bearing full life cycle processes remains a significant challenge. To overcome the challenge, this paper proposes a data and model synergy-driven RUL prediction framework that includes two data and model synergy strategies. First, a convolutional stacked bidirectional long short-term memory network with temporal attention mechanism is established to construct Health Index (HI). The RUL prediction is achieved based on HI and polynomial model. Second, a three-phase degradation model based on the Wiener process is developed by considering the evolutionary pattern of different degradation phases. Then, two synergy strategies are designed. Strategy 1: HI is adopted as the observation value for online updating of physics degradation model parameters under Bayesian framework, and the RUL prediction results are obtained from the physics degradation model. Strategy 2: The RUL prediction results from the data-driven and physics-based model are weighted linearly combined to improve the overall prediction accuracy. The effectiveness of the proposed model is verified using two bearing full life cycle datasets. The results indicate that the proposed approach can accommodate both short-term and long-term RUL predictions, outperforming state-of-the-art single models.

Rolling element bearings are essential components in heavy-duty machinery such as drilling rigs, mining equipment, wind turbine rotors, and helicopter swash plates. Their failure can cause costly downtime and major repairs, making early fault detection and continuous monitoring critical. Traditional vibration-based methods, although widely applied, often struggle to identify incipient faults under variable operating conditions. Acoustic Emission (AE) sensing has emerged as a promising alternative, offering high sensitivity to microscopic damage and transient events. However, many AE-based diagnostic approaches rely on characteristic fault frequency extraction, which requires precise information about rotational speed and bearing geometry, limiting their practicality. This study introduces an AE-based fault localization method that combines wavelet-derived features with a Random Forest classifier. Experiments on the UORED-AFCLS dataset show that energy-based features achieve near-perfect classification accuracy across different fault types. The results highlight AE’s potential, when integrated with machine learning, to provide an effective and reliable framework for early fault detection in complex industrial environments.

Water-lubricated bearings are critical in ship propulsion, and wear accelerates under low-speed, heavy-load conditions, making early detection essential. Existing studies mainly focus on rolling bearings and employ time- or frequency-domain analyses. While effective for obvious impacts, these methods often fail to capture weak transient signals associated with early wear. Directional dependence is also neglected, although wear responses of water-lubricated stern bearings vary with direction, reducing the sensitivity and interpretability of conventional frameworks. To address these issues, this study proposes a direction-sensitive detection framework that integrates time-domain features with Ensemble Empirical Mode Decomposition (EEMD)-based multi-scale energy features for multi-dimensional evaluation. Features are standardized using Z score normalization and reduced via Principal Component Analysis (PCA), and their contributions quantified using Spearman correlation and SHapley Additive exPlanations (SHAP) values to identify direction-sensitive indicators. Results show that vertical radial Kurtosis features are most sensitive to early wear, while horizontal radial features provide robust supplementary information. High-frequency time–frequency features respond more sensitively to early wear, whereas time-domain features exhibit stronger directional specificity. Based on these findings, a multi-dimensional detection framework is established, with vertical radial vibrations as core early-warning indicators and horizontal radial vibrations as robust supplements, providing a high-sensitivity, adaptable solution for early wear detection of water-lubricated bearings.

Safety evaluation is important during both the pre-market clinical trials and post-market surveillance. In either a pre-market or post-market setting wherein the safety of a device is compared to that of a control device, it is desirable to identify any difference in the safety between two devices as expeditiously as possible. Here, we introduce the Bayesian hierarchical framework for the safety assessment in two-arm clinical trials, with signal detection accomplished by evaluating the effect size of each adverse event (AE) measured by odds ratio or relative risk. The framework starts with a standard hierarchical Bayesian model with a parametric distribution as a common prior for the effect sizes of all AEs. Then, it is extended with a non-parametric prior, Dirichlet Process Prior, to allow for more flexibility. After that, to account for the rare events in some trials, it is further extended with the option of additional zero-inflated parameters and calculation of regularized effect size. Extra incorporation of exposure-time information is available under each framework. The performance of the proposed technique, along with its extensions, is studied by simulation. The application of the proposed Bayesian framework is demonstrated by data from a two-device clinical trial, the newer left ventricular assist device (LVAD) and the existing LVAD. The Bayesian analysis result is then compared to a traditional frequentist technique. Through both simulation and application, the proposed Bayesian technique is shown to be robust to the selection of priors of the variance component, and has comparative and under some scenarios even better performance than the frequentist technique. Overall, the developed Bayesian framework is a feasible alternative to the frequentist method for safety evaluation of medical device clinical trials.

Early fault detection and diagnosis of rolling element bearings (REBs) is crucial for preventing unexpected machine failures, thereby ensuring operational reliability and lowering maintenance costs. Component signals decomposed by local mean decomposition (LMD) are effective tools for describing non-stationary and nonlinear signals across multiple frequency scales. Inspired by recent advancements in graph modeling, a LMD enhanced graph spectrum analysis approach is proposed for condition monitoring of REBs. Specifically, the component signals decomposed by LMD are utilized as inputs to construct the graph spectrum, through which the degradation characteristics of REB’s health status are extracted based on multi-scale correlation information, and health indicators describing the dynamic characteristics of REBs are obtained. According to the generated health indicator, a null hypothesis testing is conducted for early fault detection. Once an early warning is triggered, subsequent data will be marked as suspicious segments. The graph spectrum will be further used to provide discriminative features, and fault identification for suspicious segments will be achieved using k-Nearest Neighbors (KNN). By constructing multi-scale graph spectrum, the proposed approach effectively captures the degradation information of REBs while enabling precise fault characterization, providing a unified framework for early fault detection and fault identification. Experiments for fault detection and diagnosis are conducted using the XJTU-SY and CWRU datasets. Comparative studies with state-of-the-art methods are conducted to demonstrate the effectiveness and superiority of the proposed approach.

Traditional methods for Bridge Influence Line (BIL) identification generally necessitate traffic interruption to achieve reliable results, significantly restricting their practical applicability. Under uninterrupted traffic conditions, bridge responses exhibiting substantial randomness due to unknown traffic loading, significantly complicating parameter identification. To address this issue, this study proposes an instrumented-vehicle-aided detection framework. An instrumented vehicle with known weight and speed traverses the bridge, while computer version technology tracks the positions of all vehicles. And a new physical model is developed to indicate the relationship of the measured responses, BIL, vehicle positions, known weights of instrumented vehicles, unknown BIL and interfering vehicles weights (IVWs). The identification of unknown BIL and IVWs is formulated as a multi-parameter inverse optimization problem and addressed within a Bayesian inference framework. The physical model is firstly embedded into the Bayesian framework by deriving the posterior distribution of the unknown parameters. After optimal conditional function of BIL is derived and solution set of IVWs is defined, a coupled iterative optimization algorithm, optimizing different groups of parameters iteratively, is then designed to simultaneously estimate the BIL and IVWs. At last, the uncertainty is quantified through a derived Hessian matrix. Numerical simulations confirm that the proposed method accurately identifies both BIL and IVWs under stochastic traffic flows. Additionally, two field tests validate the effectiveness and practicality of the approach under uninterrupted traffic conditions.

Sparsity enforced time–frequency decomposition in the Bayesian framework for bearing fault feature extraction under time-varying conditions

Roller bearings are critical components in various mechanical systems, and the timely detection of potential failures is essential for preventing costly downtimes and avoiding substantial machinery breakdown. This research focuses on finding and verifying a robust method that can detect failures early, without creating false positive failure states. Therefore, this paper introduces a novel algorithm for the early detection of roller bearing failures, particularly tailored to high-precision bearings and automotive test bed systems. The featured method (AFI-Advanced Failure Indicator) utilizes the Fast Fourier Transform (FFT) of wideband accelerometers to calculate the spectral content of vibration signals emitted by roller bearings. By calculating the frequency bands and tracking the movement of these bands within the spectra, the method provides an indicator of the machinery's health, mainly focusing on the early stages of bearing failure. The calculated channel can be used as a trend indicator, enabling the method to identify subtle deviations associated with impending failures. The AFI algorithm incorporates a non-static limit through moving average calculations and volatility analysis methods to determine critical changes in the signal. This thresholding mechanism ensures the algorithm's responsiveness to variations in operating conditions and environmental factors, contributing to its robustness in diverse industrial settings. Further refinement was achieved through an outlier detection filter, which reduces false positives and enhances the algorithm's accuracy in identifying genuine deviations from the normal operational state. To benchmark the developed algorithm, it was compared with three industry-standard algorithms: VRMS calculations per ISO 10813-3, Mean Absolute Value of Extremums (MAVE), and Envelope Frequency Band (EFB). This comparative analysis aimed to evaluate the efficacy of the novel algorithm against the established methods in the field, providing valuable insights into its potential advantages and limitations. In summary, this paper presents an innovative algorithm for the early detection of roller bearing failures, leveraging FFT-based spectral analysis, trend monitoring, adaptive thresholding, and outlier detection. Its ability to confirm the first failure state underscores the algorithm's effectiveness.

A semi-supervised Support Vector Data Description-based fault detection method for rolling element bearings based on cyclic spectral analysis

It is difficult to establish a classification and recognition model of machinery and equipment based on labeled samples in the actual industrial environment because of the imperfect fault modes and data missing. To solve this problem, a semisupervised anomaly detection method based on masked autoencoders of distribution estimation (MADE) is designed. First, the Mel-frequency cepstrum coefficient (MFCC) is employed to extract fault features from vibration signals of rolling bearings. Then, a group of mask matrices are set on each hidden layer to overcome the perfect reconstruction problem of the autoencoders' input, and the full-connection probability of reconstruction is used to replace the reconstruction error and adopted as the anomaly score. Finally, the diagnostic threshold is determined according to the Youden index. Experimental results show that the MADE method can extract fault-sensitive features from a noisy industrial environment and introduce mask matrices renders to make the network autoregressive, thus solving the problem of perfect reconstruction of autoencoders. It is verified based on three rolling bearing datasets that the accuracy, precision, recall, and F1-score of the proposed method are confirmed to be all 100%. Moreover, the accuracy of the proposed method is 17.19% higher than that of the memory-inhibition method on the rolling bearing dataset provided by the Center for Intelligent Maintenance Systems (IMS) in University of Cincinnati (USA). The accuracy of the proposed method is also improved compared with other state-of-the-art anomaly detection methods.

Heavy rain deteriorates the video quality of outdoor imaging equipments. In order to improve video clearness, image-based and sensor-based methods are adopted for rain detection. In earlier literature, image-based detection methods fall into spatio-based and temporal-based categories. In this paper, we propose a new image-based method by exploring spatio-temporal united constraints in a Bayesian framework. In our framework, rain temporal motion is assumed to be Pathological Motion (PM), which is more suitable to time-varying character of rain steaks. Temporal displaced frame discontinuity and spatial Gaussian mixture model are utilized in the whole framework. Iterated expectation maximization solving method is taken for Gaussian parameters estimation. Pixels state estimation is finished by an iterated optimization method in Bayesian probability formulation. The experimental results highlight the advantage of our method in rain detection.

An artificial-intelligence (AI)-based method for fault diagnosis is a strong candidate for industrial applications in the health management of rolling bearings. However, traditional fault diagnosis methods fail to improve the detection accuracy because they only extract a single feature and have limitations in feature representation. In addition, advanced object detection frameworks such as region-based convolutional neural networks have not yet been applied in fault diagnosis. To this end, a fault diagnosis model using a Time-Frequency Region-Based Convolutional Neural Network (TF-RCNN) is proposed in this paper. This method was mainly adopted to extract multiple regions that can characterize fault features from the Time-Frequency Representation (TFR). Specifically, an attention module was introduced so the model could focus on representative features. The existing classification strategy was also enhanced to perform multiple types of fault classification. Finally, an end-to-end rolling bearing fault diagnosis framework based on the TF-RCNN was developed with the aforementioned improvements. The effectiveness of this method was proven experimentally on artificial faults and real faults. The superiority of the proposed method is demonstrated using a comparison with the typical object detection method and an advanced fault diagnosis method.