Abstract Accurate fault detection and diagnosis are critical components of any fault-tolerant control system, especially for Unmanned Aerial Vehicles (UAVs) where reliability is paramount. Traditionally, both model-based and data-driven approaches have been applied for fault diagnosis. However, the increasing complexity of high-dimensional UAV systems has shifted focus toward data-driven methods, which leverage advanced classification algorithms to enhance fault identification and isolation. This study builds on this evolution by developing a sophisticated condition-based monitoring (CBM) system specifically designed for multirotor UAVs. In contrast to earlier studies that primarily relied on raw data for classifier training, this work introduces advanced preprocessing techniques and multi-domain feature extraction, significantly improving the robustness and accuracy of fault detection. A comparative analysis is performed between feature selection methods, including Recursive Feature Elimination with Cross-Validation (RFECV) and Variational Autoencoder (VAE), to extract critical insights into UAV operational behavior. Through testing and evaluating various classification models on data from a hexarotor UAV under diverse actuator fault conditions, this research identifies optimal approaches for real-time fault detection and diagnosis. Results demonstrate notable improvements across all evaluation metrics, establishing this approach as a substantial advancement in UAV fault tolerance.

Abstract Accurate diagnosis of outer race defects and their interaction with secondary faults remains a critical challenge in bearing condition monitoring. This study presents a physics-based diagnostic approach that integrates extended hamilton's principle (EHP) with an improved one-against-all multiclass support vector machine (OAA-MCSVM) for identifying and classifying complex bearing faults. A dynamic model of the rotor-bearing system is developed using EHP to capture the influence of outer race defects under combined fault scenarios, including misalignment, unbalance, and radial clearance variation. Vibration responses are acquired from a controlled test rig under isolated and coupled fault conditions. Fault signatures are extracted through time-frequency analysis and mapped to system dynamics derived from the variational formulation. The extracted features are classified using the improved OAA-MCSVM framework, which enhances boundary discrimination between closely interacting faults. Experimental validation shows that the proposed method achieves high classification accuracy across all tested fault conditions, with improved sensitivity to outer race-related compound faults. The integration of physics-based modeling with machine learning enables a more interpretable and reliable fault diagnosis scheme, suitable for real-time application in rotating machinery.

Abstract Due to the complexity of the operating environment and the diversity of components, fault maintenance and detection of aircraft engines have always been a huge challenge. To improve the reliability and maintenance efficiency of aircraft engines, a multi-channel, multi-scale, and graph relational meta learning intelligent detection method is proposed. By combining multi-channel data fusion, multi-scale feature extraction, graph neural networks, and meta learning strategies, a comprehensive understanding of fault characteristics and their propagation relationships is achieved. In the research results, the proposed multi-channel multi-scale and graph relation meta learning intelligent detection method has better performance, with a localization accuracy of 0.94 and an average localization error of 0.11. Meanwhile, its training time is 15.2 minutes and inference time is 49.65 milliseconds, demonstrating high computational efficiency. In addition, the multi-channel multi-scale and graph relation meta learning intelligent detection method has shown good robustness under different testing conditions. The results show that this method not only has significant advantages in the accuracy and efficiency of fault detection, but also has excellent fault localization performance, significantly reducing the safety risks and economic losses caused by aircraft engine failures. The research provides a theoretical basis for real-time fault monitoring and maintenance decision-making of aircraft engines, which is beneficial for improving the efficiency and quality of maintenance work.

Abstract Threaded pipe structures are critical components in equipment used in various fields. The threaded sections are prone to induce defects due to stress concentration, threatening the safe operation of the equipment. Consequently, the inspection of these structures is essential. While the ultrasonic guided wave (UGW) method has been applied to inspect threaded pipes, the influence of thread parameters on UGW propagation characteristics is yet to be examined. This study analyzed the dispersion characteristics of UGWs. It reveals that the group velocity dispersion curve of the L(0,2) mode initially increases and then decreases over frequency range of 60–140 kHz. The group velocity is higher in trapezoidal threads than in rectangular threads. Dispersion curves for different thread heights exhibit a crossover near 85 kHz. Dispersion curves for different pitches intersect around 83 kHz. The effect of thread parameter variations on the reflection characteristics was investigated. It was found that the trapezoidal thread exhibits a higher reflection coefficient than the rectangular thread. The reflection coefficient increases with thread height and decreases with pitch. The influence of thread parameters on defect detection sensitivity was examined. Results demonstrate that the L(0,2) mode offers high sensitivity for defect detection in threaded pipes featuring trapezoidal threads, a thread height of 1 mm, and a pitch of 4.5 mm. Finally, the effectiveness of the L(0,2) mode in detecting defects of varying depths within threaded pipes was validated. This research provides a novel method for the inspection of threaded pipe structures.

Abstract High-density polyethylene (HDPE) is used in several critical applications, ranging from cooling water pipelines in nuclear power plants and distribution pipelines for natural gas to biomedical implants. Embedded crack-like flaws form within HDPE during fabrication or operations, which may grow over time and can cause catastrophic failure if undetected. Large structures such as pipelines, where the location of a flaw is not known, require a fast, non-destructive evaluation (NDE) where the sensor can move rapidly across the structure with a fast microseconds data collection window at each location. This is only possible if the flaw is evaluated using a microsecond time signal. Ultrasonic A-scan (time signal) allows for the rapid scan, while B-scans are limited as they are slow and depend on post-processing algorithms, where subtle information can be lost. We propose a method for training a convolutional neural network (CNN) using computer simulations of ultrasound on HDPE and applying the trained CNN to real-life experiments to decipher crack characteristics in HDPE or other polymer structures using ultrasound time signals, utilizing very small measurement windows. We show that fully finite element simulations trained CNN can accurately predict crack lengths (MAPE 3.3%) and positions (MAPE 3.8%) in HDPE from experimentally measured ultrasound A-scan microsecond signals. The success of a 100% simulation trained CNN without exposure to any prior experimental data in accurately predicting crack sizes from experimental time signal data underscores a promising path for next-generation NDE methodologies.

Abstract In this article, a novel method for modal response estimation and response reconstruction is proposed. The method is based on a newly developed modal response estimation method and transformation equations derived from the modal information of the target. The modal response estimation method is designed to acquire optimal modal responses using responses from only one sensor. Based on the derived optimal modal responses of convenient locations, the modal responses of critical locations can be extrapolated using the transformation equations. To demonstrate the overall reconstruction procedure, two numerical examples are presented, including cases of well-separated modes and closely spaced modes. The effects of mode numbers, sensor numbers, sensor locations, and noise levels are investigated in detail. Following this, a realistic turbine blade structure is used to validate the effectiveness and accuracy of the method in practical applications. The results indicate that the proposed method is accurate and reliable for response reconstruction, offering a viable alternative for structural health monitoring of various structures.

Abstract Electric vehicles (EVs) are a favorable tactic for reducing carbon emissions. However, the most used power source in EVs, lithium-ion batteries (LIBs), can pose a significant safety risk in the form of thermal runaway. This is a rapid failure mode that may lead to fires and explosions. To address this issue, the authors' previous work developed a temperature-sensing composite battery enclosure with embedded temperature microsensors to provide the LIB condition monitoring. The prior work produced extensive experimental and simulation results, characterizing an enclosure-embedded battery management system. It was found that the top composite layer causes a time lag in the temperature detection, impeding an early warning signal. This current study aims to create a regression model leveraging machine learning (ML) strategies to predict battery enclosure interior surface temperatures when trained on the prior study's data. The temperature inference model predicts the enclosure's surface temperatures using embedded temperature measurements in real-time, compensating for the time lag. Random forest and recurrent neural network ML models are compared, considering performance and computational costs. Mean absolute error and mean absolute percentage error are utilized to quantify the prediction accuracy. The temperature inference model enhances the practicality of a temperature-sensing composite battery enclosure as a battery management system, mitigating risks associated with LIB thermal runaway events. By monitoring embedded temperature changes and predicting the temperatures on the interior surface of the enclosure, the system provides insights into potential hazards, enabling timely interventions and ensuring EV safety.

Abstract In this study, a machine learning-based detection and localization of localized damage in lead zirconate titanate (PZT) ceramics is developed. A point-contact excitation and detection method is employed to excite and detect acoustic wave signals from the PZT sensor. The signals are analyzed using non-destructive evaluation techniques. The significant features of wavelet transform coefficients, auto-regressive modeling parameters, peak amplitude, peak location, and wave energy are extracted from the waveforms. These features capture the salient properties of the acoustic response that change in the presence of structural damage. A trained support vector machine classifier is used to distinguish between damaged and healthy regions based on the extracted features. Classification achieved a recall of 92.7% and a precision of 86.0% for the minority damaged class. However, the method is compromised at the center of the samples, where the wave energy is the highest and the signal originates. Furthermore, the thresholding method used in data labeling can be sensitive to local anomalies, potentially leading to misclassification. Despite these challenges, the proposed framework supports a scalable and robust real-time damage detection system. By integrating machine learning, point-contact acoustic sensing, and signal processing, this study contributes to the development of automated and accurate structural health monitoring techniques for smart sensing systems.

Abstract This study demonstrates a novel non-destructive evaluation method that combines geometric phase sensing with cross-correlation processed acoustical responses, enabling the detection of structural anomalies even when the acoustic excitation is stochastic and the source characteristics are variable. Traditional ultrasonic techniques, which depend on impulse responses from known source locations to capture wave transmission behavior, often fail under stochastic excitations due to incoherent phase alignment and unpredictable wave paths. The proposed method applies cross-correlation between a fixed reference site and other sensor locations to refine the acoustic field representation, enabling physically meaningful geometric phase extraction through the dot product of two state vectors representative of the acoustic field in a high-dimensional complex Hilbert space. This allows detection of both excitation-induced field asymmetries and subtle non-linearities. Experimental validation using laser Doppler vibrometry on a circular IN625 plate demonstrates that this approach preserves excitation-induced field asymmetries while remaining sensitive to structural perturbations such as mass defects. The cross-correlation geometric phase change (CC-Δφ) spectra reveal modal differences across excitation conditions, even under white noise, where geometric phase without cross-correlation (Δφ) remains centered near 90°, obscuring structural insights. The method also detects mass-induced effects, showing increased average CC-Δφ compared to the no-mass case under the same excitation condition. These results establish a foundation for a robust, non-contact, source-independent NDE technique suitable for operation under variable and uncontrolled excitation scenarios.

Abstract Corrosion in engine connecting rods is a critical issue in the automotive industry, potentially leading to catastrophic engine failure, monetary losses, and safety hazards. The labor shortage in the industry further emphasizes the need for fast, accurate, and automated corrosion detection methods to ensure appropriate surface treatments can be applied to restore component integrity. We present an unsupervised image-based framework for classifying corrosion severity in automobile engine connecting rods using short-wave infrared (SWIR) and telecentric grayscale imaging. We employ the structural similarity index measure (SSIM) as a dissimilarity metric and the k-medians clustering algorithm for classification. Our algorithm achieves an overall accuracy of 80.64% for SWIR images, with 100% accuracy in classifying highly corroded samples. For grayscale images, the method attains an overall accuracy of 77.42%, with 90.91% accuracy for highly corroded samples. The method’s ability to work with different imaging modalities and its high accuracy in identifying severe corrosion cases make it a promising tool for automated corrosion assessment in the automotive industry, potentially improving efficiency and safety in engine component maintenance.