ABSTRACT Ensuring high-quality production of glass products is essential in modern manufacturing, where even minor defects such as scratches, cracks, or surface irregularities can significantly affect safety, performance, and usability. Traditional visual inspection methods are labour-intensive, time-consuming, and prone to human error, creating a strong need for automated and reliable defect detection systems. The main objective of this research is to develop an efficient Deep Learning (DL) based framework for accurate binary classification of defective and non-defective glass products while addressing limitations such as subtle defect visibility, variation in lighting conditions, and dataset imbalance. Furthermore, the study aims to evaluate model performance under different hyperparameter settings using an intelligent optimisation technique. The proposed work employs the Residual Network 50 Version 2 (ResNet50V2) Convolutional Neural Network (CNN) for defect classification, supported by a comprehensive preprocessing pipeline consisting of resizing, normalisation, greyscale conversion, and data augmentation. Bayesian Optimisation (BO) is integrated to fine-tune key hyperparameters, ensuring optimal learning efficiency and improved generalisation. The framework is trained using a curated glass defect dataset and evaluated using standard performance metrics to demonstrate its robustness and suitability for real-world industrial inspection scenarios. Experimental results show that the optimised ResNet50V2-BO model achieves 98.36% accuracy, along with a precision of 98%, recall of 93%, and F1-score of 95% for the defective class. The confusion matrix and PR curves further confirm the model’s ability to reliably distinguish subtle surface defects. These findings highlight the potential of the proposed approach as a fast, reliable, and scalable solution for automated quality control in glass manufacturing industries.

ABSTRACT To analyse the structural characteristics and evaluate fatigue life of spiral bevel gear transmission with crack faults, a comprehensive and in-depth analysis is conducted based on the optimisation analysis flow from the aspects of static contact, transient dynamics, modal and fatigue life. The related investigations are carried out, and some findings are revealed as well. First, a parametric spiral bevel gear FEA model with different crack parameters including crack depth and penetration length is established. Second, the bending stress and natural frequency under the parameters of gear crack depth and penetration length are analysed based on three different aspects including static analysis, transient dynamics and modal analysis. The influences caused by different crack parameters are obtained, and the variation trend is analysed as well. Finally, a fatigue life prediction model for cracked gear based on fracture mechanics and fatigue cumulative damage theory is established, and the fatigue life curve under different crack parameters is obtained and analysed. Key factors such as crack growth rate, stress concentration factor and material fatigue properties are considered in this fatigue model. This research not only provides a scientific basis for fault diagnosis and maintenance of helicopter intermediate gearboxes, but also serves as a valuable reference for the structural characteristic analysis and fatigue life assessment of rotating machinery.

ABSTRACT Projectile impact damages could chang the mass distribution and stiffness characteristics of the tail drive shaft. However, the effect mechanisms of its effect on mass distribution and stiffness characteristics are unclear. Hence, this paper conducts research on the projectile impact damage of the helicopter tail drive shaft and its effect on mass distribution and stiffness characteristics. The finite element simulation model of the projectile impact damage is established, and the stiffness simulation model is further proposed. The residual velocity, projectile impact duration, and projectile impact damage morphology have been analysed in detail. The effect of projectile impact damage on mass loss, centre of mass displacement, stiffness reduction, stiffness asymmetry, and cross stiffness is evaluated. The projectile impact experiment bench is established. The maximum error between the experimental incidence velocity and the ideal incidence velocity is 3.4%. The projectile impact damage morphology obtained from the experiment is larger than the simulated damage, with a maximum error of 26.5%, because the armour and lead sheath expand the damage area. Experimental results effectively validated the accuracy of the finite element simulation model of the projectile impact damage. This paper provides important theoretical guidance and technical support for the helicopters’ survival ability.

ABSTRACT This paper studies the free vibration analysis of a nanocomposite conical shell with an adhesive lap joint. As the adherents, two polymer-based nanocomposite conical shells enriched with graphene nanoplatelets (GNPs) are considered, which are connected with an elastic adhesive. The distribution patterns and mass fractions of the GNPs in two adherents are not necessarily the same. The set of coupled partial differential equations is solved analytically by considering appropriate trigonometric functions in the circumferential direction, and approximately by applying the differential quadrature method (DQM) in the meridional direction. Numerical results demonstrate that the natural frequencies increase as the lap joint length increases. However, the variation of each natural frequency versus the variation in the thickness of the adhesive depends on the vibrational mode. It is demonstrated that the natural frequencies reach their maximum values when the GNPs are distributed as far away as possible from the middle surfaces of the outer and inner parts of the shell. The presented study is the first theoretical work regarding to the free vibrational analysis of a nanocomposite conical shell with an adhesive lap joint.

ABSTRACT This study investigates the effectiveness of microwave-assisted debinding as a sustainable and efficient post-processing technique for 316 L stainless steel components fabricated via material extrusion additive manufacturing (MEAM). Traditional debinding methods, such as solvent and thermal debinding, are often energy-intensive, time-consuming, and prone to causing part defects due to incomplete binder removal and dimensional distortion. Microwave-assisted debinding offers uniform volumetric heating, which can enhance the binder removal process and reduce processing time. Eight different microwave debinding parameter sets were tested, varying in power level, heating duration, and heating mode, applied to green parts produced using BASF Ultrafuse 316 L filament. The performance of each condition was evaluated based on binder removal rate, dimensional stability, relative density, microhardness, phase composition, and microstructural integrity. Among the tested conditions, specimen S_026 (parameter 8) showed optimal performance, achieving a binder removal rate of 5.14%, a dimensional expansion of below 2%, and a microhardness of 279.08 HV. X-ray diffraction (XRD) and scanning electron microscopy (SEM) analyses confirmed the presence of strong phase formation with minimal porosity in S_026. In contrast, suboptimal parameters led to increased porosity, microcracks, and compromised dimensional stability. The study also revealed that debinding conditions significantly influence the mechanical properties of the sintered parts. Overall, the findings demonstrate that optimised microwave-assisted debinding is a viable alternative to conventional methods, offering improved efficiency and quality in metal AM processing. This method holds significant promise for industrial-scale MEAM applications by enabling the production of dense, mechanically robust components with reduced processing time and energy consumption.

ABSTRACT Erosion wear caused by the impact of solid particles in centrifugal slurry pumps presents a significant engineering challenge, affecting pump performance. This study investigates the effects of particle shape, size, and mass flow rate on erosion wear and pump efficiency using ANSYS CFX simulations with the Finnie erosion model. The results demonstrate that increasing the shape factor from 0.2 to 0.8, while keeping the particle mass flow rate at 0.5 kg/s and particle size at 500 µm, reduces erosion wear rate density from 5.23 × 10-5 to 3.26 × 10-5 kg/m2, improving pump performance from 56.87% to 67.35%. Conversely, when the particle size increases from 500 µm to 1500 µm, with a fixed mass flow rate of 0.5 kg/s and shape factor of 0.2, the erosion wear rate density rises from 5.23 × 10-5 to 9.75 × 10-5 kg/m², resulting in a performance drop from 65.85% to 54.45%. Furthermore, increasing the particle mass flow rate from 0.5 kg/s to 1.5 kg/s, with a particle size of 500 µm and shape factor of 0.2, elevates the erosion wear rate density from 5.23 × 10-5 to 7.87 × 10-5 kg/m2, causing pump efficiency to decline from 56.87% to 50.28%. The study concludes that more spherical particles and smaller particle sizes lead to lower erosion rates and better pump performance.

ABSTRACT This paper details a hybrid operational and computational study for improving the thermodynamic performance of a Gas-Steam Combined Cycle Power Plant (CCGT) through a novel hybrid optimisation framework Adaptive Annealed NSGA (AANSGA. This framework integrates Non-Dominated Sorting Genetic Algorithm II (NSGA-II) and Simulated Annealing (SA). The study applies second-law (exergy) analysis and optimised heat integration approaches such as the pinch point method and the approach temperature difference method to reduce exergy destruction and improve thermal efficiency. Operational results were obtained from a laboratory-based CCGT study, where the gas turbine inlet temperature was 1400 K, the Heat Recovery Steam Generator (HRSG) outlet steam pressure was 20 bar, and the steam mass flow rate was 25 kg/s. The results show an increase in thermal efficiency from 46.96% to 54.12%, and an increase in the exergy efficiency from 84.95% to 85.55%. Similarly, fuel consumption improved from 2.425 to 2.405 kg/s, and CO2 emissions stabilised at 352 kg/MWh. The HRSG pinch point temperature difference was improved between 9.35°C − 9.47°C. The hybrid AANSGA approach achieved a reduction in total exergy destruction per cycle and improved operational stability. Overall, AANSGA provides a useful decision-support tool for the development of sustainable, high-performance power systems under dynamic conditions.

ABSTRACT The nuclear industry is exploring applications of machine learning, including autonomous control and management of reactors and nuclear power plant and their components. The accurate diagnosis and classification of motor bearing faults under diverse operating conditions remain a significant challenge due to the complex nature of signal patterns, overlapping features, and dynamic environments. This study presents a comprehensive comparative analysis of multiple machine learning algorithms, including Random Forest, Gradient Boosting Machine, Decision Tree, Support Vector Machine (SVM), and K-Nearest Neighbours (KNN), applied to the HUST bearing dataset. This work also underscores the importance of load-dependent analysis, as fault signatures in bearings vary significantly with operating conditions. Experimental results indicate that ensemble models, particularly Random Forest, deliver superior performance across both binary and multi-class classification tasks, achieving up to 99.70% accuracy for 7-class cases and 99.37% for 21-class scenarios. Performance metrics further highlight the Random Forest model’s robustness, achieving 99.37% precision, recall, and F1-score with 15 features, confirming its suitability for real-time predictive maintenance applications. This study emphasises the importance of appropriate segmentation strategies and model selection, offering a reliable and scalable framework for industrial fault diagnostics and condition monitoring systems.

ABSTRACT Faults such as rotor unbalance and shaft cracks pose serious risks to the integrity of high-speed rotating machinery, often leading to catastrophic failures if not identified early. While conventional bearings have long been the standard, air foil bearings are rapidly gaining traction due to their superior reliability, durability, and minimal maintenance requirements. In this paper, a machine learning-based diagnostic framework has been presented for fault classification in a Jeffcott rotor system supported by air foil bearings. The rotor model features a centrally mounted rigid disc and is analysed under three fault scenarios: unbalance, crack, and a combination of both. Using Newton’s second law and incorporating the equivalent stiffness and damping characteristics of the shaft and foil bearings, the system’s equations of motion are derived. Displacement responses at the disc location are simulated via a MATLAB Simulink model, capturing dynamic behaviour under varying fault conditions. These time-domain signals are then processed using the HistGradientBoostingClassifier, a robust and efficient machine learning algorithm to accurately classify fault types. Comparative analysis with Support Vector Machine, Deep Neural Network, and Random Forest models demonstrates superior performance of the proposed approach in both accuracy and computational efficiency. It is revealed that Random Forest and HistGradientBoosting consistently achieved the most generalisable and robust performance. The proposed machine learning approach not only enhances predictive maintenance strategies but also contributes to the reliability and safety of advanced rotating machinery systems.

ABSTRACT Casting is a fundamental manufacturing process for producing components with complex geometries; however, surface and subsurface defects continue to compromise product reliability andproduction efficiency. To support automated and consistent quality inspection, this paper presents a hybrid deep learning framework termed SCNNBN – TBiG for intelligent casting defect classification. The proposed approach integrates stacked convolutional neural networks with batch normalisation to extract stable and discriminative spatial features, followed by a Transformer encoder that captures long-range contextual relationships through multi-head self-attention. The resulting representations are compressed using global average pooling and subsequently analysed by stacked bidirectional gated recurrent unit layers to model sequential dependencies within the learned feature space. The framework is evaluated on a publicly available industrial casting image dataset comprising 7,348 samples under both defective and non-defective categories. Experimental results demonstrate that the proposed model achieves a testing accuracy of 99.44%, outperforming several existing deep learning and hybrid architectures. The findings confirm that the synergistic integration of spatial, global, and sequential feature learning provides a robust and efficient solution for high-precision industrial quality inspection.