This study presents a physics-informed Gaussian Process (GP)-based modelling framework for predicting time-dependent corrosion and assessing the structural reliability of steel members exposed to aggressive tropical environments. The proposed approach integrates an empirical power-law model, serving as a physically consistent trend function, with a heteroscedastic GP to capture both systematic deviations and time-dependent variability in corrosion evolution. The framework enables probabilistic prediction of corrosion depth by combining a GP for modelling residual behaviour with a second GP for representing input-dependent variance. This procedure provides a practical decomposition of predictive uncertainty and allows uncertainty propagation into structural performance analysis. The model is calibrated using a five-yr atmospheric corrosion dataset collected from seven representative sites in Vietnam. Results demonstrate that the proposed approach achieves improved predictive accuracy and uncertainty calibration compared to conventional power-law and homoscedastic GP models (R2 = 0.976, continuous ranked probability score = 0.0076). Long-term corrosion behaviour is evaluated in a probabilistic sense, where predictive uncertainty increases with extrapolation horizon, reflecting limited observational data and inherent variability in environmental exposure. Moreover, the predicted corrosion evolution is incorporated into a finite element model of tapered steel columns to evaluate time-dependent degradation of buckling resistance. In addition, Monte Carlo simulation-based reliability analysis and Sobol sensitivity analysis reveal a shift in dominant factors governing structural safety, from initial geometric imperfections to corrosion-related parameters over time. The proposed framework provides a transparent and computationally efficient tool for probabilistic corrosion modelling and risk-informed structural assessment. The results highlight its potential for supporting inspection planning and long-term maintenance strategies for steel structures in corrosive environments, while also indicating the need for future extensions incorporating environmental covariates.
Solar photovoltaic (PV) system optimisation plays an important role in maintaining an energy production efficiency at a maximum level. The objective of this study is to maximise the tilt angle of PV system according to the parameters of the site feature, weather parameters, solar path parameters and analysis method. The study was conducted using a 50 Wp monocrystalline PV module with an azimuth of 0° and different tilt angle with winter condition for geographic coordinates of Girijananda Chowdhury University, Guwahati and Photovoltaic System Simulation Software (PVsyst) was used for high precision simulation by taking seasonal variations and solar irradiation levels. The optimal orientation of the PV panels in order to maximise energy absorption was discovered with the aid of sun path analysis. The study looked at the effects of tilt angle variations based on Meteonorm 8.1 weather data, especially during off-peak winter months. According to the results, the most effective orientation under the conditions was one with a 27° inclination, which generated the maximum specific energy output, reduced system losses, and improved overall performance whereas it was observed that incorrect inclination drastically lowered system efficiency. The system's viability was evaluated through an economic analysis in addition to its technical performance. A payback period of 8.75 years, a levelised cost of electricity (LCOE) of ₹7/kWh, and an investment cost of ₹12,600 (after subsidy) were among the important financial metrics. The system is expected to save ₹25,725 in total over its 25-year operating lifespan when compared to traditional grid electricity. Under a daily average load of 0.56 kWh, the standalone PV system, set up with a 12 V battery system (300 Ah total capacity), guarantees an autonomy of roughly 5.1 days. Energy efficiency was increased through the direct coupling of the PV array to the battery, which was controlled by a universal controller. This optimisation strategy lowers system losses, boosts energy yield, and makes solar PV installations more economically viable.
In current manufacturing, obtaining outstanding product quality and process efficiency increasingly relies on data-driven modelling and optimisation. The response surface methodology (RSM) and Gaussian process regression (GPR) are two well-established methods for modelling processes. They are based on different ideas about deterministic and probabilistic learning. RSM provides analytical interpretability and statistical rigour through polynomial regression, whereas GPR offers a versatile Bayesian framework that is adept at modelling nonlinear and uncertain interactions. This research conducts a methodical comparison of RSM and GPR using actual industrial production data to analyse the impact of film thickness and weight on critical quality metrics, including tensile strength, coefficient of friction (COF), and heat-seal strength, in packaging manufacturing. A central composite design (CCD) was employed to ensure balanced experimentation, and both models were evaluated using R2, RMSE, and MAE measures. The results indicate that RSM attained a high local fit (R2 > 0.98) inside the design domain, while GPR exhibited enhanced generalisation capabilities, achieving cross-validation R2 values of up to 0.96. Additionally, the incorporation of SHapley Additive exPlanations (SHAP) improved model interpretability by quantifying the relative impact of inputs, identifying film thickness as the most significant predictor for all responses. In general, the results indicate that RSM and GPR work well together, rather than against each other. When combined, they create a robust and comprehensible hybrid framework that integrates modern data-driven learning and statistical experimental design. This framework can be utilised as a methodological guide for improving quality in the future.
Multi-link operation (MLO) is a distinctive feature of the IEEE 802.11be (Wi-Fi 7) standard that enables multi-link devices (MLDs) to coordinate the use of multiple frequency bands through a single association. This feature significantly improves the performance of IEEE 802.11be wireless local area networks (WLANs). In this paper, we propose MLO-based novel service differentiation scheme for Wi-Fi 7 to achieve service differentiation with minimal additional overhead. In the proposed scheme, initially the nodes in the network are grouped based on their transmission requirements, and then links are allocated to the MLDs based on their priority. The analytical and simulations results confirm that the proposed scheme can deliver efficient service differentiation in terms of throughput and delay performance in Wi-Fi 7, without compromising the aggregate throughput of the network.
The problem of choosing the best port in a slow-fluid antenna multiple access (s-FAMA) system, where the port remains fixed until channel conditions change, is the first task addressed by the proposed approach. To address this, a limited number of observed channels are used to estimate the signal-to-interference plus noise ratios (SINRs) across all accessible ports using the advanced pyramidal dilation and attention-driven convolutional network (APDASCN). Despite having few data, this network improves prediction accuracy by prioritizing important channel information and using pyramidal dilation to collect multi-scale spatial patterns. The starfish optimization algorithm (SOA), which ensures reliable performance with little computing cost, is used to optimize system settings and choose the port after the SINRs have been deduced. Lastly, simulations are used to assess the overall scheme, validating its potential for next-generation multiple access systems by demonstrating that it successfully lowers the outage probability and enhances multiplexing advantages.
