Laser Powder Bed Fusion (LPBF) is widely used to produce lightweight aluminium alloys such as AlSi10Mg for aerospace and defence applications. Yet, the as-built parts often suffer from excessive surface roughness, which reduces fatigue resistance and impedes fluid flow problems that are especially critical in narrow slots and channels. Conventional finishing techniques find it difficult to access and polish such confined features, making alternative strategies necessary. In this work, we investigated laser-based surface remelting as a geometry-aware polishing method for LPBF AlSi10Mg slots with widths between 1 and 5 mm. The experiments, carried out on 30 samples, achieved up to ~ 87% reduction in average surface roughness (Ra), with profilometry confirming the removal of powder-induced asperities. Although this degree of improvement is in line with earlier reports, the novelty of the present study lies in introducing the first predictive framework that combines slot geometry particularly slot width ith physics-inspired descriptors in an artificial neural network (ANN). To overcome the limitations of the experimental dataset, around 2000 synthetic yet physically consistent data points were generated to guide the model. The resulting ANN achieved excellent predictive accuracy (R² ≈ 0.98–0.99) and closely matched experimental polishing trends. While demonstrated here for slot geometries, this integrated experimental–computational approach offers a pathway toward AI-assisted, geometry-sensitive surface optimisation for aerospace and defence components where conventional polishing is not feasible.

Purpose: Pressure vessels are vital components in manufacturing and energy systems, where early detection of degradation is essential to prevent catastrophic failures. Conventional acoustic emission (AE)-based prognostic methods depend heavily on experimental testing and finite element simulations, which are costly and time-intensive. This study proposes a fully computational, artificial intelligence (AI)-driven digital twin framework capable of performing both health-state classification and remaining useful life (RUL) prediction using physics-inspired synthetic AE data. Design/methodology/approach: A multi-task convolutional neural network–long short-term memory (CNN–LSTM) architecture was developed to extract spatio-temporal features from synthetically generated AE waveforms. The AE signals were produced through a parameterized, physics-informed model representing energy attenuation and frequency variation during progressive material degradation. CNN layers capture spectral energy features, while LSTM layers learn temporal evolution patterns associated with structural deterioration. The framework, implemented entirely in Python, enables an end-to-end digital twin without external simulation platforms. Model convergence, residual analysis and t-distributed stochastic neighbor embedding visualizations were used to assess robustness and interpretability. Findings: The model achieved 99% classification accuracy and a low RUL prediction error (root mean square error ˜ 5), demonstrating strong predictive capability and stable convergence. Latent-space visualization revealed distinct health-state clusters, validating interpretable degradation learning across synthetic AE data. Originality/value: This work presents one of the first physics-informed, Python-based digital twin frameworks for AE-driven prognostics of pressure vessels. It eliminates dependence on costly experiments or finite element simulations, offering a scalable, data-efficient approach suitable for Industry 4.0 applications.

The rapid proliferation of unmanned aerial vehicles (UAVs) in defence and surveillance zones presents an escalating challenge to national security, demanding the deployment of agile, real-time interception systems. While high-energy laser-based directed energy weapons (DEWs) provide a non-kinetic alternative to missile defence, their performance is hampered by actuator latency, beam jitter, and inability to track erratically manoeuvring drones. To overcome these challenges, this study proposes a novel AI-augmented predictive beam control framework that anticipates UAV trajectories by estimating future positions using velocity-informed regression modelling. The system, developed in MATLAB, simulates and predicts evasive UAV motions sinusoidal, zigzag, and spiral through a shallow neural network trained on synthetic trajectory data. It incorporates beam delay modelling and sinusoidal jitter to replicate real-world laser transmission imperfections. The predictive model achieves a tracking accuracy of 94.12% and a mean squared error (MSE) of 0.1834, validated across diverse UAV paths. Key simulation outcomes include six comparative trajectory plots, error histograms, and cumulative accuracy profiles that highlight the system’s robustness. This research contributes (i) an integrated beam latency compensation module for predictive DEWs, (ii) a simulation dataset curated from recent UAV motion and targeting literature, and (iii) a control framework extendable to reinforcement learning (RL) and Kalman filtering approaches for adaptive targeting. The proposed framework advances the design of autonomous DEW systems by enhancing hit accuracy, energy efficiency, and responsiveness in contested aerial environments. The framework’s modular design enables extensions to real-world testing and broader applications in high-speed target tracking beyond defence systems.

Digital twins increasingly rely on autonomous self-updating mechanisms to remain synchronized with physical systems; however, repeated self-updating can lead to error accumulation, numerical instability, and progressive loss of physical consistency when models iteratively learn from their own predictions. To address this challenge, the study proposes a physics-stabilized self-updating digital twin framework based on Physics-Informed Neural Networks (PINNs) and demonstrates its core principles on a canonical thermal field prediction problem. The framework integrates adaptive physics-loss weighting, a physics-only stabilization stage, and second-derivative smoothness regularization within the self-updating loop, enabling controlled data assimilation while explicitly enforcing governing equation constraints. Numerical results show a monotonic reduction in root mean square error (RMSE) from approximately 1 × 10 ⁻³ in the first update cycle to 8 × 10 ⁻⁵ after four update cycles, accompanied by effective suppression of model drift and a substantial reduction in partial differential equation (PDE) residuals compared to a naïve self-updating strategy. Furthermore, the analysis reveals the existence of an update saturation point, beyond which additional autonomous updates yield diminishing accuracy improvements, providing a physically motivated stopping criterion for autonomous updating. By establishing a stable and physically interpretable self-updating architecture, the study provides a foundational framework for the development of reliable digital twins, with clear potential for extension to more complex thermal and Multiphysics systems.