Abstract Shape memory alloys (SMAs) constitute a distinct class of smart materials characterized by their remarkable deformation capability and the ability to recover their original geometry upon exposure to an appropriate stimulus. Among these, nickel–titanium (NiTi) alloys exhibit superior superelasticity and shape memory effect (SME), making them promising candidates for advanced structural rehabilitation and retrofitting applications. The present study investigates the integration of NiTi SMA wires with an average ultimate tensile strength of 900 MPa in the austenitic phase within high-strength cementitious composites (HSCC) as a retrofitting strategy, focusing on enhancing the self-centering capacity, i.e., the ability to restore its original shape after unloading or deformation owing to the recovery stresses in structural elements. NiTi SMA wires of 0.8 mm diameter, embedded in bundles (consisting of 1 to 5 wires) within the tensile zone of HSCC prisms, were evaluated under monotonic three-point bending tests. The influence of the number of SMA wire bundles on key performance indicators—namely, maximum midspan deflection, crack initiation and propagation, and energy dissipation—was systematically assessed. Furthermore, the crack-closing efficiency of the embedded NiTi SMA was examined under controlled thermal activation, with image processing. Experimental observation and quantitative evaluation using ImageJ demonstrated that two-strand HSCC samples (S2) consistently achieved more than 90% crack-width recovery in the tensile zone, whereas three-strand samples (S3) showed mixed performance (in the range of 65–78%). The S4 and S5 samples relatively exhibited poor recovery with less than 25% crack-width closure. Crack closure in S2 specimens was observed to accelerate in the range of 2.4–3.2 V and in a temperature zone of 55–75°C, consistent with the reported austenitic phase ( A f ) temperature of 68°C for the NiTi SMA wire. These findings underscore the feasibility of NiTi SMA–reinforced HSCC as a ready-to-deploy, high-performance retrofitting solution for extending the service life and resilience of structural components.
The growing demand for high-performance power conversion has led to the widespread adoption of multilevel inverters (MLIs), which are capable of delivering high-quality output voltages with reduced harmonic distortion. However, their reliance on a greater number of semiconductor switches makes them susceptible to failure, thereby compromising overall system reliability. This work proposes a diagnostic and fault-tolerant control strategy for a three-phase three-level T-type (3L-T2) MLI. The diagnostic framework uses a discrete Fourier transform (DFT) to obtain the fundamental frequency components of pole voltages and phase currents, which are then normalized to detect faulty phases and perform switch-level fault diagnosis accurately. Unlike conventional AI-based techniques that impose high computational demands, the proposed approach directly leverages spectral features, providing a computationally simple signal-processing and threshold-based diagnostic framework. Furthermore, a fault-tolerant mechanism is incorporated, which utilizes an auxiliary leg and bidirectional switches to enable inverter reconfiguration during fault events, thereby ensuring uninterrupted operation. The effectiveness of the proposed methodology is evaluated through MATLAB/Simulink simulations and real-time hardware-in-the-loop validation using the OPAL-RT4510 platform, demonstrating fault detection, switch-level diagnosis, and post-fault-tolerant operation under varying operating conditions. The abovementioned work contributes to the United Nations Sustainable Development Goal 7 (SDG 7) by advancing clean and sustainable energy solutions.
Deep neural networks trained under severe class imbalance often exhibit degraded performance, typically attributed to statistical bias. In this work, we identify a complementary optimization-level pathology: inter-class gradient interference within shared representations, where gradients from majority classes suppress minority-class learning. To analyze this phenomenon, we introduce a diagnostic framework based on layer-wise gradient flow analysis and a Gradient Conflict Matrix, which quantifies interference using cosine similarity between class-specific gradients. Using this framework, we study multi-branch convolutional architectures and propose a lightweight modification, Class-Specific Branch Attention (CSBA), that enables branch-specific channel reweighting to reduce gradient coupling. This mechanism promotes implicit feature decoupling across branches while preserving architectural simplicity. Empirically, CSBA improves minority-class performance, increasing the F1 score for the Physical-Damage class from 0.261 to 0.522 under severe imbalance, while maintaining comparable overall accuracy. These results suggest that mitigating gradient interference is associated with improved minority-class representation. Validation on CIFAR-10-LT is consistent with this behavior extending to other imbalanced visual recognition settings, with Macro-F1 improving from 0.595 to 0.655. More broadly, our findings highlight the importance of considering optimization dynamics alongside statistical methods when designing architectures for imbalanced learning.
Transition to a low-carbon energy future not only requires decarbonization but also integration of principles of a circular economy (CE). Electrochemical energy transition technologies (EETTs) can be seen through the lens of CE by enabling clean energy generation and storage. This review examines the electrochemical energy transition technologies with the approach of integrating energy transition systems into the principles of CE, with focus on two areas: (1) Green fuel production and (2) electrochemical energy storage. In green fuel production, we highlight the use of electrochemically generated green fuel for sustainable feedstock, and for the decarbonization of industrial processes. In energy storage, batteries and supercapacitors as the essential components of clean energy transition are discussed. The review also examines the new opportunities, such as the new catalytic materials, integration with renewable energy sources, implementation of digitalization, artificial intelligence, and machine learning to optimize the system. Large-scale deployment is also evaluated in terms of policy and market trends. According to the literature reviewed, a framework is proposed that will help researchers, political and corporate decision makers toward synergistic pathways for CE conscious clean energy transition.
Geo-distributed computing requires geo-distributed Datacenters (DC)s to be interconnected through high-speed optical edges for Inter-Datacenter (Inter-DC) communication. Backbone optical networks are evolving towards Elastic Optical Network (EON) and, more recently, Space Division Multiplexing-Elastic Optical Network (SDM-EON) to support the exponential growth of Inter-DC traffic. However, upgrading all network edges to Multi Core Fiber (MCF) will incur huge CAPEX. Therefore, it is imperative to identify a subset of edges for upgrading to MCF to improve Elastic Optical Inter-Datacenter Network (EODCN) performance in terms of communication delay. This enables a trade-off between cost-efficiency and network performance. We propose to form a virtual spanning tree across geo-distributed DC. This ensures a single path between two DCs, thereby eliminating path uncertainties and redundant routing delays. Further, we append chords to the virtual spanning tree to ensure every node is at least two connected, thus ensuring survivability. To carry out virtual tree embedding on physical tree topology, we adopt graph theoretic approach. We explore three centrality metrics, namely edge-betweenness, edge-closeness, and edge-eigenvector, to select a set of important edges for upgrade under the influence of an increasing number of DCs in the network. Simulation results demonstrate that the proposed upgrade strategy reduces Communication to Computation Ratio (CCR) by 79%, 54%, and 41% with three, four, and five geo-distributed DC configurations compared to Single Core Fiber (SCF) network. Techno-economic analysis shows that our proposed hybrid network provides a trade-off between upgrade cost and CCR. It improves CCR by 41.66% while reducing the cost of full MCF upgrade by 45%. Thus, substantial improvements can be achieved through a cost-effective gradual network upgrade.
