Molecular representation learning and molecular property prediction are crucial tasks in AI-driven drug discovery. Learning the representation of molecules helps to understand various properties of the molecules, such as their toxicity. As molecules can be represented as graphs, we can learn the molecular graph and predict the molecular properties. Many graph neural network models have been proposed for this purpose. However, most of the works have focused only on the aggregation of the atom features; they do not incorporate bond features. To address this issue, we develop an Edge Convolution Network (ECN) model to predict newly discovered molecules’ characteristics. ECN is capable of effectively handling the features of atoms and bonds together, and by combining them interactively, it makes a representation of the entire molecule, i.e. molecular representation. We employ self-supervised approaches to help the model learn molecular representations and perform better in several downstream tasks. ECN has been evaluated on seven molecular property prediction tasks, including four classification and three regression tasks. ECN exhibits up to 8–10
Catalyst material design is one of the strategies to improve the performance of oxygen reduction, especially using non-precious metal. Certainly, each component material has synergistic role to improve reduction reaction. This research analyses composition study of dual metal including Fe, Co, and Ni with nitrogen doped carbon building block. Optimisation of nitrogen doping on carbon surface was carried out during aniline polymerisation for 24 h via chemical oxidative polymerisation, followed by pyrolysis after the addition of metal precursor. The formation of dual metal and oxide compounds is confirmed by X-ray diffraction analysis. The combination of dual metal composition produces varying electrochemical properties as seen on cyclic voltammetry (CV) analysis result. Current density is one of the main properties of material selection for catalyst. NiCo–C–N exhibits higher current density than CoFe–C–N and FeNi–C–N. The formation of C–O bond from Fourier-transform infrared spectroscopy analysts on NiCo–C–N catalyst plays an active role in adsorbing oxygen which generates superior catalyst performance. Two catalyst component materials also affect the morphological structure and composition from energy dispersive X-ray spectroscopy measurement.
Foam concrete (FC) is increasingly recognized as a lightweight construction material due to its low density and advantageous properties. However, the coupled effect of the type of surfactant and polymer additives on the performance of medium density FC is not well established. In this study, FC with a target fresh density of 1800 kg/m3 was produced using two surfactant systems, i.e., alpha-olefin sulphonate (AOS) and cocamidopropyl betaine (CAPB), in combination with polyethylene glycol (PEG) and carboxymethyl cellulose (CMC). The effects of these combinations were assessed in terms of fresh density, spreadability, flow time, compressive strength and water absorption. The results showed that the AOS–PEG system showed comparatively better performance which resulted in 18
Thermal Barrier Coatings (TBCs) are essential for ameliorating the performance and durability of high-temperature segments in applications essentially gas turbines, jet engines, and diverse advanced thermal systems. Despite the widespread use of conventional deposition methods like Plasma Spraying (PS), Electron Beam Physical Vapor Deposition (EB-PVD), and Chemical Vapor Deposition (CVD), challenges remain in achieving uniformity, adhesion strength, and long-term thermal stability. In recent years, machine learning (ML) and artificial intelligence (AI) integration have emerged as promising approaches to optimize TBC deposition processes, predict material behavior, and automate design and quality control. This paper reviews the application of ML and AI techniques in enhancing TBC deposition, focusing on their use in process optimization, predictive modelling, and defect detection. ML methodologies, such as neural networks, support vector machines, and genetic algorithms, have been leveraged to fine-tune key deposition parameters, such as spray distance, substrate temperature, and gas pressure, improving coating quality and performance. Furthermore, AI-driven techniques, including image recognition and deep learning, have been employed for microstructural analysis, defect detection, and lifetime prediction of TBCs under varying thermal and mechanical conditions. This review outlines the current advancements in TBC technology, emphasizing the role of ML in refining deposition processes, and discusses future directions for developing more efficient, cost-effective, and durable thermal barrier coatings.
The increasing demand for clean air has further accentuated the need for the development of efficient air filtration technologies. This article describes the design, fabrication, and testing of a novel hollow-structured nonwoven air filter designed to improve particle capture efficiency with optimal airflow and low pressure drop. The filter design combines a tubular, porous structure with a nonwoven substrate to increase surface area for effective aerosol capture. The critical fabrication parameters, such as polyacrylonitrile (PAN) polymer concentration, polyurethane foam (PUF) weight, and tube circumference, were optimised using a Box–Behnken design to investigate their effects on air filtration performance. Fifteen hollow tube air filter samples were fabricated and tested for particulate matter removal efficiency (0.3 – 5 µm). The results showed that higher concentrations of polymers and higher weights of polyurethane foam increased the filtration efficiency, and smaller circumferences of the tubes resulted in better performance because of higher packing density. The maximum filtration efficiency of 94.12
