Vol 3 · Issue 11
Open Access
Ayush D. Karande, Sahil U. Ghare, Shreyash C. Mali, Ajinkya B. Patil
This project focuses on the development of a smart water purification and cooling system that provides clean drinking water at a controlled temperature. The proposed system integrates multiple water filtration techniques with sensors to monitor water temperature and tank level. A microcontroller, such as an ESP32 or Arduino, processes the sensor readings and controls the pump, solenoid valve, Peltier cooling module, and cooling fan according to the system requirements. The Peltier module cools the purified water through thermoelectric technology, eliminating the need for conventional refrigerants. The cooling process is automatically regulated based on the water temperature, helping to reduce unnecessary power consumption. This integrated design provides a compact, automated, and convenient solution for supplying purified and cooled drinking water in residential buildings, offices, educational institutions, and laboratories.
Smart water purificationPeltier coolingthermoelectric system.
Vol 3 · Issue 11
Open Access
Kandi Sandya, Gudivada Aravind, Kalla Satyasai, Datti Manikanta
Commonly railway defect is detected by periodic inspection of tracks and structures such as fallen trees, rocks, debris, water logging, vegetation growth and track damage. Conventional systems for monitoring large sections of railways or those that are hard to access are cumbersome, tedious, and labour intensive. This paper proposes an automated railway infrastructure monitoring system integrating anomaly detection, risk assessment, and maintenance advice, powered by AI. Using the drones, the system takes photographs and videos of railway tracks and infrastructure, and then uses a deep learning algorithm developed by YOLO to identify and classify anomalies. The anomalies identified are evaluated according to type, severity, geographical location and track coverage. Environmental factors, like the weather and past history, are also added to the machine learning risk assessment module which produces risk scores. The system takes these scores and estimates maintenance requirements and suggests maintenance activities (priority, people required, equipment, and estimated completion time). A GIS interface presents anomalies and related risks to maintenance teams, enabling them to identify areas that need attention and plan their work. The framework combines automated inspection, anomaly detection, risk assessment and maintenance planning, facilitating to safer and more efficient railway infrastructure management.
Drone-Based Railway InspectionRailway Anomaly DetectionYOLO
Vol 3 · Issue 11
Open Access
Chichula Abhishek,Kotilingala Jeevitha,Attada Anvek, Gudla Sai Lahari…
There is a great difficulty in communicating between Indian sign language users and nonusers. The paper introduces a multilingual sign language translation framework, which is based on computer vision, deep learning, and Generative AI to facilitate communication. The sign recognition module is responsible for capturing data from the video using OpenCV, detecting the hands in the image using MediaPipe Hands, extracting the spatial features using EfficientNet-B0 and learning the temporal features using Bidirectional Gated Recurrent Unit network. For the recognition model, the training and evaluation set consisted of INCLUDE which contains a total of 262 ISL sign classes. 32 frames were extracted for each input video sequence and 1,280 dimensional feature vectors were used. The performance obtained from the experimental evaluation for 619 samples was found to be 86.75% with 537 correct predictions. The framework also includes a user-controlled gloss buffer where in known sign labels are gathered and sent for sentence generation when the user has manually triggered translation. After the user manually triggers translation, Generative AI is used to convert the buffered sign labels into grammatically meaningful sentences. The generated sentences are then translated into English, Telugu, Hindi, and Tamil using the multilingual translation module and converted into speech using the Text-to-Speech module.
The following components for additional sentence generation, multilingual translation and speech are part of the framework, but have not been quantitatively assessed in this study. The results confirm the viability of the proposed architecture for the recognition of isolated signs using the ISL, and serve as a basis for developing the system towards multilingual sign language communication.
Indian Sign LanguageIsolated Sign RecognitionComputer Vision
Vol 3 · Issue 11
Open Access
Suhani Shinde, Kalpesh Kharat and Sunil Shinde
Industrial activities associated with construction and infrastructure development contribute to ambient air pollution through material handling, fuel combustion and high-temperature processing. This review compares the principal sources and control approaches for air pollutants associated with four sectors: cement manufacturing, ready-mix concrete (RMC) production, brick kilns, and iron and steel production. The discussion focuses on particulate matter (PM), including PM₁₀ and PM₂.₅, sulphur dioxide (SO₂), nitrogen oxides (NOₓ), carbon monoxide (CO), and carbon dioxide (CO₂). A narrative review approach was used to synthesise the technical literature, regulatory material and guidance documents cited in this paper. The comparison indicates that fugitive and process dust is a major concern in cement and RMC operations; combustion-related emissions are particularly important in brick kilns; and the integrated iron and steel production route has multiple emission-generating stages, including coke making, sintering, ironmaking and steelmaking. The review also shows that emission profiles vary with fuel, raw-material composition, production technology, operating conditions and the effectiveness of pollution-control equipment. Accordingly, no single control strategy is suitable for all four sectors. Source enclosure, dust extraction, fabric filters, electrostatic precipitators, improved combustion, flue-gas treatment, energy efficiency and regular monitoring are among the measures relevant to specific processes. Because this paper is based on secondary sources and does not report original stack measurements, its comparison is qualitative rather than a ranking of measured emission rates.
Air pollution; brick kilns; cement industry; emission control; iron and steel industry; ready-mix concrete.
Vol 3 · Issue 11
Open Access
Ramkumar T, Hemanth G, Jaivarshan G, Kamalesh S, Veerasamy A
The temperature of a lithium-ion battery decides how much power it can deliver, how safe it is and how long it lasts, and uneven heating inside the pack shortens its life further. This paper presents the design of a liquid-cooled thermal management system for a 24 V, 10 Ah lithium-ion pack built as seven series groups of four parallel cells (7S4P), together with the prototype assembled from it. An aluminum panel carries heat from the pack to a copper tube, and a 12 V pump moves an inhibited propylene-glycol/water coolant from the tube to a fan-cooled radiator. Six DS18B20 sensors read the pack temperature and an ACS758 sensor reads the battery current. An ESP32 runs a PID controller whose gains are scaled by a fuzzy supervisor using the highest temperature, its rate of rise and the load current, and it drives the pump and fan with PWM signals. A separate check sets full cooling and raises an alert on overtemperature or a sensor fault. Design calculations put the pack heat at about 5 W at 10 A and 47 W at 30 A, which needs only 0.04–0.4 L/min of coolant. In a lumped thermal-model run with the load falling from 10 A to 5 A at 30 °C ambient, fuzzy–PID control peaked at 32.77 °C, 0.77 °C above the 32 °C set-point, and ended at 31.98 °C, while the uncooled pack reached 36.51 °C. In the same model ON/OFF control held the temperature closer to the set-point but used about 42 % more auxiliary energy, and at 30 A the cooling saturated with the pack settling at 38.0 °C. These values come from a model and not from measurements. The prototype has been assembled and its display reports a normal-temperature status, but systematic measurements on it are still to be done.
Battery thermal managementFuzzy logicLithium-ion battery
Vol 3 · Issue 11
Open Access
Emekwisia, Chukwudubem C, Oyesetan, Kolade O, Onuoha, Precious C, Eso…
Polymer matrix composites (PMCs) have become important structural and functional materials because their matrix chemistry, reinforcement type, architecture, interface, and processing history can be tailored to meet application-specific requirements. Nevertheless, the development of reliable composite systems remains difficult because material performance emerges from coupled phenomena spanning molecular, microstructural, laminate, component, and environmental scales. Conventional experimental campaigns are consequently expensive and time-consuming, while high-fidelity numerical simulations can become computationally demanding when many formulation and process variables must be explored simultaneously. Machine learning (ML) and broader artificial intelligence (AI) methods provide a complementary approach by learning relationships from experimental, computational, imaging, and process-monitoring data and using those relationships for prediction, optimization, classification, and inverse design. This review synthesizes recent developments in AI-assisted polymer-composite research, with emphasis on literature published from 2018–2024. The review covers material design and selection, mechanical and functional-property prediction, multiscale modelling, manufacturing-process optimization, additive manufacturing, defect detection, damage prognosis, fire performance, tribology, explainable AI, physics-informed learning, and digital-twin-oriented workflows. The literature indicates that ML is most mature for tabular prediction problems and increasingly capable in image-based inspection and surrogate modelling of expensive simulations. However, apparently high prediction accuracy does not necessarily establish generalizability: many studies are limited by small datasets, correlated observations, inconsistent testing protocols, leakage between training and test sets, narrow material families, and insufficient external validation. Accordingly, AI should be viewed as a decision-support and acceleration layer rather than a replacement for materials science, mechanics, or experimental validation.
Polymer Matrix CompositesMachine learningArtificial Intelligence