AISSMS College of Engineering

The detection of toxic gases at trace levels is crucial for environmental monitoring and human safety. In this work, we report the synthesis and gas-sensing performance of a novel redox-engineered nanocomposite comprising gold (Au ^0 ) nanoparticle cores encapsulated within benzene sulphonic acid (BSA) doped polypyrrole (PPy) shells, denoted as Au ^0 @PPy ^+· C_6H_5SO_3^- . The nanocomposites were synthesized via in situ oxidative polymerization of pyrrole in the presence of HAuCl _4 and BSA, resulting in a uniform core–shell architecture with enhanced redox activity. The synergistic combination of catalytic Au ^0 cores, conductive PPy shells, and acid doping yielded a material with superior surface reactivity and charge transport properties. Chemiresistive sensors fabricated using this composite exhibited highly sensitive, reversible, and reproducible responses toward selected toxic gases, including Cl _2 , NH _3 , CO, SO _2 , NO _2 , and LPG under ambient conditions. Notably, the sensor displayed a rapid increase in resistance upon exposure to oxidizing gases (Cl _2 , SO _2 , NO _2 ), and a sharp decrease upon interaction with reducing gases (NH _3 , CO, LPG), confirming the p-type nature of the composite. The device demonstrated excellent sensitivity, long-term stability, and full recovery upon air purging, highlighting its potential for real-time, low-power toxic gas monitoring applications. This study provides valuable insight into redox modulation and core-shell engineering strategies for the design of high-performance gas-sensing materials.

The application of remote sensing data, specifically Nighttime Light (NTL) imagery derived from the Visible Infrared Imaging Radiometer Suite (VIIRS), has rapidly emerged as an indispensable methodology for measuring fine-grained economic activity. This approach is particularly critical in developing nations like India, where traditional socioeconomic data may be sparse, outdated, or inconsistently reported across heterogeneous sub-national regions. The inherent objective nature of NTL data offers a crucial advantage over conventional survey or census methods, providing a reliable and frequently updated proxy for urban expansion, electrification, and consumption.While existing academic literature has effectively established correlations between NTL and broad macroeconomic indicators, such as national GDP or overall crime rates, a significant research gap persists. The majority of these studies rely on coarse state or district-level aggregations and older, less precise DMSP-OLS data. This level of aggregation critically fails to capture the spatial inequality, heterogeneity, and highly localized development that occurs within India’s rapidly growing metropolitan areas. Furthermore, relying predominantly on traditional econometric models overlooks the enhanced predictive and pattern-recognition capabilities offered by modern ensemble machine learning techniques.To bridge this crucial gap, this paper introduces a novel machine learning-based framework designed for assessing regional economic development at a granular, sub-district resolution. The core methodology involves the fusion of multi-source geospatial data: monthly VIIRS NTL composites (serving as a high-frequency economic proxy) are integrated with specific ground-truth indicators derived from OpenStreetMap (OSM) Points-of-Interest (POI) data, which track concentrated consumption and nightlife amenities.The analytical pipeline is multi-faceted and robust: it includes rigorous preprocessing and noise filtering, followed by density-based spatial clustering (DBSCAN) for reliable hotspot detection and temporal decomposition (STL) for isolating seasonality and identifying anomalies. The core predictive component utilizes an ensemble regression model (XGBoost) to forecast short-term economic momentum (proxied by radiance growth). The empirical validation achieved a strong predictive accuracy (R2 = 0.6828) on the unseen test set, confirming the superiority of the machine learning ensemble approach for this complex time-series forecasting task. Ultimately, this work contributes an open-source reproducible pipeline and yields actionable insights critical for evidence-based urban governance and policymaking.

Underwater image analysis often suffers from low visibility, color distortion, and scattering effects, which reduce the performance of object classification and detection models. We propose a hybrid augmentation pipeline combining Gaussian Blur, Gaussian Noise, and Generative Adversarial Networks (GANs) to synthetically enhance training datasets. The proposed method first applies Gaussian-based degradations to simulate real-world underwater distortions and then employs an Attention-Guided Self-Adaptive GAN (AGSA-GAN) based generator to produce realistic augmented samples. This improves dataset diversity and robustness against varying underwater conditions. Experimental evaluation on Ocean Dark dataset (low light images) demonstrates improvements in object detection. The proposed augmentation approach is particularly effective for deep learning model such as YOLO, making it suitable for underwater robotics, marine debris detection, and ecological monitoring. Ablation study is done on augmented images using YOLO model and evaluation metrics, Precision, Recall and mAP are measured which shows better performance on combined images of Gaussian blur and noise when applied to AGSA-GAN, Evaluation metrics obtained are precision, Recall and F1 score, mAP0.5, mAP0.5-0.95 and Accuracy as 0.853, 0.849, 0.851, 0.867, 0.548, and 0.718 respectively.

Dysarthric speech recognition (DSR) is very challenging due to the vast variability and low intelligibility of the voice, which is associated with distinct neurological disorders. Traditional speech recognition systems fail to identify speech with dysarthria due to the wide variation in prosody and intonation. Various systems have been developed for detecting DSR, but they are limited by lower generalization, suboptimal feature representation, and lower recognition rates. This paper presents the DSR using a Deep Convolutional Neural Network and Long Short-Term Memory (DCNN-LSTM), which provides better feature depiction, superior temporal correlation, and long-term connectivity in the features. The DCNN-LSTM helps boost the discriminative properties of the voice to characterize the prosodic, intonational, spectral, and temporal variations associated with dysarthria. The effectiveness of the DSR scheme is evaluated on the UASpeech dataset for digit recognition, achieving 93.88% accuracy.

This study presents an integrated experimental and SYSWELD-based thermo-mechanical investigation of TIG and MIG welded AISI 316 L stainless steel joints. Multi-pass TIG and MIG welding experiments were conducted on 12 mm thick SS316L plates using a Taguchi L4 orthogonal array by varying welding current, filler wire diameter, and shielding gas flow rate. The optimized welding condition of 170 A current, 3 mm filler wire diameter, and 14 L/min shielding gas flow rate resulted in a fusion zone microhardness of 342 HV and a maximum ultimate tensile strength of 682 MPa. Different base metal (BM), heat-affected zone (HAZ), and fusion zone (FZ) sections were identified by optical microscopy with dendritic solidification and heat-input-dependent grain coarsening. SYSWELD-based finite element analysis predicted concentrated residual stresses in the fusion zone and compressive stresses in adjacent regions. The study establishes a correlation between welding parameters, heat input, microstructural evolution, residual stress behavior, and mechanical properties. The combined experimental and simulated investigation demonstrates the suitability of TIG and MIG welding for producing high-integrity SS316L joints for structural and high-temperature engineering applications, with simulation predictions showing good agreement with experimental observations and a maximum deviation of 1.18