Latha Mathavan Engineering College

The latest robotic manipulators need light weight and strong mechanical properties of materials to enhance precision, speed and energy saving in the contemporary industrial applications. Nanocomposite materials have been considered as a promising structural based material due to their high strength to weight ratio, increased stiffness, and enhanced vibration resistance as compared to the conventional metallic ones. Nevertheless, robotic manipulators that have to work under dynamic and high-speed conditions are susceptible to structural deformation including vibration, bending, and buckling, which may decrease positioning accuracy and system reliability. This study suggests an AI-based Digital Twin to optimize and predict the structural deformation behavior of a robotic manipulator made of nano-composites to manage such challenges. The suggested design includes finite element analysis, real-time sensor control, deformation forecasting through machine learning, and motion optimization through reinforcement learning. The Digital Twin model is a continued process of real-time sensor values integrating with simulation results in order to track structural responses throughout robotic operation. Machine learning algorithms process vibration, stress and displacement data to predict deformation patterns and reinforcement learning controls robot motion to minimize vibration and structural stress. It has been experimentally demonstrated that the proposed framework had a vibration prediction accuracy of 96.3 %, buckling detection accuracy of 95.9 % and bending prediction accuracy of 96.1 % and its higher than the traditional monitoring techniques. Moreover, the system that is based on reinforcement learning lessened the vibration amplitude to 0.48 mm and robotic movement precision to 96.8 % which increased structural stability when using the robotic at the high-speed mode. These findings indicate that the modeling of a nanocomposite material, Digital Twin, and Artificial Intelligence make it possible to control the structural control, enhance the performance of robots, and be safer in the advanced manufacturing conditions.

The fast changing nature of connected vehicles requires smart systems that can monitor health in real-time and predict failures. This paper introduces an IoT-based Smart Vehicle Monitoring and Predictive Maintenance System that involves onboard sensors, edge computing, and cloud analytics to ensure high reliability and reduce unscheduled downtimes. Multi-modal transportation data (temperature, vibration, fuel consumption, and brake parameters) are continuously measured with the help of ESP32/Arduino devices connected to the CAN bus. An Extended Kalman Filter (EKF) is an algorithm that does nonlinear sensor fusion at the edge to enhance data reliability and noise minimization. The degradation prediction and Remaining Useful Life estimation of the fused time-series data are done with a Long Short-Term Memory (LSTM) network. A new adaptive thresholding system is used to dynamically regulate anomaly sensitivity according to driving context and historical trends. It was evaluated experimentally on 100,000 real-time sample divided into 70% training, 15% validation and 15% testing sample. The proposed framework was found to have 97.4% prediction accuracy with Precision (94.8%), Recall (95.6%), and F1-score (95.2%). EKF preprocessing minimized RMSE by 0.84, and it is a 56-percent improvement in estimation accuracy. The adaptive detection module reduced false positive rate to 4.1% which was 63% lower than the approaches to static threshold. The system forecasted failures almost 22 minutes earlier than it was actually detected which enhanced early detection by almost 30 percent. The edge deployment decreased latency by a factor of 4 to 160 ms, and thus, allowed the creation of nearly real-time alerts. The implementation of visualization and alert management was realized based on the use of Node-RED and Grafana dashboards, along with MQTT-based secure communication. All in all, the framework is statistically proven to be robust, scaled and applicable to next-generation intelligent vehicular IoT ecosystems.

Leakage in oil, gas, and water pipelines poses severe environmental, financial, and operational challenges, especially when traditional monitoring systems fail to detect early or small leaks due to noisy sensor data and limited modeling intelligence. To address these limitations, this study introduces a Whale Optimization Assisted Deep Metaheuristic Learning (WOA-DML) framework designed for smart, IoT-enabled pipeline leakage detection and predictive failure analytics. The proposed system integrates heterogeneous pressure, acoustic, vibration, and flow-rate sensors with a hybrid CNN-BiLSTM model to effectively capture spatial-temporal signal patterns related to pipeline anomalies. The Whale Optimization Algorithm is employed to optimize hyperparameters and feature-selection weights, enabling faster convergence, improved anomaly sensitivity, and stable learning under uncertain operating conditions. Experimental evaluation using benchmark datasets demonstrates that the WOA-DML framework achieves a superior leakage detection accuracy of 98.46 %, with corresponding precision, recall, and F 1 -score values of $97.92 \%, 98.63 \%$, and 98.27 %, respectively. Comparative analysis shows statistically significant improvements ($\mathbf{p}<0.05$) over conventional CNN, LSTM, Hybrid CNN-LSTM, PSO-DL, and GA-DL models. The framework further reduces training time by $23-31 \%$ and converges within 17 iterations, highlighting its computational efficiency. Robustness testing under noise levels up to 30 % confirms that the model maintains accuracy above 94 %, demonstrating high resilience to sensor fluctuations and dynamic pipeline conditions. Overall, the statistical findings validate WOA-DML as a scalable, energy-efficient, and highly reliable solution for next-generation smart pipeline monitoring and proactive maintenance systems.

A smart wheelchair with IoT based interactive health monitoring system is proposed. IoT based live monitoring of elderly people health parameters like temperature, pressure, fall detection and emergency notification button to intimate the emergency situation to the care taker so that the care taker can give instructions from a IoT app to an Alexa device connected with a camera and monitor in the wheelchair system and it also helps elderly people is sitting and getting up from a chair using hydraulics. This smart system can support the elderly and disabled people with high accuracy and efficiency. The proposed Smart Wheelchair consists of various sensors to monitor the body parameters from the wheelchair connected to the central Raspberry Pi 4. In addition to this an emergency notification system is included to intimate the caretaker so that the care taker can give instructions from their Android mobile to an Alexa device, which in turn connected to a display with camera in the wheelchair system. For easy seating and getting off from the chair a mechanical hydraulics system is designed to lift the seat while sitting and to get up from the seat.

During the pandemic, online classes are predominated. However, the new normal needs effective analysis of students’ classroom engagement. Offline classes also have a potential threat to students’ engagement before and especially after the post-covid. Facial Expressions Analysis has become essential in the learning environment, whether it is online or offline. The offline classroom environment is considered a problem environment. Since, it can be easily adapted to the online environment. Notably, in the PTZ camera environment, the recognition becomes more challenging due to varying face poses, limited Field-of-View (FOV), illumination conditions, effects of the continuous pan, zoom-in, and zoom-out. In this paper, facial expression-based student engagement analysis in a classroom environment is proposed. Face detection has been achieved by YOLO (You only look once) detector to find multiple faces in the classroom with maximum speed and accuracy. Consequently, by adopting the Ensemble of Robust Constrained Local Models (ERCLM) method, landmark points are localized in detected faces even in occlusion, and therefore, feature matching is performed. Besides, the matched landmark points are aligned by an affine transformation. Finally, having different expressions, the aligned faces are fed as input to Faster R-CNN (Faster Regions with Convolutional Neural Network). It recognizes behavioral activities such as Attentiveness (Zero-In (ZI)), Non-Attentiveness (NA), Day Dreaming (DD), Napping (N), Playing with Personal Stuff in Private (PPSP), and Talking to the Students’ Behind (TSB). The proposed approach is demonstrated using the TCE classroom datasets and Online datasets. The proposed framework outperforms the state-of-the-art algorithms.