The "INDIAai" National AI Portal of India offers weekly articles that showcase the AI research contributions of universities across the country. Each article provides an in-depth report on the work of a specific university, allowing researchers and students to present concise explanations of their research. 

This week, the portal highlights the top AI research contributions from Kalasalingam Academy of Research and Education, Krishnankoil, Tamilnadu, India.

Kalasalingam Academy of Research and Education (KARE) (Deemed to be University), formerly known as Arulmigu Kalasalingam College of Engineering, was founded in 1984 by Kalvivallal Thiru T. Kalasalingam under the Kalasalingam and Anandam Ammal Charities trust. The Founder Chairman was a philanthropist and freedom fighter. Kalasalingam is situated in the picturesque Western Ghats of southern Tamilnadu, at the pristine foothills. In 2006, the college was granted the status of a Deemed University. For thirty-seven years, the institution has been providing services to the community and meeting the requirements of students from all backgrounds. 

Kalasalingam provides undergraduate, postgraduate, and doctoral programs in a variety of fields, including engineering, science, technology, and humanities. It is the first institution in India to offer a specialized B.Tech program in engineering for students who are differently abled, including those who are speech and hearing impaired. NAAC has re-accredited the institution with a 'A++' grade. NBA has accredited six undergraduate programs under Tier-1. 

Optimal deep learning model for classification of lung cancer on CT images

Authors:

• Lakshmanaprabu S.K - Department of Electronics and Instrumentation Engineering, B.S.Abdur Rahman Crescent Institute of science and Technology, Chennai, India
• Sachi Nandan Mohanty - Department of Computer Science & Engineering, Gandhi Institute for Technology, Bhubaneswar, India
• Shankar K - School of Computing, Kalasalingam Academy of Research and Education, Krishnankoil, India
• Arunkumar N - Department of Electronics and Instrumentation Engineering, SASTRA University, Tanjavur, India
• Gustavo Ramirez - Department of Telematics, University of Cauca, Colombia

Lung cancer remains one of the most fatal cancers worldwide, where early detection significantly enhances survival rates. The study presents an innovative automated diagnosis classification system for lung cancer using Computed Tomography (CT) scans. The proposed methodology integrates Optimal Deep Neural Networks (ODNN) with Linear Discriminant Analysis (LDA) to improve the accuracy of diagnosing lung cancer. The process involves extracting deep features from CT images of the lungs, followed by dimensionality reduction using LDA to classify lung nodules as either malignant or benign. The ODNN approach is further optimized through a Modified Gravitational Search Algorithm (MGSA) to enhance classification performance. The system shows promising results, achieving a sensitivity of 96.2%, specificity of 94.2%, and overall accuracy of 94.56%.

The AI-based framework not only facilitates the detection of cancerous lung nodules but also offers a precise and efficient method for distinguishing between benign and malignant tumors. This approach underscores the transformative role of AI in improving diagnostic accuracy and timely intervention in lung cancer treatment.

Comprehensive learning particle swarm optimization for reactive power dispatch

Authors

• K. Mahadevan - Electrical & Electronics Engineering, Kalasalingam University, Krishnankoil – 626 190, Tamilnadu, India
• P.S. Kannan - Electrical & Electronics Engineering, Thiagarajar College of Engineering, Madurai – 625 015, Tamilnadu, India

Reactive Power Dispatch (RPD) plays a crucial role in optimizing power grid operations by reducing system losses, improving voltage stability, and enhancing overall grid security. It is an optimization problem aimed at minimizing active power losses and alleviating grid congestion by adjusting reactive power control variables such as generator voltages, transformer tap-settings, and capacitor banks. Effective RPD ensures better voltage control, an improved voltage profile, enhanced system security, and increased power transfer capabilities, which contribute to more efficient power system operation.

This study explores the solution of the RPD problem using Particle Swarm Optimization (PSO), a well-known optimization technique. To address the issue of premature convergence often associated with PSO, a comprehensive learning strategy, termed Comprehensive Learning Particle Swarm Optimization (CLPSO), is introduced. The CLPSO approach improves upon traditional PSO by enhancing its exploration capabilities, leading to more robust optimization results. Three different test cases—minimizing real power losses, improving voltage profiles, and enhancing voltage stability—were conducted using standard IEEE 30-bus and 118-bus test systems. Comparative results show that both PSO and CLPSO are feasible and effective in solving the RPD problem, with CLPSO demonstrating superior performance in overcoming the limitations of traditional PSO.

This AI-driven optimization approach underscores the potential of advanced algorithms like CLPSO to improve grid efficiency and reliability, offering promising solutions for large-scale power system operations.

Hybridization of firefly and Improved Multi-Objective Particle Swarm Optimization algorithm for energy efficient load balancing in Cloud Computing environments

Authors

• A. Francis Saviour Devaraj - Department of Computer Science and Engineering, Kalasalingam Academy of Research and education, India
• Mohamed Elhoseny - Faculty of Computers and Information, Mansoura University, Egypt
• S. Dhanasekaran - Department of Computer Science and Engineering, Kalasalingam Academy of Research and education, India
• E. Laxmi Lydia - Computer Science and Engineering, Vignan’s Institute of Information Technology (Autonomous), India
• K. Shankar - Department of Computer Applications, Alagappa University, Karaikudi, India

Load balancing in Cloud Computing (CC) is essential for optimizing resource utilization, ensuring efficient task execution, and managing workloads across servers, networks, and computers. This study introduces a novel load balancing algorithm, termed FIMPSO, which is a hybrid of the Firefly Algorithm (FF) and Improved Multi-Objective Particle Swarm Optimization (IMPSO) technique. The FIMPSO approach combines the advantages of both algorithms: FF minimizes the search space, while IMPSO identifies optimal solutions by selecting the global best (gbest) particle using a minimum distance from a point to a line technique, improving precision and convergence.

The proposed FIMPSO algorithm optimizes cloud resource allocation by effectively balancing load distribution and improving key performance metrics such as response time and resource utilization. In simulations, FIMPSO outperformed traditional methods, achieving an average response time of 13.58ms, CPU utilization of 98%, memory utilization of 93%, reliability of 67%, and throughput of 72%, with a make span of 148. These results highlight its superior efficiency in managing cloud environments compared to other algorithms.

The application of AI-based hybrid algorithms like FIMPSO demonstrates significant potential in enhancing cloud computing systems, driving improvements in computational efficiency, and ensuring reliable resource management. This study reinforces the growing role of AI-driven optimization in addressing complex load balancing challenges in dynamic cloud environments.

Artificial neural network approach for fault detection in rotary system

Authors

• S. Rajakarunakaran - A.K College of Engineering, Anand Nagar, Krishnankoil 626190, Tamilnadu, India
• P. Venkumar - A.K College of Engineering, Anand Nagar, Krishnankoil 626190, Tamilnadu, India
• D. Devaraj - A.K College of Engineering, Anand Nagar, Krishnankoil 626190, Tamilnadu, India
• K. Surya Prakasa Rao - Department of Industrial Engineering, Anna University, Chennai 600025, Tamilnadu, India

Fault detection and diagnosis in technical systems is critical for ensuring safe and efficient operations, particularly in environments where machinery failure can lead to significant damage or risk to human lives. This research focuses on the centrifugal pumping rotary system and presents an AI-based model for fault detection using artificial neural networks (ANNs). Two different neural network approaches are employed: a feed-forward network with backpropagation algorithm and a binary adaptive resonance theory network (ART1).

The study involves training and testing the ANN models using data generated under various operating conditions, including fault scenarios, via real-time simulation through an experimental model. The models are designed to detect and diagnose seven categories of faults in the centrifugal pumping system. The performance of both the backpropagation and ART1 models is evaluated and compared based on their fault detection accuracy.

The results demonstrate the effectiveness of AI-based models in identifying faults in complex mechanical systems, showcasing the potential of neural networks to enhance system reliability and prevent major machinery failures. The comparative analysis provides insights into the strengths of each model, contributing to the development of more robust fault detection mechanisms in industrial applications. This study highlights the transformative role of AI in predictive maintenance and fault detection, underscoring its importance in improving safety, reducing downtime, and optimizing the performance of critical technical systems.

Application of modified NSGA-II algorithm to multi-objective reactive power planning

Authors

• S. Ramesh - Instrumentation and Control Engineering Department, Arulmigu Kalasalingam College of Engineering, Anand Nagar, Krishnankoil, Tamilnadu, India
• S. Kannan - Electrical and Electronics Engineering Department, Kalasalingam University, Anand Nagar, Krishnankoil, Tamilnadu, India
• S. Baskar - Electrical and Electronics Engineering Department, Thiagarajar College of Engineering, Madurai, Tamilnadu, India

This paper presents the application of the Modified Non-Dominated Sorting Genetic Algorithm-II (MNSGA-II) to address the multi-objective Reactive Power Planning (RPP) problem, focusing on minimizing operating and VAR allocation costs, improving the bus voltage profile, and enhancing voltage stability. The MNSGA-II incorporates a Dynamic Crowding Distance (DCD) procedure to ensure better diversity among solutions. Performance analysis was conducted on standard IEEE 30-bus, practical 69-bus Indian, and IEEE 118-bus systems, and results were compared with the original NSGA-II and validated against the reference pareto-front using Covariance Matrix Adapted Evolution Strategy (CMA-ES). 

The study shows that MNSGA-II outperforms NSGA-II based on various multi-objective performance metrics. A decision-making process using the TOPSIS method was applied to select the best compromise solution from the pareto-optimal set, highlighting MNSGA-II’s potential to effectively solve multi-objective RPP challenges.