International Journal of Engineering Systems Modelling and Simulation (16 papers in press)

Regular Issues

• Automated recognition of power quality disturbances for internet of power quality things  
  by V. Jomole Varghese, M.P. Vidhya, B. Smitha 
  Abstract: Real-time power quality (PQ) monitoring has become most essential to evaluate the severity of voltage variations for timely protecting distributed energy systems, appliances and equipment connected with internet of things (IoT) networks. In this paper, we attempt to present a low-complexity PQ disturbance (PQD) recognition method for automatically detecting variation events, such as sags, swells, interruptions and transients according to the IEEE Std. 1159. The proposed PQD event recognition method consists of digital filtering, Hilbert transform (HT) and decision tree. The proposed PQD recognition method is evaluated using the simulated PQ signals according to the IEEE Std. 1159 and the real-time PQ signals. The proposed method had a recognition accuracy of 96-100% for detecting the voltage sag, swell, momentary interruption, transient and combined disturbances.
  Keywords: power quality; PQ; power quality disturbance; PQD; Hilbert transform; HT; internet of power quality things; IoPQT.
  DOI: 10.1504/IJESMS.2026.10076587
   
  
• Performance of single elliptical-shaped three-blade Savonius micro-wind turbine installed on car roof: simulation and experimental study  
  by Hemantchandra N. Patel, Kalpesh V. Modi, Manthan A. Modhia, Falak T. Makwana 
  Abstract: The implementation of wind energy conversion devices in automobiles to harness/harvest the energy from wind faces significant challenges, as it affects the vehicles aerodynamic characteristics and overall performance. The simulation and experimentations were conducted to study the effect of installing single elliptical-shaped three-blade Savonius micro-wind turbine (MWT) on car roof on aerodynamics and power generation. The simulations on car model were carried out at various car speeds (20 to 120 km/h) for two cases case-I: car without MWT, and case-II: car with single MWT in enclosure. The simulation results indicated that single MWT generated 2.3481 and 7.6674 W power at MWT rotation of 2,123 and 3,185 RPM (car speeds of 40 and 60 km/h). Experimental results indicated that single MWT generated 1.1562 and 5.41 W effective power at MWT rotation of 1,765 and 2,365 RPM and relative velocity of 16.36 and 23.9 m/s (car speed 59 and 86 km/h).
  Keywords: aerodynamics; drag coefficient; lift coefficient; computational fluid dynamics; CFD; wind energy; micro-wind turbine; MWT.
  DOI: 10.1504/IJESMS.2026.10077182
   
  
• Experimental analysis of soil quality using internet of LoRaWAN and predicting the soil nutrient using federated learning for next generation sustainable agriculture  
  by M. Vinodhini, C. Neeladharan 
  Abstract: Federated Learning has become an emerging technology for the analysis of the soil nutrient index (SNI). To address the existing issue, the proposed hybrid long range (LoRa) with federated learning (LFL)-based real-time soil quality management system in a specific zone, Ambur, Vellore District. The process begins with the collection of real-time soil samples from the specified zone using a LoRaWAN-enabled prototype system, ensuring continuous monitoring and data acquisition. Second, the soil samples undergo scanning electron microscopy (SEM) analysis to identify key organic indicators like biochemical oxygen demand (BOD), chemical oxygen demand (COD), phenol, chloride, and phosphate, which provide insights into soil health and contamination levels. The gathered real-time soil nutrient data is utilised to train a federated learning (FL) model, which predicts the soil nutrient index efficiently while maintaining data privacy. The proposed system integrated approach combines LoRaWAN-based monitoring with advanced AI-driven analytics, enabling effective soil health evaluation.
  Keywords: federated learning; soil nutrient index; SNI; internet of LoRaWAN prototype; sustainable agriculture; real-time soil data; machine learning classifiers; scanning electron microscopy; SEM analysis; federated learning; SNI; LoRaWAN; machine learning.
  DOI: 10.1504/IJESMS.2026.10077681
   
  
• Experimental and multi-objective optimisation study on impact strength and surface roughness in fused filament fabrication of flexible TPU component  
  by Rituparna Saha, Subhash Chandra Panja, Sankar Narayan Patra, Sunil Kumar Sharma, Sovan Sahoo, Sumit Dhar 
  Abstract: The performance of fused filament fabrication (FFF) components is strongly influenced by printing parameters, which govern deposition, interlayer bonding, and surface morphology. For flexible TPU, widely used in energy-absorbing applications, the combined effects of these parameters on impact strength and surface quality remain unclear. This study evaluates layer height, infill angle, and builds orientation using Taguchi L9 design, ANOVA, and optical microscopy, with Grey relational analysis applied for multi-objective optimisation to identify balanced settings. Results show layer height is the most influential factor, improving impact strength through better interlayer diffusion but increasing surface roughness due to stair-stepping. Build orientation affects impact strength, while infill angle has little effect. The optimal combination of 0.17 mm layer height, +-22.5 infill angle, and 0 build orientation achieves a balance between mechanical strength and surface quality. The study demonstrates the role of process parameters and layer morphology in optimising flexible TPU components.
  Keywords: fused filament fabrication; FFF; thermoplastic polyurethane; TPU; Taguchi; analysis of variance; ANOVA; single- objective optimisation; grey relational analysis; GRA.
  DOI: 10.1504/IJESMS.2026.10077850
   
  
• Enhanced image manipulation detection using lightweight MobileNet and meta graph neural networks  
  by Mahejabi Khan, Samta Gajbhiye, Rajesh Tiwari 
  Abstract: Digital manipulation of images has been made widespread by the availability of advanced editing and generative tools. While these tools make it easy to create visual content, they also pose a risk to digital authenticity, security verification, journalism credibility and forensic investigations. The increasing number of manipulation methods calls for accurate and scalable methods for detection. Many existing deep learning models such as conventional CNN, fully convolutional networks, hybrid autoencoders and CNN-LSTM frameworks still suffer from high computational cost, slower inference and limited generalisation on manipulation types. To overcome those limitations, a lightweight detection framework using MobileNet architecture is proposed and augmented with Meta Graph Neural Networks to learn relational features from each other region in the image. Depth-wise separable convolutions allow efficient feature extraction using few parameters. Using the CASIA2 dataset with 12,614 samples, the model achieved 91.67% training accuracy and 99.29% validation accuracy with 0.0220 validation loss, which is better than CNN, FCNN, CNN-AE and CNN-LSTM baselines.
  Keywords: image manipulation detection; lightweight MobileNet; meta graph neural network; Meta-GNN; digital forensics; computational efficiency.
  
  
• Interactive hydraulic analysis of side obstacles and composite structures with a comparative machine learning study for discharge prediction  
  by Rafi M. Qasim, Ammar Salman Dawood, Ahmed Sagban Khudier 
  Abstract: This study investigates hydraulics of a combined weir-gate structure and how lateral side obstacles alter the downstream flow field. Side-obstacle effects were embedded in inputs. Experiments in a rectangular flume produced 75 observations of upstream water level, gate-weir spacing, and flow cross-sectional area to relate downstream flow area and discharge to flow rate, weir and gate flow areas, discharge coefficient, and Froude and Reynolds numbers. The dataset was split into 80% training and 20% testing to evaluate eight machine-learning regressors (decision tree, random forest, bagging, AdaBoost, gradient boosting, XGBoost, KNN, and ANN) for discharge prediction. Gradient boosting performed best (test R2 = 0.9596), whereas the ANN failed (R2 = -0.30), indicating poor suitability for this problem. XGBoost and KNN achieved moderate accuracy (R2 0.75-0.78). As the first systematic comparison for composite weirs-gates with/without lateral obstacles, the work highlights MLs potential to support hydraulic design and water-resources management.
  Keywords: artificial neural network; ANN; discharge prediction; hydraulic engineering; gradient boosting; composite hydraulic structure; machine learning.
  DOI: 10.1504/IJESMS.2026.10078514
   
  
• An intelligent lifecycle framework for strengthening the resilience of critical infrastructure systems  
  by R. Rajesh Kanna, P. Praba Devi, L. Subha, T. Sindhu, K. Sackthivel, S. Namachivayam, S. Anitha 
  Abstract: Critical infrastructure networks (CINs) like power, transport, and communications networks are vital to modern society. These networks can fail catastrophically. A novel paradigm for CIN systemic resilience through intelligent lifecycle management is presented in this study. Proactive vulnerability assessment and repair across the infrastructure asset lifetime from design and building to simulated operation and decommissioning using real-time monitoring, predictive analysis, and automated reaction systems. The current work uses a synthetic dataset of 492 examples reflecting regional power grid operating regimes under varying stress levels. Dataset inputs include component condition, load distribution, and environmental conditions. The study used Pandas for data management, Scikit-learn for prediction models, and Matplotlib for charting in Python. The suggested machine learning system can forecast failures, schedule maintenance, and recommend adaptive control actions to improve global network resilience. The study shows that smart lifecycle management can extend the steady operation of fundamental infrastructure by improving disruptive resistance and resilience. In a quickly changing, uncertain global context, the study provides policymakers and infrastructure stakeholders with a solid, scalable strategy to improve critical infrastructure security and resilience.
  Keywords: systemic resilience; electricity grids; intelligent lifecycle management; predictive analysis; autonomous response; global environment; applied energy; real-time monitoring; vulnerability analysis.
  DOI: 10.1504/IJESMS.2026.10078691
   
  
• Intelligent paddy leaf disease detection and automated pesticide spraying using AI and IoT  
  by Sheeba Santhosh, G. Hari Krishnan, G. Mohandass, R. Sreelakshmi, E. Sivanandam, Ashok Kumar Srinivasan 
  Abstract: Technological innovations that enhance farmland quality and productivity have led to the development of smart farming systems with automation. One of the longstanding threats to food security is paddy leaf disease, which can significantly reduce crop yield and quality. Accurate diagnosis of these diseases has traditionally been challenging; however, recent advances in deep learning and machine vision have enabled more precise detection. In this work, a novel approach for identifying paddy leaf diseases is implemented using convolutional neural networks (CNNs). Neuron-wise and layer-wise visualisation techniques were employed to analyse the networks decision-making process. The model was trained on a publicly available paddy leaf disease image dataset, allowing it to accurately recognise the textures and patterns of lesions associated with different diseases, effectively mimicking human expert evaluation. Additionally, an Arduino-based automated pesticide spraying system was developed and integrated with the disease detection model, enabling targeted and efficient pest management.
  Keywords: deep learning; TensorFlow; Keras; convolutional neural network; CNN.
  DOI: 10.1504/IJESMS.2026.10078931
   
  
• Deep learning-driven forecasting of cloud resource utilisation using PSO-enhanced LSTM for cost-effective scaling  
  by Karthikeyan Sivanandi, C.Sathish Kumar, D. Chitra, S.Silvia Priscila, G. Rajasekaran, S. Suman Rajest 
  Abstract: Modern IT design relies on cloud computings on-demand, scalable resources and services. Cloud computing is expensive, and resource allocation controls it. Traditional resource scalability methods like threshold-based or human supervision are expensive and inefficient. Deep learning for resource forecasting and cost management is growing. In this research, a deep learning-driven algorithm predicts cloud memory and CPU utilisation. The model forecasts CPU and memory usage to enable proactive resource allocation. The suggested paradigm is implemented in Python. The experiment model was trained, validated, and tested using 2019 Google Cluster Workload Traces. The models performance was also compared to conventional models. The PSO-optimised LSTM model forecasted cloud resource use better than all others with an MSE of 0.0398 and an MAE of 0.0135. This was slightly less successful than linear regression but better than CMA-ES. The trained model can be deployed to cloud infrastructure and configured with an automated mechanism that auto-scales based on forecasts. This proactive method improves resource use and could cut costs. The model can reduce manual monitoring by integrating a feedback mechanism for retraining or fine-tuning. This article offers a cost-cutting and resource-efficient technique for cloud service providers and businesses.
  Keywords: LSTM model; cloud resource consumption; lower costs; proactive strategy; manual monitoring; cloud computing.
  DOI: 10.1504/IJESMS.2026.10078937
   
  
• Nature-inspired load balancing in cloud systems: a PSO-GA hybrid optimisation model  
  by S. Balaji, K. Krishna Prasad, S.Silvia Priscila, P.M. Praveen 
  Abstract: Computing optimisation has become an industry standard and is growing rapidly. It provides fast, affordable computing using massively virtualised data centres. Todays apps must serve millions of users with accurate text, photos, videos, and other data quickly and consistently. Cloud load balancing (LB) or resource scheduling is the most critical problem to solve due to resource heterogeneity, interdependencies, and load unpredictability. LB distributes network traffic equitably among software application resources. LB directs workloads to computing resources to boost performance. In complicated, unpredictable situations, simple LB methods fail. Metaheuristics are used for solving complex problems. Nature-inspired metaheuristic (MH) algorithms are being applied in many domains to solve difficult optimisation problems. This paper uses PSO + GA, a hybrid metaheuristic algorithm, to optimise resource scheduling for LB. Comparisons include PSO, ACO and genetic algorithms. Metaheuristic algorithm performance is measured by response time, makespan, and average resource use. Using performance factors, the CloudAnalyst simulator shows PSO + GA, PSO, ACO and GA for LB. The results suggest nature-inspired algorithms may improve load-balancing reliability and efficiency.
  Keywords: computing optimisation; load balancing; particle swarm optimisation; PSO; genetic algorithm; ant colony optimisation; ACO.
  DOI: 10.1504/IJESMS.2026.10079047
   
  
• Intelligent correlation and optimisation of CNC machining parameters for enhanced 3D modelling accuracy in mould manufacturing  
  by Meihua Shi 
  Abstract: This study proposes an intelligent hybrid framework for determining optimal CNC machining parameters to improve the precision of 3D mould modelling. Using statistical experimentation, machine learning algorithms, and multi-objective optimisation, the research shows that machining inputs such as spindle speed, feed rate, depth of cut, and step over affect key performance indicators, including surface roughness (Ra) and dimensional deviation (Dd). A large factorial experimental design included 81 CNC trials. We have lots of data to train and test our models. Spindle speed and feed rate accounted for 61.6% of the variation in machining quality, as determined by statistical analysis and PCA. The study predicted machining outcomes using advanced modelling methods, such as RSM, RF, SVM, and ANN. The ANN model outperformed others (R2 = 0.948, RMSE = 0.0072 mm) by accurately representing nonlinear interactions. The SHAP explainable AI study indicated that spindle speed and feed rate most affect Ra and Dd. The study found the optimal parameter combinations using NSGA-II and Monte Carlo simulations. The optimal spindle speed was 3850 rpm, and the feed rate was 220 mm/min. These settings conserved energy and increased machining accuracy by 15.2%. This framework creates clever, eco-friendly, and adaptable CNC machining systems.
  Keywords: CNC machining optimisation; 3D modelling accuracy; machine learning; ML; surface roughness; Ra; multi-objective optimisation.
  DOI: 10.1504/IJESMS.2026.10079467
   
  
• Optimising railway delays: a comparative analysis of advanced algorithms for minimising delays  
  by Shyamal Sundar Banerjee, Avijit Bhattacharyya, Subhash Chandra Panja 
  Abstract: Railway delays can significantly impact the efficiency of train operations and the satisfaction of passengers. This paper provides a comparative analysis of advanced algorithms for minimising railway delays. The algorithms examined include the Hungarian method, integer linear programming, genetic algorithm, simulated annealing, Tabu search, ant colony optimisation, and frog optimisation. The objective is to determine the most effective algorithm for optimising railway schedules and minimising delays.
  Keywords: railway delays; optimisation algorithms; Hungarian method; integer linear programming; genetic algorithm; simulated annealing; Tabu search; ant colony optimisation; ACO; frog optimisation.
  DOI: 10.1504/IJESMS.2026.10080088
   
  
• Wavelet-enhanced deep learning framework for real-time defect identification in friction stir welding  
  by B. Sudha, Kathiravan Srinivasan 
  Abstract: Friction stir welding (FSW) is widely used in shipbuilding, automotive, and aerospace to weld light, high-strength materials without melting. The processs tendency to generate voids, tunnelling flaws, and insufficient fusion weakens the joint. Detecting subtle or subsurface faults with visual inspection, X-ray radiography, and ultrasonic testing is time-consuming, expensive, and ineffective. Traditional machine learning methods that use manually collected characteristics cannot generalise across weld settings. The work offers a wavelet-based deep learning model for autonomous FSW joint defect inspection to overcome these limits. Weld signal time-domain and spatial-frequency characteristics from sensor readings are modelled using the DWT. Multi-resolution characteristics fed into deep Q-network learning aid fault determination and spatial feature extraction. Wavelet decomposition greatly improves the models sensitivity to tiny signal fluctuations indicating hidden defects. A vast database of FSW samples with varying defects and process parameters was used to evaluate the suggested approach. The studies had 97.8% flaw-detection accuracy, much greater than many sophisticated baseline models. Results show that Wavelet-DQL enrichment improves results. This study validates the Wavelet-DQL as a scalable, effective, and highly accurate real-time, non-destructive FSW defect diagnosis method for Industry 4.0 production system quality assurance.
  Keywords: deep learning; defect identification; friction stir welding; FSW; non-destructive evaluation; NDE; wavelet transform; discrete wavelet transform; DWT.
  DOI: 10.1504/IJESMS.2026.10080121
   
  
• Energy consumption reduction in architectural by using smart thermobimetal to as a sun shading optimal transmission of natural light  
  by Xiaodan Liang, Dan Wang 
  Abstract: This study examines the application of smart thermobimetals to architectural facades to reduce energy consumption and enhance daylighting in buildings. Thermobimetals, which deform in response to temperature fluctuations, offer a passive approach to energy-efficient architectural design by adaptively responding to ambient conditions. This study integrates experimental data and computational software (DesignBuilder, EnergyPlus) to assess the efficacy of thermobimetal sunshades in a Tehran-based building, emphasising the reduction of cooling loads and enhancement of natural light infiltration. The research indicates that incorporating thermobimetals into the facade design achieved a 41% decrease in energy usage, equating to an annual savings of about 350,000 kWh, alongside a reduction of 6,000 kg of CO2 emissions. These findings underscore the considerable potential of thermobimetals in diminishing the energy requirements of buildings, especially in areas with elevated solar radiation. The findings further illustrate the advantages of employing thermobimetals in improving occupant comfort through the regulation of ideal daylighting levels and the mitigation of glare. This study provides significant insights into sustainable building design, demonstrating thermobimetals as an efficient alternative for energy-efficient architecture. The results are especially pertinent for structures in regions with significant sun exposure, where intelligent materials can greatly influence energy conservation and environmental sustainability.
  Keywords: natural light; reduction of energy consumption; architectural space; thermo bimetal; smart material.
  DOI: 10.1504/IJESMS.2026.10080426
   
  
• Enhancing smart metre security in smart grids using Spritz stream cipher  
  by S. Shabana Banu, M.S. Sujatha 
  Abstract: Supervisory control and data acquisition (SCADA) systems have been integrated into todays power grid infrastructure, achieving the automation of electricity generation, transmission and distribution and is now known as the Smart Grid. But this greater level of inter-connectivity also creates new cyber security risks such as spoofing, man-in-the-middle (MITM) attacks and data manipulation. The confidentiality, integrity and availability CIA of critical data transferred between smart metres and substation by these risk sensors and substations may be compromised resulting in financial loss, operations degradations, reduced public confidence. To address these concerns, the cryptographic stream cypher Spritz is proposed as a suitable encryption algorithm for using in this work. Spritz provides better security with higher randomness and is lightweight enough for smart metres. By integrating Spritz into smart grid communication systems, the proposed approach enhances data confidentiality and integrity while maintaining computational efficiency, thereby supporting a more secure and resilient smart grid infrastructure.
  Keywords: cybersecurity; data integrity; SCADA; Spritz encryption; smart metres.
  DOI: 10.1504/IJESMS.2026.10080474
   
  
• Adaptive protection strategies for critical infrastructure systems leveraging real-time threat intelligence  
  by P.V. Bhuvaneshwari, P. Sudhakar, Abhishek Julka, V.S. Devi Priya, G. Ragu, A. Anithalakshmi 
  Abstract: Critical infrastructure (CI) networks such as power grids, water supply systems, and transport networks become more and more susceptible to new-generation physical and cyber threats. A new adaptive protection policy architecture that uses real-time threat intelligence to dynamically adjust security is presented and examined in this study. A hidden feature of the proposed structure is an AI system that feeds threat data to predict attacks. We used a synthetically generated dataset, CI-Threat-Sim-2024, consisting of 480 threats against a simulation of a regional electric power grid model. Each instance includes a richer set of features, including network traffic abnormalities, sensor physical notices, and threat intelligence sources. Python was used to develop and train an adaptive defensive model utilising Scikit-learn for machine learning, Pandas for data manipulation, and Matplotlib/Seaborn for data visualisation. A random forest classifier classifies threats nature and intensity, and the model dynamically allocates defensive resources very accurately. The report states that real-time threat intelligence boosts critical infrastructure resilience by creating a dynamic, intelligent protection mechanism. This study is the best proof that threat intelligence and intelligent systems can create an invulnerable and more secure system for our nations most important assets.
  Keywords: adaptive security; critical infrastructure protection; machine learning; anomaly detection; cybersecurity; real-time threat intelligence; risk management; resilience.
  DOI: 10.1504/IJESMS.2026.10080600