Control Systems & Robotics (CSR)
Track A3F2 CSR 3: Control Systems & Robotics (CSR) 3
Room: F2. 501 Kadamaian
Chair: Siow Cheng Chan (University Tunku Abdul Rahman, Malaysia)
4:30 Fault-Tolerant Control for a Four-Wheeled Independently Steerable Power-Driven Mobile Robot
Gopika Gireesh (College of Engineering Trivandrum, India & Nil, India); Hari Kumar R and Lal Priya P S (College of Engineering Trivandrum, India)
This work extends a fault-tolerant control (FTC) strategy based on kinematic control, previously applied to four-wheel Mecanum wheel systems, to an omnidirectional mobile robot equipped with four independently steerable powered wheels, a configuration that remains underexplored at the kinematic level. Such a robot offers unparalleled maneuverability by independently controlling both the steering and rolling of each wheel, allowing effortless movement in any direction. This advanced mobility makes these robots highly suitable for real-world applications that require precise navigation within confined or dynamic spaces, including warehouse automation, healthcare, and search and rescue operations. The proposed approach maintains trajectory tracking effectively by exploiting kinematic redundancy, without relying on dynamic compensation. The system can successfully complete the task despite faults that affect up to two wheels; however, it cannot accommodate faults beyond this limit. The effectiveness of the control scheme under single and two-wheel actuator fault conditions has been validated through MATLAB simulations.
4:45 Conversational Simulator-Based Study on the Difference in Impressions of Turn-Taking Behaviors in Video and/or Audio
Masahide Yuasa (Shonan Institute of Technology, Japan)
To develop conversational robots or agents that can communicate smoothly with humans, it is essential to conduct studies that deepen our understanding of the social and emotional aspects of human communication. Previous studies have focused on turn-taking behaviors, which inherently involve social and emotional elements. These studies have explored how varying the timing of turn initiation or termination can create different emotional impressions. Although they have yielded valuable insights, a key issue remains: these studies either used a mix of visual and auditory stimuli or relied solely on auditory input. Therefore, the distinct influences of visual and auditory modalities have not been sufficiently examined. In this study, we investigated the differences in impressions created by visual and auditory stimuli in turn-taking behaviors using a conversational simulator. The experiment involved three types of turn-taking behaviors-Overlap, No-gap-no-overlap, and Gap-and three media conditions: Video+Audio, Video-only, and Audio-only. Participants watched or listened to conversations involving three virtual characters and rated their impressions. The results showed that impressions of agreeableness and politeness were rated significantly higher in the Video+Audio condition compared to the Audio-only stimulus. Moreover, there were no significant differences between the Video+Audio and Audio-only conditions for the factors of relaxation, modesty, respect, and closeness. This suggests that nonverbal cues expressed through video during turn-taking play a significant role in conveying politeness and agreeableness. These findings can inform the design of conversational robots or agents by highlighting the importance of incorporating visual cues into turn-taking behaviors, especially when considering differences between visual and auditory aspects.
5:00 Coordinated Dual-Arm Manipulation Using Reinforcement Learning: a Soft Actor-Critic Approach on the Poppy Humanoid
Allen Jacob George (Birla Institute of Technology and Science Pilani, India); Abhishek Sarkar (Birla Institute of Technology and Science Pilani, India & Hyderabad Campus, India); Joyjit Mukherjee (BITS Pilani Hyderabad Campus, India)
In this work, we present a reinforcement learning-based framework for dual-arm object manipulation using the Poppy humanoid robot platform. Our approach addresses the challenges of coordinated grasping and object transport by decomposing the task into two sequential stages. In the first stage, a single policy is trained using the Soft Actor-Critic (SAC) algorithm to simultaneously control both arms and the torso to perform object grasping. In the second stage, a separate SAC policy is trained to move the grasped object to a specified target location. To improve learning efficiency, we incorporate human demonstration data during training. This integration of expert guidance with sample efficient deep reinforcement learning enables the system to achieve robust, coordinated manipulation behavior. Further validation of the simulation results were tested by putting the Poppy robot joint trajectories generated from the simulation results. Both simulation and experimental results are shown to demonstrate that our method can successfully perform dual-arm object grasping and relocation with stable and synchronized motion.
5:15 Local Acceleration on Planar Mobile Platform with Front Differential Drive and Rear Omni Wheels
Chayapat Leardngammongkolkul, Kasem Hutapornprasert, Natchanont Phanphakdeewong and Ronnapee Chaichaowarat (Chulalongkorn University, Thailand)
Differential drive mobile platforms are widely applied in mobile robotics. Omni wheels provide a practical alternative to caster wheels, enhancing mobility and stability by enabling lateral motion while preserving the ground contact point. This research examines a planar mobile platform featuring front differential drive wheels and rear omni wheels, intended for an electric wheelchair. The chassis was constructed from carbon fiber tubes joined by 3D-printed components. Two 4-inch geared BLDC hub motors with integrated hall sensors, controlled by the ODrive motor controller, drive the front wheels, while two 125-mm omni wheels serve as passive supports on the rear axle. Experiments were conducted by varying the speeds of the front left and right wheels to observe lateral and longitudinal acceleration at three distinct positions: centrally located between the left and right wheels, slightly posterior to the front axle, and adjacent to the left and right omni wheels at the rear axle. During the rotation of the wheelchair around an instantaneous center of rotation, the magnitudes of centrifugal acceleration in the longitudinal and lateral components may vary depending on the passenger's location. The findings from this study can serve as a framework for regulating motor speeds during wheelchair turns to reduce discomfort.
5:30 Planetary Geared CVT Using Electromagnetic Brake to Adjust Slipping Torque of Ring Gear
Jormpoom Sukdaeng, Pacharawit Yokyong, Supanat Hathanglarn and Ronnapee Chaichaowarat (Chulalongkorn University, Thailand)
Planetary gears offering high reduction ratios with concentric input and output shafts are widely applied in various mechatronic systems. For optimizing the efficiency of machines across their range of operating conditions, this paper presents a design concept of the planetary geared continuously variable transmission (PG-CVT) using an electromagnetic (EM) brake to adjust the slipping torque of the rotatable ring gear. The input torque is applied to the sun gear while the output shaft is connected to the planet carrier. The secondary velocity source for driving the ring gear is replaced by the brake resisting the rotation, which requires less energy to operate. When the brake is fully locked, the planetary gear operates with stationary ring gear. The brake torque required to support the ring gear is the linear combination of the input torque at the sun gear and the output torque at the carrier. When the brake is freely slipped, the output torque is limited to zero although the sun gear is rotating. The variable transmission ratio can be achieved by adjusting the torque applied to the ring gear, which results in changing the torque conversion ratio. In this paper, the experimental prototype of the PG-CVT was designed and built. At different output torque and brake torque conditions, the constant speed tests were conducted to observe the input torque required for maintaining the constant speeds. The static friction and the damping of the PG-CVT were characterized. It is worth noting that the zero or negative damping phenomenon was observed from a constant brake torque. The findings of this study are fundamental for implementing velocity feedback torque control for achieving desired dynamic responses.
5:45 Fuzzy PID Control Modeled by T-s Fuzzy System for Train Speed Tracking in Virtual Coupling
Yiting Liang (Beijing Jiaotong University, China & 无, China); Jian Wang, Debiao Lu, Jiang Liu and Bai-gen Cai (Beijing Jiaotong University, China)
To address the challenge of stability analysis for traditional nonlinear fuzzy PID controllers in train virtual coupling, this paper proposes a dynamic modeling approach for speed errors based on the T-S fuzzy model. The T-S fuzzy model is constructed via fuzzy rules to achieve local linear approximation of nonlinear error dynamics. Leveraging Lyapunov stability theory, local asymptotic stability conditions are derived using Linear Matrix Inequalities (LMIs), and Particle Swarm Optimization (PSO) algorithm is employed for multi-objective optimization of PID parameters, considering response speed, control smoothness, and gain robustness comprehensively. Simulation results in virtual coupling following scenarios demonstrate that the controller achieves high-precision tracking, the mean error shows a 19.67% reduction compared to traditional fuzzy PID, the mean squared error exhibits a 20.51% reduction and control effort fluctuations are significantly reduced. This study establishes a local stability framework for T-S fuzzy control in virtual coupling systems, providing a theoretical basis for energy-efficient multi-objective optimization under complex dynamics.
Power, Energy & Electrical Systems (PES)
Track A3F3 PES 3: Power, Energy & Electrical Systems (PES) 3
Room: F3. 502 Mesilau
Chair: Yu Zheng Chong (Universiti Tunku Abdul Rahman, Malaysia)
4:30 Design and Optimization of a Highly Efficient Dual-Absorber Heterostructure for Next-Generation Photovoltaics
Venkateswarlu G and Umakanta Nanda (VIT-AP University, India); Pratap Kumar Dakua (Vignan's Institute of Information Technology, India & Duvvada, Vishakapatnam, AP, India)
Perovskite solar cells (PSCs) face significant challenges such as instability, high recombination, poor charge transport, and mismatched band alignment, which limits their power conversion efficiency (PCE). These issues are more pronounced in multi-junction or heterojunction configurations. This study addresses them by proposing an advanced and novel dual-absorber thin-film heterojunction structure (THSC) is Au/CBTS/BiFeO 3/CIGS/PDINO/FTO/Ni using the Key Materials such as CBTS and BiFeO 3 improves high carrier mobility, optimal layer thickness, controlled defect levels, tailored doping concentrations, and minimized interface defects, CIGS enhances light absorption and provides efficient charge carrier generation due to its tunable band gap. This contributes to reduced recombination losses, improved photocurrent, and overall device efficiency. The structure of the THSC device was optimized and simulated to have remarkable photovoltaic parameters, impressive performance, with a PCE of 40.10%, Jsc of 35.81 mA/cm 2, Voc of 1.31V and FF of 88.51%. The novel architecture proposed guidance for future experiments and simulations aimed at developing high-efficiency, stable thin-film heterojunction solar cells in next-generation photovoltaics.
4:45 Simulation of M13 Bacteriophage-Based Piezoelectric Nanogenerators
Glenn C. Virrey, Himeko Andrei Noto, Mariah Jane Sicam, Mike Denver Tolentino, John Josea Umali and Leanne Vince Vergara (University of Santo Tomas, Philippines)
Piezoelectric nanogenerators (PENGs) based on biological materials, such as the M13 bacteriophage, present a promising avenue for sustainable and biocompatible energy harvesting. This study investigates the influence of electrode material, geometry, and thickness on the performance of M13-based piezoelectric nanogenerators through Finite Element Analysis (FEA) and MATLAB Simulink circuit simulations. Four electrode materials-gold, silver, copper, and stainless steel-were examined with varying thicknesses (20 µm, 120 µm, and 200 µm) and geometries (square and rectangular). Simulation results revealed that gold square electrodes at 20 µm achieved the highest voltage output, reaching 10.383 V. Statistical analysis using one-way ANOVA confirmed the significance of material selection and geometry in optimizing energy conversion efficiency. The test displayed scores of 0.2017, 0.1845, and 0.2953 for the square electrodes with thicknesses of 20 µm, 120 µm, and 200 µm, respectively, indicating minimal discrepancies between them. Furthermore, circuit simulations demonstrated that energy harvested using the optimal configuration could be effectively stored in a lithium-ion battery. Among the chosen electrode materials, gold generated the quickest charging time of 371.42 hours with and without load, while stainless steel charged the slowest. Notably, thinner electrodes produced better voltage and power outputs, while increased thickness resulted in diminished performance. These findings highlight the critical role of electrode design in enhancing the performance of biologically inspired PENGs and pave the way for their integration into self-powered wearable electronics and biomedical sensors. Future research should explore experimental validation and assess long-term environmental durability for real-world applications.
5:00 Multiport Current-Fed Asymmetric Bidirectional DC-DC Converter for Hybrid Polar DC Microgrid System
Shangyi Li, Mingjun Jiang and Ma Jianjun (Shanghai Jiao Tong University, China)
To promote sustainable development and reduce energy loss, hybrid polar DC microgrid system has been studied and put into use worldwide. However, most traditional symmetric DC-DC converters fail to take the capacity difference between PV and battery into consideration. It results in extra component cost for converters in PV energy storage system. This paper presents a multiport current-fed asymmetric bidirectional DC-DC converter (CF-ABC) to handle the capacity difference. The converter consists of a full-controlled bridge in primary side and two semi-controlled bridges in parallel in secondary side under forward operation mode. Under reverse operation mode, it works as a dual active bridge (DAB). The working principles and zero voltage switching (ZVS) conditions under both operation modes are analyzed and presented in this paper. Compared to traditional bidirectional DC-DC converter, the proposed CF-ABC can reduce the total cost of the converter. The effectiveness of the proposed CF-ABC is verified through experiments.
5:15 Sensor-Driven Solar Power Forecasting Using No-Code LSTM in KNIME: A Scalable Deep Learning Framework
Lakshmi Boppana (National Institute of Technology Warangal, India); Raghuram Kornepati (RVR & JC College of Engineering, India)
As a predominant renewable energy source, solar energy has gained a significant prominence due to its sustainable characteristics and environmental benefits. Precise forecasting of photovoltaic power generation constitutes a critical requirement for optimal energy management and the maintenance of stability in electrical grid operations. This work proposes a novel predictive modeling framework that implements long-short-term memory (LSTM) neural networks through the KNIME Analytics Platform to predict solar energy production utilizing high-resolution sensor data. The comprehensive data set incorporates time series measurements of key solar parameters, including solar irradiance, global horizontal irradiance (GHI) power density, plane-of-array (POA) energy yield, and module temperature, acquired from an industrial-grade solar monitoring infrastructure. Data preprocessing techniques and visualizations were used to improve model performance and interpretability. The LSTM model effectively captured temporal dependencies in the data, with evaluation metrics including mean absolute error (MAE), mean squared error (MSE), and root mean squared error (RMSE). The implemented LSTM architecture successfully learned complex temporal patterns inherent in the solar generation data, with rigorous performance validation conducted using standard evaluation metrics: mean absolute error (MAE), mean squared error (MSE) and root mean squared error (RMSE). The results demonstrate that a well-optimized LSTM can effectively model long-term solar patterns without the computational overhead of hybrid architectures while maintaining competitive accuracy, making it more scalable and deployable for real-world solar energy forecasting. This contribution addresses a fundamental research gap in the field by providing a simplified yet highly accurate forecasting solution, particularly beneficial for grid operators and renewable energy planners seeking efficient and reliable predictions.
5:30 Early Detection of Internal Short Circuit in Lithium Ion Battery Using Time-Frequency Analysis
Anup Appasaheb Kshirsagar and Jithendra Mani Kumar Kommuru (Chang Gung University, Taiwan); Cher Ming Tan (Chang Gung University & Center for Reliability Sciences and Technologies, Taiwan); Shuen-Lin Jeng (National Cheng Kung University, Taiwan)
Early identification of internal short circuits (ISCs) in lithium-ion batteries (LiBs) is critical for ensuring system-level safety and operational reliability in applications such as electric vehicles and grid-scale energy storage. Conventional diagnostic techniques-such as resistance monitoring, voltage deviation analysis, and statistical health modeling-typically detect faults only after substantial degradation has occurred, often overlooking transient electrochemical anomalies that precede ISC onset. To address this diagnostic limitation, the present study introduces a high-resolution, voltage-only analytical framework employing the Smoothed Pseudo-Wigner-Ville Distribution (SPWVD) for time-frequency spectral decomposition of discharge voltage signals. Experimental results demonstrate that the SPWVD technique can detect early ISC-related spectral changes as much as 96 seconds before the appearance of conventional ISC indicators. These findings highlight SPWVD's effectiveness in revealing subtle, early-stage anomalies associated with ISC development. The proposed framework contributes a low-intrusion, high-sensitivity strategy for real-time prognostic health management of LiBs. By enabling predictive fault detection, it offers significant potential to enhance battery safety protocols and supports in the development of advanced diagnostic tools for early warning and preventive control in energy storage systems.
5:45 Power Quality and Condition Monitoring of Inverter Duty Transformers at Grid-Connected Solar Photovoltaic Plants in India
M V Chilukuri (VIT University, Vellore, India); Kavita Sao (VIT University & Vellore Institute Technology, India)
A key area of concern is the performance of grid-connected solar photovoltaic (SPV) plants, particularly in maintaining power quality and operational reliability. As of 23 May 2025, India had installed 107.95 GW of solar PV capacity, much of which was through grid-connected solar photovoltaic (SPV) plants incorporating thousands of Inverter Duty Transformers (IDTs). However, a number of these IDTs have experienced failures during commissioning and operation, despite compliance with existing national and international standards. Currently, there is limited publicly available data on IDT failures at both the national and global levels. To address this gap, a technical study was carried out involving (a) surveys of SPV plants and (b) Monitoring of power quality and condition at selected sites. The aim is to gain a deeper understanding of the nature of these failures, particularly those involving IDTs, and to identify the root causes through modeling and analysis. Furthermore, CIGRE has recently established Working Group A2.68 to investigate these issues globally and publish its findings and recommendations. These insights will help national committees to revise standards and implement best practices. This paper discusses the challenges and opportunities identified during the SPV plant survey, as well as the results of monitoring power quality and conducting condition assessments. Highlighting the importance of power quality and condition monitoring in the smart grid for power supply reliability and quality for critical infrastructure.
6:00 Dynamic Solar Energy Optimization: Implementing DiTO-Based MPPT Control for Grid-Connected PV System
Praveen Kumar Balachandran (Vardhaman College of Engineering, India); Muhammad Ammirrul Atiqi Mohd Zainuri (Universiti Kebangsaan Malaysia, Malaysia)
Presently, solar PV systems are considered as an efficient and sustainable technology because of the ability by the systems to convert sunlight into electricity. But, due to the unpredictability of solar irradiation, there is a requirement for complex control methods to enhance the effectiveness of these systems. However, one of these mechanisms takes a crucial role to guarantee the maximum operating electrical power from the solar PV system according to the environmental conditions, which is called Maximum Power Point Tracking (MPPT). However, most of the previously mentioned techniques are still confronted by issues like slow response to sudden variations in irradiance, high computation, and higher implication costs. These are the discrepancies that this paper is set to eliminate by adopting a newly introduced Dipper Throated Optimization (DiTO)-based MPPT control strategy. The DiTO-MPPT control approach is therefore a dynamic real-time method that tries to maximize the power extracted from the PV system by changing its functioning parameters to settle for the MPP. The combination of the DiTO algorithm with the MPPT control is indeed a novel approach that promises to enhance the power output and contour the reliability of the SPV systems that are connected to the grid. Through experimental outcome and simulation it has been established that the proposed method is effective in achieving maximum energy conversion efficiency as compared to the existence approach used in solar PV systems.
Electronics, Circuits & Devices (ECD)
Track A3F4 ECD 3: Electronics, Circuits & Devices (ECD) 3
Room: F4. 503 Dinawan (Level 5)
Chair: Mazlina Mamat (Universiti Malaysia Sabah, Malaysia)
4:30 Guidelines and Logistics for Manufacturing RISC-V Vanilla Silicon Chips Using SkyWater 130nm OpenPDK
Anand K Vinu (Cochin University of Science and Technology, India); Sahil Athrij (NVIDIA, USA); Asna V A (Cochin University of SCience and Technology, India)
The design and fabrication of Reduced Instruction Set Computer - Five (RISC-V) based silicon chips using the SkyWater 130nm Open Process Design Kit (PDK) and the OpenLane toolchain represent a significant step toward democratizing semiconductor development. Leveraging the open-source nature of the RISC-V instruction set architecture (ISA) offers substantial advantages in terms of flexibility, extensibility, and cost reduction. This open approach enables designers to fully customize their hardware without the licensing constraints typically associated with proprietary ISAs, making it an ideal candidate for academic, research, and low-volume commercial applications.
The design process begins with a clear definition of the desired functionality of the chip, followed by the development of Register Transfer Level (RTL) code implementing the five-stage pipelined RISC-V core. This includes modular components for the instruction fetch, decode, execute, memory access, and write-back stages. The RTL code is validated through extensive simulations using tools such as Verilator and ModelSim to ensure functional correctness under a variety of test scenarios. GTKWave is used to analyze simulation waveforms, offering visibility into signal transitions and pipeline behavior, and aiding in the detection and resolution of subtle design issues.
Synthesis is performed using Yosys as part of the OpenLane flow, which translates the RTL code into a gate-level netlist. The structured and modular design of the pipelined core allowed for relatively smooth synthesis, with only minor issues such as unconnected ports or redundant signals, which were resolved with small code refinements. Once synthesized, the design undergoes floorplanning, placement, clock tree synthesis, and routing using tools integrated into the OpenLane environment. Final checks and verification steps ensure design rule compliance and functionality before the layout is exported as a Graphic Data System II (GDSII) file, ready for fabrication.
Despite the challenges commonly associated with open-source flows-particularly in handling complex Verilog constructs-the project demonstrates that with careful design practices, efficient RTL structuring, and iterative verification, it is entirely feasible to produce manufacturable RISC-V silicon on the 130nm technology node. This work underscores the viability of using open-source tools and platforms for custom chip development and contributes to broader efforts aimed at making silicon design more accessible and innovative.
Index Terms- RISC-V, GDSII, SkyWater, OpenPDK, OpenLane, RTL Design, Physical Design
4:45 Advanced Three-Axis Shake Table System for Comprehensive Earthquake Simulation and Structural Dynamics Research
Pasindu A. Iddamalgoda and Dulan Sashik (Sri Lanka Institute of Information Technology, Sri Lanka); Jayakody Arachchilage Don Chaminda Anuradha Jayakody (Curtin University Technology, Sri Lanka & Sri Lanka Institute of Information Technology, Sri Lanka); Nalin Manchanayake (LinkNlabs, Sri Lanka); Raj Prasanna (Massey University, New Zealand); Pradeep Abeygunawardhana (Sri Lanka Institute of Information Technology, Sri Lanka)
The development of a three-axis shake table is essential for advancing research in sensor validation, earth- quake engineering, and structural dynamics. Traditional shake tables that exist provide one or two-dimensional motion where it limits the capability of three-dimensional (3D) movements in experiments conducted by engineers, researchers and users. This paper presents the design and development of a cost- effective 3-axis shake table which has the ability to produce independent motions in the X, Y and Z directions. The shake table consists of three dynamic plates, each axis supported by dual linear guides and driven by separate DC motors with a 12V voltage rating. This independent configuration of the axis allows the users for a precise control and generation of vibrations across all three axes, making it suitable for a wide range of testing applications. The design utilizes motor drivers for varying power to the motors and control the movement of the dynamic plates independently, which omits the requirement of encoders and reduce the production cost. The developed shake table offers a versatile and cost-effective solution for laboratories, earthquake engineering and structural health monitoring, where 3D motion capabilities are essential. Index Terms-Structural Health Monitoring (SHM), Earth Quake Simulation, Structural Dynamics, Vibration Analysis, Three Dimensional Motion
5:00 Cryptographically Secure Random Number Generator Utilizing Environmental Radiation
Tee Hui Teo (Singapore University of Technology and Design, Singapore); Maoyang Xiang (8 Somapah Rd & Singapore University Technology and Desgin, Singapore); Qianrui Lin, YiKai Zhang, Ziyue Pan, Au Hern Ng, Khant Htet Kaung and Kevin Chua (Singapore University of Technology and Design, Singapore); Hei Lam Shum (Auckland University of Technology, New Zealand)
This paper proposes a cost-effective Cryptographically Secure Random Number Generator (CSRNG) utilizing a Field-Programmable Gate Array (FPGA) integrated with a Geiger counter. The front end of this system is designed to capture ambient radioactivity through a Geiger counter system that emits pulses in response to such environmental stimuli. To complement this setup, a neoTRNG-based ring oscillator RNG is incorporated to enhance the randomness of the generated numbers. By combining these elements, the CSRNG can achieve a high level of unpredictability crucial for cryptographic applications. Moving to the system's backend, utilizing the BLAKE2s hash function is pivotal in counteracting any potential biases introduced by the frontend components. This hashing function ensures that the output random numbers remain robust and secure, free from discernible patterns or vulnerabilities. In essence, the methodology outlined in this paper offers a comprehensive guide to conceptualizing and implementing a cost-effective CSRNG. The challenges inherent in generating cryptographically secure random numbers can be effectively navigated through a meticulous integration of hardware components and cryptographic techniques. This approach sheds light on the intricate processes in creating a reliable source of randomness, essential for safeguarding sensitive information in various digital systems.
5:15 Fingerprint Tarot Fortune Teller Game Utilizing Hénon Map-Based Pseudorandom Number Generator
Tee Hui Teo (Singapore University of Technology and Design, Singapore); Maoyang Xiang (8 Somapah Rd & Singapore University Technology and Desgin, Singapore); YiKai Zhang, Junhan Li, Keith Zhengxian Lee, Wyndham Tian, Yi Sun and Yew Rei Leow (Singapore University of Technology and Design, Singapore); Miranda Chen (University of Waterloo, Canada)
The project is about integrating biometric security, hardware-based randomness, and symbolic visualization through tarot for a unique user experience. It utilizes a field-programmable gate array (FPGA) for biometric authentication using fingerprint input to enhance security. Sensitive data is encrypted within the FPGA, ensuring tamper resistance and mitigating threats such as replay attacks. The system utilizes the entropy from the fingerprint as a seed for a pseudorandom number generator (PRNG) to select a tarot card displayed on a Raspberry Pi and a narrative. The project explores the fusion of digital security and human meaning by combining secure biometrics, hardware-accelerated cryptography, and symbolic storytelling. It aims to provide personalized authentication, interactive installations, and secure entertainment interfaces for users. A custom enclosure CAD design was developed for the FPGA-integrated biometric system to improve user-friendliness. The design focused on touch-based interaction using a touchscreen and fingerprint scanner to simplify the user interface and enhance user enjoyment. This approach allowed the team to concentrate on perfecting the PRNG code rather than dealing with moving parts and manual updates for user instructions. The Hénon Map-based PRNG is implemented in the FPGA for real-time applications.
5:30 Entropy-Rich One-Time Password Generation Utilizing Sensors in a Hardware-Realized Chaotic Chua's Circuit
Tee Hui Teo (Singapore University of Technology and Design, Singapore); Maoyang Xiang (8 Somapah Rd & Singapore University Technology and Desgin, Singapore); Zhengyao He, Qianrui Lin, RK Suriya Varshan, Yee Kiat Lim, Jing Ting Leow and Ahmad Danish Bin Azli (Singapore University of Technology and Design, Singapore); Matthew Wong (University of Waterloo, Canada)
This paper introduces a novel hardware-based solution for generating one-time passwords (OTPs) using a field-programmable gate array (FPGA). By leveraging real-world analog noise sources like light, temperature, and sound sensors, the system ensures a high level of entropy to seed the random number generation process in a dedicated FPGA chaotic Chua's circuit. The design of this OTP generator is capable of producing secure 5-digit OTPs ranging from 00000 to 99999. These OTPs can serve various purposes, such as wireless applications when transmitted to an ESP32 microcontroller or authentication in access control systems. By integrating these OTPs directly into access control systems, organizations can enhance their security measures significantly. This integration allows for seamless and secure authentication processes, ensuring that only authorized individuals gain access to restricted areas. The proposed approach prioritizes high randomness and resistance to prediction, essential characteristics for secure embedded systems. By incorporating multiple noise sources and utilizing FPGA technology, the OTP generator guarantees a robust level of security. Overall, the hardware-based OTP generator presented in this paper stands as a reliable and innovative solution for enhancing security in embedded systems.
5:45 An Electrical Behavior Analysis Method of ESD Protection Structure in Chip Based on SPICE
Siyuan Shen, Xiangfen Wang, Bo Wan and Guicui Fu (Beihang University, China)
With the continued scaling of integrated circuits (ICs), strong electrostatic-discharge (ESD) protection is increasingly essential for device reliability. This paper presents a method for analyzing the behavior of SCR-based protection structures using SPICE. The workflow is driven by Transmission Line Pulse (TLP) measurements, which are used to characterize the electrical behavior and to set up the simulations. Key metrics such as triggering voltage, holding voltage, and snap-back are examined through both experiments and SPICE. The Silicon-Controlled Rectifier (SCR) device is used as the representative protection structure, known for its reliable and efficient performance. The proposed methodology provides a practical framework for simulation and experimental verification, supporting the design and optimization of ESD protection in ICs. In particular, hybrid SPICE models are shown to capture the overall protection performance of SCR structures. The analysis also offers a systematic process for comparing and balancing alternative ESD solutions, providing a reliable basis for optimization in semiconductor device and packaging design.
Communication Systems (CS)
Track A3F5 CS 3: Communication Systems (CS) 3
Room: F5. 504 Madai (Level 5)
Chair: Azwan Mahmud (Multimedia University & Telekom Malaysia, Malaysia)
4:30 VAE-BiLSTM-IDS: A Two-Phase Deep-Learning Framework for Enhanced IoT Security
Siddhant Gond (Indian Institute of Information Technology, Guwahati, India); Bishal Chhetry, Rajdeep Kumar Dutta, Rakesh Matam and Ferdous Barbhuiya (Indian Institute of Information Technology Guwahati, India)
The widespread adoption of Internet of Things (IoT) devices has transformed industries such as healthcare, manufacturing, and smart cities. However, these devices often possess limited resources and weak security mechanisms, making them vulnerable to cyberattacks. Traditional Intrusion Detection Systems rely on known attack signatures and are ineffective against novel or unknown threats. Although recent machine learning (ML) approaches aim to detect anomalous activity, many continue to suffer from high false alarm rates and degraded performance under dynamic network conditions. To address these challenges, we propose a two-phase deep learning framework, VAE-BiLSTM-IDS, designed specifically for IoT networks. In the first phase, Variational Autoencoders (VAEs) learn normal traffic patterns and detect anomalies using adaptive thresholds that adjust to changing network behavior. In the second phase, a CNN-BiLSTM model leverages both spatial and temporal features to classify anomalies and identify specific attack types. Evaluated on the Edge-IIoT dataset, which contains realistic IoT traffic and zero-day attacks, our framework yields a detection accuracy of 98.89%. It significantly reduces false positives compared to state-of-the-art methods. This approach offers a robust and adaptive solution for improving IoT security.
4:45 Game-Theoretic Optimal Channel Allocation for LoRaWAN in Dynamic IoT Environments
Soham Raju Kadtan and Biraja Nanda Mohanty (BITS Pilani, India); Alekhya Gorrela, Anakhi Hazarika and Nikumani Choudhury (BITS Pilani Hyderabad Campus, India); Dipamani Choudhury (Royal Global University, India); Syed Mohammad Zafaruddin (BITS Pilani, India)
Low-power wide-area networks (LPWAN) have substantially improved the Internet of Things (IoT). LoRaWAN is a potential technology for IoT applications because it uses low-power, long-distance communication and offers excellent availability with low energy consumption. LoRaWAN power consumption can be reduced using the pure Aloha protocol at the MAC level. Optimizing orthogonal transmission parameters is still a major difficulty for enhancing network performance, even though they reduce packet loss and prevent collisions, especially in dynamic and heterogeneous networks. However, the challenge of random channel selection in LoRaWAN communication often leads to inefficient resource utilization and degraded network performance. This paper proposes a novel game-theoretic approach for optimal channel selection in LoRaWAN networks. Our method leverages real-time Received Signal Strength Indicator (RSSI) data and a non-cooperative game theory model to dynamically select channels, thereby improving throughput and reducing packet loss. Through extensive simulations and a real-world testbed, we demonstrate that our proposed mechanism outperforms existing approaches such as the Online Decision algorithm and MFMSF. Specifically, it achieves up to 22% improvement in throughput, 15% higher packet delivery ratio, and 18% reduction in latency, while consuming up to 25% less energy under heavy and dynamic traffic conditions. This work offers a significant advancement in enhancing the scalability and reliability of LoRaWAN networks, paving the way for more efficient IoT communications.
5:00 An Ensemble Learning Approach for Malicious Traffic Detection System for Computer Networks
Jay Fel Quijano (Mapua University, Philippines & Nexus Technologies, Inc., Philippines); Ramon Garcia (Mapua Institute of Technology, Philippines)
Threats in the digital landscape increased as different emerging technologies progress. Despite the various implementations of different defensive measures and tools to monitor and detect traffic anomaly, complexity and cost are the main challenges for small to medium enterprise networks. This study developed a malicious traffic detection system that utilized signature-based and anomaly-based techniques through packet analysis and network flow using machine learning algorithms. The system consists of a Raspberry Pi-based packet capturing tool, website application, implored with Ensemble Learning Model to detect and classify malicious network traffic. The ensemble learning model which includes weak learners like Decision Tree, Naïve Bayes and Support Vector Machine (SVM), in which individual results are combined and boosted using XG Boost algorithm. CTU-13 and CSE-CIC-IDS2018 datasets were used for training and validating the model. The ensemble learning model achieved an accuracy of 96.42%, precision of 98.99%, recall of 93.79%, and F1-score of 96.32% for the binary classification using the validation data. Moreover, the model also achieved an accuracy of 96.39%, precision of 95.82%, recall of 96.39%, and F1-score of 95.66% for the multiple classification. The ensemble machine learning model was also evaluated using generated traffic, resulting in a decline in performance: accuracy dropped to 60.94%, precision to 44.44%, recall to 30.77%, and the F1-score to 61.54%.
5:15 Age Optimal Scheduling for a Linear Multiflow Network with Transmission Constraints
Teena Mary Treesa (IIITDM Kancheepuram, India); Premkumar Karumbu (Indian Institute of Information Technology Design and Manufacturing Kancheepuram, India)
In this Paper, we consider a scheduling problem in a network with three nodes and two flows using an Age of Information metric. The paths of the flows are different, and hence, the flows affect the scheduling metrics differently. Each node is equipped with a one-buffer for each flow that passes through it, and the network has general interference constraints. For stochastic packet arrivals, we pose a problem with constraints on energy at each node that aims to minimise the Expected Weighted Sum Age of Information. We formulate the problem as a Constrained Markov Decision Process. The problem can be solved by a linear program, which results in the optimal policy. We also propose a near-optimal heuristic policy and a low-complexity Drift-Plus-Penalty based scheduling policy for the problem. The proposed policies are shown to have a near-optimal average age performance. The proposed heuristic and Drift-Plus-Penalty based scheduling policies have a complexity that increases linearly with the number of feasible scheduling actions.
5:30 ML Based Traffic and Packet Size Prediction for Scalable DBA Algorithms in 50G PONs
Shobhit Khurana and Ayushman Rathor (Birla Institute of Technology and Science Pilani, India); Malhaar Goswami, Nishit Prabhakar Shetty and Sukriti Garg (BITS Pilani, India)
With increasing demand for low-latency and energy-efficient communications in Passive Optical Networks (PONs), classical dynamic bandwidth allocation (DBA) algorithms have found it challenging to accommodate the dynamic nature of traffic flows. The study explores the trend of using machine learning (ML) models, namely, convolutional neural networks (CNNs) and extreme gradient boosting (XGBoost), for timestamp and packet size prediction that would allow for an intelligent DBA. The CNN models that we propose in this work are effective at predicting the packet inter-arrival time, with low mean absolute errors across the different periodic traffic scenarios. Additionally, we propose using XGBoost, an ensemble-based gradient boosting method, to predict packet size and bandwidth demand for large users and a network capacity of 50 Gbps. We can produce an effective root mean square error (RMSE) exhibiting strong predictive power at low computational cost. Our comparative analysis also reveals that XGBoost outperforms conventional DBAs and deep learning (DL) models such as long- and short-term memories (LSTMs) and artificial neural networks (ANN) from the perspective of reduction in delay and training time efficiency. This hybrid architecture allows the implementation of DBA algorithms driven by scalable predictions that achieve delay reduction of up to 10% upstream, a considerable extent that we can consider for next-generation optical access networks.
5:45 Optimal Active/Passive Beamforming and Positioning of RIS for Ship-Shore Communications
Deepthi M (Signal Processing and Communication, India); Poornima S (Cochin University of Science and Technology, India); Deepak S (Signal Processing and Machine Learning, India)
RIS offers a cost-effective solution to overcome the coverage issues of using next-generation communication systems. UAVs with mounted RIS can flexibly provide reliable connectivity to remote areas. In this paper, an RIS-mounted UAV is employed to improve the reliability of ship-to-shore communication. The proposed work aims to jointly optimize the beamforming vector of the transmit antenna (active beamforming), the phase shifts of RIS unit elements (passive beamforming) and the placement of an RIS-mounted UAV to maximize the receiver's capacity. The exact joint optimization for active/passive beamforming and positioning of RIS-mounted UAV is obtained by using Maximum Ratio Transmission (MRT) and Particle Swarm Optimization (PSO). An approximate joint optimization, which maximizes an upper bound of the SNR, is also solved using MRT, semi-definite programming (SDP) with a closed-form solution for optimum position. It is found that when the distance between BS and user is equal to twice the height of UAV (D = 2h), RIS should be placed vertically at the mid-point; for D >> 2h the RIS may be placed directly above the BS or the user. For lower distances, RIS placed vertically at the mid-point, at a fixed UAV-safe height, is best. The two distance optimized joint optimizations show superior performance compared to a benchmark scheme [3] and the case of random beamforming.
Computing & Computational Intelligence (CCI) 1
Track A3F6 CCI 3.1: Computing & Computational Intelligence (CCI) 3.1
Room: F6. 505 Sepilok (Level 5)
Chair: Yu Beng Leau (Universiti Malaysia Sabah, Malaysia)
4:30 A Hybrid Technique for Object Recognition
Rakesh Gangula (SRM University-AP, India); Rituparna Choudhury (International Institute of Information Technology Bangalore, India)
Object detection from a given image is a very important application for drones. The high accuracy in correctly recognizing the objects in an image is also paramount in this crucial application. So, in this paper, an efficient algorithm for the accurate detection of 7 different objects in a frame is proposed. In this model, a hybrid of Convolutional Neural Network (CNN) and Long-Short Term Memory (LSTM) is used for detection. The CNN is used for feature extraction and LSTM is used to perform classification. The hybrid use of two deep learning models integrates the strength of both the deep learning models to achieve high F1-score for all 7 classes. This model is proven to achieve around 96% accuracy which is observed to be much higher than the accuracy obtained by the existing models for the proposed application. This model is also found to perform better than other existing machine learning algorithms tested on the same application.
4:45 Detecting Adversarial Attacks on HiDDeN Watermarked Images
Huan Lin Oh (Nanyang Technological University, Singapore)
Adversarial attacks pose a significant threat to deep learning systems by introducing imperceptible perturbations that cause models to make incorrect predictions. This paper proposes a method to detect such attacks in watermarked images using transfer learning with a pre-trained convolutional neural network (CNN). Instead of focusing on message recovery, the approach detects tampering by analyzing visual distortions in images watermarked using the HiDDeN framework. A ResNet-based classifier was trained on watermarked images to distinguish clean from adversarial inputs. Projected Gradient Descent (PGD), a strong white-box attack, was used to generate adversarial examples. Experimental results show that such perturbations introduce detectable patterns, enabling the classifier to reliably differentiate between clean and tampered images. While the detection model was primarily evaluated on PGD attacks, training with additional examples generated using the Fast Gradient Sign Method (FGSM) improved generalization to weaker perturbations. This work demonstrates a promising direction for integrating neural watermarking with adversarial detection to strengthen the robustness of image-based deep learning systems.
5:00 Lightweight Deep Learning Models for Classification and Adulteration Detection of Philippine Rice Varieties
Joesmart Apan, Jose Miguel D. Domingo, Melvin K. Cabatuan and Edwin Sybingco (De La Salle University, Philippines)
Rice is an essential food source in the Philippines, yet ensuring its quality remains challenging due to the visual similarities among rice varieties. These similarities often lead to mislabeling and adulteration, undermining consumer confidence and affecting market integrity. This study presents a deep-learning approach to automated rice variety classification and adulteration detection. A custom image dataset containing 2,400 samples was acquired, comprising four classes: Dinorado, Malagkit, Sinandomeng, and adulterated rice. Three models were evaluated - DenseNet121, EfficientNetV2S, and Mobile Vision Transformer (MobileViT). EfficientNetV2S achieved the highest performance with a test accuracy of 100%, demonstrating a superior balance of accuracy, training time, and inference speed. It was found to be 45.69% as fast in inference as DenseNet121, which achieved a 99.5% test accuracy. The lightweight MobileViT model, while indicating underfitting, provided a test accuracy of 92% and the fastest inference time, proving to be at least 31% faster than EfficientNetV2S. The results highlight EfficientNetV2S as the most effective and reliable solution for accurate rice variety identification, while also indicating the significant potential of lightweight models like MobileViT for future real-time applications in resource-constrained environments with additional optimization.
5:15 Classification of Arrhythmia by Adopting the Hybrid Convolutional Neural Network and the Long Short-Term Memory
Adam Mohd Khairuddin and Siti Armiza Mohd Aris (Universiti Teknologi Malaysia, Malaysia); Ku Nurul Fazira Ku Azir (Universiti Malaysia Perlis, Malaysia); Noor Jannah Zakaria (Universiti Teknologi Malaysia, Malaysia)
In this research, the hybrid convolutional neural network (CNN) and the long short-term memory (LSTM) algorithms were adopted to classify five types of ECG arrhythmia: (1) normal (N); (2) ventricular ectopic (V); (3) supraventricular ectopic (S); (4) fusion (F); and (5) unknown (Q). The framework consisted of the following three main phases: (1) pre-processing; (2) feature extraction; and (3) classification was used to develop the hybrid classification model. The proposed hybrid CNN-LSTM model was evaluated by using the MIT-BIH arrhythmia database that contains 48 recordings of ECG signals. Random under-sampling was utilized to address the imbalanced classes in the database. The experimental results of the study achieved precision of 98.00%, recall of 98.00%, F1-score of 98.00%, and accuracy of 98.00%. Comparisons with prior studies revealed that the proposed hybrid CNN-LSTM model was able to attain comparable performance. However, challenges in interpreting the hybrid model as well as its generalizability to diverse dataset remains.
5:30 Smart Audio Surveillance System for Real-Time Violence Detection and Alarm Response Using Long Short-Term Memory
Robert G. de Luna (Polytechnic University of the Philippines, Philippines & University of Sto. Tomas, Philippines); Khristel B Biscocho (Polytechnic University of the Philippines, Philippines); Charles Adriel A. Del Rosario (Polytechnic University of the Philippines - Sto. Tomas Campus, Philippines); Mary Mizzy Clare O Elipse (Polytechnic University of the Philippines & PUP -STC, Philippines); Piolo O. Pecho (Polytechnic University of the Philippines, Philippines); Sophia Noviel O. Silvestre (Polytechnic University of the Philippines - Sto. Tomas Campus, Philippines)
Public safety has become the main priority nowadays, and one major area of research interest is violence detection. This study shows the development of a deep learning-based monitoring system for real-time violence detection. The audio classification employs Recurrent Neural Network (RNN), particularly Long Short-term Memory (LSTM), having a dataset of 13,141 audio data gathered using a microphone. The methodology incorporates Data Augmentation, Feature extraction using Mel-frequency Cepstral Coefficients (MFCCS), and Hyperparameter Tuning using Optuna and Hyperband pruning. Evaluation metrics including accuracy, precision score, F1 score, and recall were used to assess the model. The hardware implementation utilizes Jetson Nano, along with a microphone and alarm system for violence detection. The Recurrent Neural Network exhibited the best model, achieving outstanding accuracy of 99.85%. This study depicts the potential of a deep learning-based device in enhancing public security, however, further improvement requires recognizing limitations such as the need for diverse datasets. The study contributes to the increasing potential of AI-powered security systems that offer future advancements in audio-based violence detection.
5:45 Assessing the Impact of Atmospheric CO2 Concentrations on Rainfall Patterns
Rangani Anjana Wijesinghe and Deepika Suranjini Silva (Sri Lanka Institute of Information Technology, Sri Lanka)
This study examines how atmospheric levels of CO2, along with factors such as temperature, humidity, wind, and pressure, influence rainfall patterns in Colombo over 17 months. Using data from the National Building Research Organization and the Sri Lanka Meteorological Department, three machine learning models, such as Random Forest, XGBoost, and LSTM, were tested to predict rainfall. Among them, Random Forest delivered the most accurate results. The inclusion of CO2 data significantly improved the performance of the model. With CO2, the Random Forest model showed lower error rates and a higher R2 value, indicating more accurate predictions. Specifically, R2 improved from 0.886 to 0.921, and with further tuning, reached 0.998. XGBoost exhibited improvements, with R2 rising from 0.570 to 0.657 when CO2 was included, while LSTM saw a modest but consistent gain from 0.266 to 0.346. Even in predictive tests, the R2 increased from 0.558 to 0.660 when CO2 was considered. These results highlight the importance of incorporating CO2 data into rainfall prediction models, especially in regions like Sri Lanka, where it's not yet commonly used. Beyond academic insight, the findings have real-world implications for sectors such as agriculture, aviation, and fisheries, where accurate weather forecasts are crucial and can support more informed climate adaptation strategies.
Engineering Technologies & Society (ETS)
Track A3F7 ETS 3: Engineering Technologies & Society (ETS) 3
Room: F7. 506 Selingan (Level 5)
Chair: Nur Ashida Salim (Universiti Teknologi MARA, Malaysia)
4:30 Automated Lung Lesion Segmentation with Minimal Number of Labeled CT Images
Ishita Maiti, Namrata Kadasi, Nihar Domala and Aman Soni (Indian Institute of Technology Kharagpur, India); Manjunatha Mahadevappa (Old NCC Building & Indian Institute of Technology Kharagpur, India); Nirmalya Ghosh (Indian Institute of Technology Kharagpur, India)
An arduous task in medical image analysis is the automated segmentation of infected regions in a organ (for example, in lungs, segmentation of lung lesion), which is crucial for assessing the volumetric severity of infections and evaluating the effectiveness of subsequent treatments. The affected regions in diseased lung cases, called lung lesions, vary widely in their intensity, texture, contrast, location, and size. The similarity of intensities between the lesion and the healthy surrounding tissue bordering the lung makes lung as well as lesion segmentation more challenging, especially by classical image processing techniques. One solution is machine learning (ML) or deep learning (DL)-based methods, that learn segmentation models from datasets with lung lesion labeled by experts. Unfortunately, acquiring any labeled data in the healthcare sector for ML/DL algorithms is prohibitively costly due to the participation of skilled clinicians. Hence, the current study proposes a Random Forest (RF) based ML-method that utilizes a significantly reduced number of expert-labeled data for lung lesion segmentation. Eight different experiments with varying amounts of labeled data from two challenging benchmark datasets demonstrated encouraging results, comparable to state-of-the-art methods, reaching the highest precision values of 86.4%, recall value of 78.1%, and F1 score of 82.0%.
4:45 Dual-Path Enhancement Framework for Masses and Calcifications in Mammograms: a Quantitative Evaluation of Preprocessing Techniques
Janindu Athukorala (University of Sri Jayawardenepura, Sri Lanka); Didulangani Deepashika (University of Sri Jayewardenepura, Sri Lanka); Tirush Wickramasingha and Senal Inovin Fernando (University of Sri Jayawardenepura, Sri Lanka); Uditha L. Wijewardhana (University of Sri Jayewardenepura & Faculty of Engineering, Sri Lanka); Umaya Bhashini Balagalla (University of Sri Jayewardenepura, Sri Lanka)
Breast cancer is a leading cause of mortality among women and early detection using mammograms is important to increase survival rates. However, due to low contrast and noise in mammograms, image enhancement techniques are required to assist in accurate diagnosis of breast cancer. This study proposes a dual-path image enhancement framework evaluated using the publicly available INBreast dataset and compared against existing mammogram enhancement methods. The proposed mass enhancement pipeline is compared against Contrast Limited Adaptive Histogram Equalization (CLAHE), Haze Reduced Adaptive Technique (HRAT) and Magma colour mapping. The proposed calcification enhancement pipeline is compared against Gaussian and Laplacian filtering. The proposed approach achieved a 72% improvement in Contrast-to-Noise ratio (CNR) and nearly a 100% increase in Structural Similarity Index (SSIM) over CLAHE for enhancement of breast masses. According to the results for enhancement of calcifications, the proposed method obtained an improved peak signal-to-noise ratio (PSNR) by 46% highlighting its potential in assisting the process of accurate diagnosis of breast cancer using mammograms.
5:00 ADASYN Driven Framework for Pothole Detection Using XGBoost
Tarun Kumar and Divya Lohani (DIT University, India); Debopam Acharya (DIT University, Dehradun, India)
Pothole detection is critical for enhancing road safety and enabling proactive maintenance, particularly in developing and densely populated regions. Determining the state of the road surface accurately is crucial for preventing accidents and road infrastructure health monitoring. To address this issue, we propose an Adaptive Synthetic Sampling (ADASYN)-driven framework for pothole detection using the Extreme Gradient Boosting (XGBoost) technique. This smartphone-based sensing system leverages Inertial Measurement Unit (IMU) data, comprising accelerometer and gyroscope readings, for the classification of road surfaces. To mitigate class imbalance, the system incorporates the ADASYN technique, which improves the representation of pothole instances. The augmented dataset is processed using the XGBoost algorithm, achieving a classification accuracy of 98%. The proposed work is cost-effective, scalable, and suitable for real-time deployment in intelligent transportation systems. This work also contributes to the United Nations' Sustainable Development Goals (SDGs), particularly SDG 9 (Industry, Innovation, and Infrastructure) and SDG 11 (Sustainable Cities and Communities) by facilitating safer and more efficient transportation infrastructure.
5:15 Text Neck Syndrome: Electromyographic Analysis of Forward Head Posture Induced Cervical Strain
S Saranya and Sharadha Gopal (Sri Sivasubramaniya Nadar College of Engineering, India); Rakshana R (SSN College of Engineering, India); Shravan Kumar Subbaraman (Sri Sivasubramaniya Nadar College of Engineering, Chennai, India)
Detrimental neck postures are unconsciously adopted during prolonged usage of handheld mobile devices (HHMDs). The strain on the neck muscles and spine from this posture can lead to chronic discomfort and, over time, severe musculoskeletal complications. This condition, commonly referred to as Text Neck Syndrome, has become increasingly prevalent in today's digital world. The main purpose of this study is to develop a wearable custom-made device that actively monitors neck posture to help prevent and mitigate the effects of Text Neck. This device integrates an Inertial Measurement Unit (IMU) to detect head positioning and acts as a monitor to remind users if their bad posture remains unchanged for a fixed period of time. These time and position thresholds are validated using an EMG-based data acquisition system. Detailed EMG analysis has been performed on the data collected to track the onset of fatigue and decide the corresponding time threshold. This provides users with real-time feedback on posture, encouraging prompt correction and inculcation of pristine ergonomics.
5:30 Development of an IMU-sEMG System for Hamstring Analysis in Static Movement Protocols
Michael John M Espino, Ken Marco C. Mercado, John Jacen D. Del Mundo, Carlos Miguel S. Estrada, Aaron Sam A. Gilla, Jan Tyrone Cabrera, Reil Vinard S. Espino, Ma. Belinda C. Fidel, Timothy Nazareno, Jazzmine Gale S. Flores, Warren Denzel F. Cheng, Sophia Nicole R. De Leon, Jairo C. Estopace, Renell Arthur A. Kalalang, Augie Louis A. Pador, Ma. Madecheen S Pangaliman, Consuelo B. Gonzalez-Suarez and Jehiel D. Santos (University of Santo Tomas, Philippines)
Existing data acquisition systems for lower-limb assessment typically rely on single-sensor modalities and device-specific protocols, limits accurate measurement of both knee angles and peak muscle activations during exercises like standing leg curls. To address these limitations, we developed a wireless device integrating IMU and sEMG sensors, synchronizing data via timestamps and dynamic time warping. The Kalman filter is used to obtain the IMU signals, while an adaptive filter is used to denoise the sEMG signals. From the combined IMU and sEMG recordings, clear trends emerge in peak leg curl power as a function of knee flexion angle; to visualize these patterns, we applied K-means clustering (k = 3) to knee flexion angle (40°-115°) versus peak leg curl power (0.2-1.0 W) across all repetitions, which revealed three regimes, low (≈ 60°-80°, 0.5-0.8 W), medium (≈ 80°-95°, 0.7-0.9 W), and high (≈ 95°-115°, 0.7-1.0 W), highlighting typical performance zones. Analysis of the resulting scatter plot showed peak muscle activation during concentric phases at flexion angles of 97°-113°. Prototype knee-angle measurements achieved RMSE 7.37°, MAE 4.76°, ICC 0.98, and Spearman's ρ = 0.9765 against ground truth, confirming the system's reliability in data acquisition. Overall, the integrated system provided reliable measurements of muscle activation and knee angles with acceptable error margins and consistency.
5:45 Integrating FDES-DPSIR Framework for Evidence-Based Climate Risk Assessment and Causal Modeling in Indonesia
Muhammad Miftakhul Romadlon (Monash University Indonesia); Miya Irawati and Taufiq Asyhari (Monash University, Indonesia)
Climate change poses complex and interconnected risks that demand structured, data-driven approaches to support effective adaptation and sustainable development. This study introduces a new integrated framework for assessing climate vulnerability in Indonesia by combining the Framework for the Development of Environment Statistics and the Driving Forces-Pressure-State-Impact-Response model. The framework leverages multi-dimensional data from 2014 to 2023, integrating socio-economic statistics and satellite-derived environmental indicators, which were mapped to the DPSIR structure for subsequent analysis. To analyze causal relationships among vulnerability components, Partial Least Squares Structural Equation Modeling (PLS-SEM) is used, revealing significant pathways along the DPSIR sequence. Notably, socio-economic drivers exerted strong direct effects on environmental pressures, while environmental degradation indirectly influenced institutional responses through its impact on climate-related risks. A composite Integrated Climate Vulnerability Index is developed by applying the Entropy Weight Method to quantify spatio-temporal vulnerability patterns across Indonesian provinces. Results show a national decline in vulnerability between 2014 and 2020, followed by a modest rebound post-pandemic, while regional disparities persisted-with eastern provinces such as Papua and Maluku exhibiting consistently high vulnerability. The framework evaluation demonstrates a promising way of integration of structured frameworks with statistical modeling, envisaging evidence-based approaches in supporting sustainable, region-specific climate adaptation and planning.
Computing & Computational Intelligence (CCI) 2
Track A3F8 CCI 3.2: Computing & Computational Intelligence (CCI) 3.2
Room: F8. 507 Monsopiad
Chair: Yin Qing Tan (Universiti Tunku Abdul Rahman, Malaysia)
4:30 AI-Driven EEG Insights into Music Tempo & Mode Impact on Consumer Behavior
Kirsten Lee and Sin Tian Choong (Hwa Chong Institution, Singapore); Aung Aung Phyo Wai (Nanyang Technological University, Singapore)
This study investigates the effects of background music-specifically, tempo and mode- on consumer decision-making during utilitarian online shopping, using electroencephalography (EEG). To address the limitations of traditional self-report methods, which are often affected by social desirability bias, and to fill the gap in neuromarketing research within e-commerce, we conducted a controlled experiment with twelve healthy participants. Subjects performed a series of low- and high-complexity shopping tasks while recording EEG data. Twelve classical music excerpts, categorized into four non-lyrical tempo-mode conditions, served as auditory stimuli. We extracted band-power features from EEG to differentiate cognition states, complemented by repeated-measures ANOVA and post-hoc analyses. Results showed a significant effect of music condition on decision (F(13,11) = 1.94, p = 0.039), with fast-tempo minor-mode conditions significantly enhancing accuracy in low-complexity tasks compared to fast-tempo major-mode conditions (F(2,22) = 9.91, p < 0.001). However, mode differences were not significant under high-complexity scenarios (t(11) = 0.29, p = 0.777). EEG-based frontal alpha asymmetry revealed positive affective states in non-music, fast-major, fast-minor, and select slow-minor conditions, contrasting negative affect in most other scenarios. Interestingly, these neural affect indicators did not correlate strongly with decision accuracy or subjective valence ratings, suggesting music may have subconscious emotional effects independent of overt self-report measures.
4:45 Improving Intersection Traffic Flow Using Deep Q-Learning Algorithms
Abhinav Mishra, Prabu Mohandas, Aparna Vijayan and Anisha N K (National Institute of Technology Calicut, India)
Efficient control strategies play a vital role in mitigating traffic congestion in urban areas. This study introduces an adaptive traffic signal control approach using Deep Q-Learning (DQN) family techniques, such as, the Double DQN variant. The conventional fixed-time control strategies, whose fixed signal timings often cause inefficiency, are usually not suitable, whereas the models in this paper can use real-time traffic data to achieve the regulated phases of signals in an optimal order. DQN improves decision-making by learning optimal signal times, while Double DQN improves further stability and accuracy by rectifying the Q-value overestimation issue by means of decoupled action selection and evaluation. It learns to improve signal timings based on changing traffic patterns. Experiments conducted using the SUMO simulator demonstrate that the proposed framework significantly reduces cumulative delay by 75% and queue length by 78%. The models are considerably better than fixed-timing schemes and traditional Q-Learning methods, and Double DQN shows better adaptability and stability under oscillating traffic conditions.
5:00 Automated Cabin System for Hygiene Compliance Monitoring with Mask, Hairnet, and Handwash Detection
Francis Jann A Alagon and Collien Princess Pepito (Mindanao State University - Iligan Institute of Technology, Philippines); Elaine Krissnell Miral (Mindanao State University-Iligan Institute of Technology, Philippines); Earl Ryan Aleluya (Mindanao State University - Iligan Institute of Technology, Philippines); Cherry Mae G Villame (Mindanao State University-Iligan Institute of Technology, Philippines); Carl John Salaan (Mindanao State University - Iligan Institute of Technology, Philippines); Jeralyn Alagon (Bohol Island State University, Philippines); Shamsudin Abu Ubaidah (Universiti Tun Hussein Onn Malaysia's, Malaysia)
Non-compliance with hygiene protocols-such as wearing a mask or hairnet, and performing proper handwashing-in food manufacturing facilities contributes to food contamination, thereby compromising product quality, consumer trust, and brand integrity. Manual inspection methods used to monitor compliance are susceptible to human error and lack objectivity. Thus, the need for an automated solution is prominent. In response, this study developed a cabin-based system integrated with two YOLOv8-trained models: one for detecting mask and hairnet usage, and another for recognizing handwashing gestures. These models were deployed on a mini-computer (Dell OptiPlex 3080). The compliance system follows the protocol outlined as follows: (i) personnel identification via RFID scanning of the employee card, (ii) detection of mask and hairnet usage through camera input, (iii) sequential detection of handwashing gestures, (iv) regulation of door access to food manufacturing areas based on the evaluation outcome, and (v) recording of compliance results for supervisory review. The system achieved a mean Average Precision (mAP) of 99.2% for mask and hairnet compliance, and 92.7% for handwashing compliance. These experimental results support the system's potential for deployment in food manufacturing settings to facilitate compliance monitoring and reinforce food safety assurance.
5:15 GRU-OptiCom: Revolutionizing Computation Offloading in Edge Computing Through Meta-Reinforcement Learning with GRU
Aditya Oza (IIIT Naya Raipur, India & No, India); Yash Vardhan Gautam, Anirudh Bhakar, Mallikharjuna Rao K and Kanika Malhotra (IIIT Naya Raipur, India)
Modern mobile devices often struggle with limited computational capabilities, hindering their ability to efficiently process data-intensive applications such as augmented reality, mobile healthcare, and intelligent navigation. Multi-access Edge Computing (MEC) presents a viable solution by enabling the offloading of complex computational tasks to geographically proximate edge servers. This offloading approach alleviates the processing burden on user devices and significantly reduces end-to-end latency, thereby enabling real-time responsiveness. In this work, we propose GRU-OptiCom, a novel task offloading framework that leverages meta-reinforcement learning (MRL) to dynamically optimize offloading decisions across varying environments. The model incorporates a GRU-based sequence-to-sequence neural architecture for capturing task dependencies, and employs Proximal Policy Optimization (PPO) to ensure stable and efficient training. We evaluate GRU-OptiCom using latency as the primary performance metric and compare its performance with baseline methods such as MRLCO and HEFT-based greedy algorithms. Experimental results demonstrate that GRU-OptiCom consistently achieves lower latency and improved task distribution, setting a new benchmark for adaptive and intelligent task offloading in MEC environments.
5:30 EEG-Based Classification of Abnormal Epileptiform Patterns
RH Sanjay (Vellore Institute of Technology Vellore, India); Harshita Patel (Vellore Institute of Technology, India); Dharmendra Rajput (VIT, India)
Electroencephalography (EEG) is a widely used non-invasive technique for monitoring brain activity, especially for diagnosing neurological disorders like epilepsy. However, EEG interpretation remains a complex and time-intensive task due to the high variability and noise inherent in the signals. This study proposes a comparative evaluation of two state-of-the-art classification models-CatBoost, a gradient boosting machine learning algorithm, and WaveNet, a deep learning architecture designed for temporal sequence modeling-for automated detection of abnormal epileptiform EEG patterns. Utilizing the publicly available HMS EEG dataset, both models are assessed on classification accuracy and computational efficiency. Results show that WaveNet outperforms CatBoost in classification accuracy, achieving a cross-validation score of 0.81 versus 0.74, though it incurs significantly higher computational cost. Conversely, CatBoost offers faster inference and enhanced model interpretability, making it more suitable for real-time clinical deployment. The study highlights key trade-offs between performance and efficiency, suggesting potential for hybrid approaches that leverage the strengths of both models to advance reliable, explainable, and scalable EEG-based epilepsy diagnosis systems.
5:45 Disciplinary and Transversal Competencies at Play in Work-Integrated Learning
Chien Ching Lee (SIT, Singapore); Ryan Fraser Kirwan (Singapore Institute of Technology, Singapore)
This study explores students' ability to connect the dots in their disciplinary and transversal competencies in an internship report writing workshop held mid-way through their internship. These students have undergone two 3-month internships and are currently participating in their final 8-month internship, as part of a work-study degree programme. The pre- and post-surveys, and teacher's feedback on their writing reflect that the students were able to apply disciplinary competencies effectively and elaborate on these with instructive details. They valued the opportunity to be involved in complex, industry projects and develop their technical skills. Despite their crucial nature, teamwork and communication challenges were elaborated only briefly; with students struggling to express their difficulties in these areas. This work underlines the need for students to be encouraged to share more on their application of transversal competencies to positively impact their long-term employability. Appropriate formats for them to do so could also be examined.