Issues: Hệ thống và Thiết bị thông minh Vol 34.3 (09/2024)
1. Enhance Massive Open Online Courses Integrity: AI for Exam Proctoring

Tuan Linh Dang*, Dinh Minh Vu, Ngoc Dung Pham, The Vu Nguyen, Dinh Phu Mac, Nguyen Minh Nhat Hoang, Huy Hoang Pham

Massive Open Online Courses (MOOCs) are growing quickly, but it's challenging to ensure academic integrity during remote exams with many participants. Existing approaches to supervising students online have scalability, accuracy, and integration limitations. This paper proposes a scalable, accurate AI exam proctoring module compatible with MOOCs to address this issue. Our approach includes an AI server that handles video processing and coordinates cheating detection services. Another server uses Triton to analyze student video feeds quickly. It runs optimized deep learning models, such as face recognition. There is also an integrated MOOC client to capture, compress, and transmit video. The main innovations are the asynchronous AI server for handling multiple tasks simultaneously, efficient deep learning pipelines that use fewer computing resources, and the integration of inference pipelines into Triton for faster processing. The integration of the AI module into the MOOC has been successful. The system can monitor multiple test-takers at the same time and accurately detect any potential cheating. Evaluations showed the high accuracy on different AI models

Article Code:23075

# AI, face recognition, phone detection, face pose estimation, online proctoring, MOOCs platform

Page: 1-8

Field: Information and Communication Technology
2. Advanced Machine Learning and Deep Learning Techniques for Anomaly Detection in Industrial Control System

Le Hai Anh, Tran Le Duc Anh, Hoang Si Hong, Nguyen Thi Hue*

The study extensively examines the evolution of Industrial Control Systems (ICS), with a specific focus on Programmable Logic Controllers (PLC) within critical infrastructure, specifically mixing stations and heat treatment facilities. The research delves into the cybersecurity risks arising from the convergence of PLCs with information technology, transitioning from standalone systems to cloud integration. Noteworthy contributions from industry and academia underscore the pivotal role of machine learning and deep learning techniques in fortifying PLC-based system security. The article rigorously optimizes five classic machine learning algorithms and three deep learning algorithms, achieving an impressive accuracy of over 97%. Additionally, the proposed combined model attains over 99% accuracy on Hardware-In-the-Loop-based Augmented ICS (HAI) and ICS-Flow datasets. The study's importance lies in its thorough analysis of security implications and practical optimization of advanced algorithms, promising effective detection and mitigation of cyber threats in PLC-based ICS environments. These insights offer a compelling perspective for industry and researchers, providing nuanced understanding of cybersecurity dynamics in critical facilities. Optimized algorithms not only demonstrate remarkable threat detection accuracy but also signify a pivotal step in enhancing the cybersecurity resilience of essential infrastructure, serving as indispensable tools against emerging risks.

Article Code:24022

# Machine learning, anomaly detection, ICS, deep learning

Page: 9-16

Field: Electrical Engineering
3. HyperPolypDEQ: A Lightning Transformer-Based Deep Equilibrium Model for Colon Polyp Segmentation

Nguyen Minh Chau, Le Truong Giang, Dinh Viet Sang*

Deep Equilibrium Models (DEQs) have recently emerged as a promising approach to building implicit deep learning models that can achieve on-par accuracy with traditional explicit models but with considerably smaller sizes. However, the significant downside of DEQs is their slow inference speed, primarily due to the time cost of the fixed-point solver. This paper proposes to overcome this issue by applying HyperSolver, a novel technique that replaces traditional fixed-point solvers with a lightweight neural network. This is an extension of our previous work on PolypDeq concerning DEQs for medical image segmentation as an attempt to accelerate our existing implicit models. Experimental results show that our new models using Hyper-Solver can achieve similar results to existing DEQ models on several benchmark medical image datasets while having a significant speedup in inference time (about 9 times). To the best of our knowledge, this is the first attempt to accelerate DEQs for medical image segmentation using HyperSolver, representing a significant step towards making implicit deep learning models more practical for real-world applications.

Article Code:24007

# Semantic segmentation, polyp segmentation, implicit deep learning, deep equilibrium models

Page: 17-26

Field: Information and Communication Technology
4. Effects of the Rotation Phase on the Average Lift of an Insect-Like Flapping Wing

Vu Dan Thanh Le, Anh Tuan Nguyen*, Thanh Dong Pham, Hoang Quan Dinh, Huu The Nguyen, Cong Truong Dinh

Insect-like flapping wings are characterized by multi-degree-of-freedom motions at the wing base, which can be divided into two main movements: sweep and rotation. The phase difference between sweep and rotation motions is an important kinematic parameter that has a great influence on the wing lift. In this paper, the effect of the rotation phase on the average lift of a hawkmoth-like wing is investigated. Simulations were conducted using a Fluid-Structure Interaction co-simulation framework developed based on the multibody dynamics approach and an unsteady vortex-lattice method. The results show that maximum lift for the rigid wing is reached at an advanced phase of about 10%. For the flexible wing, maximum lift is reached at a delayed phase of about 5%. The reason for this difference could be the passive deformation of the flexible wing, which causes an advanced rotation phase at the wing tip. The obtained results are in good agreement with experimental results conducted by previous studies.

Article Code:24020

# Flapping wing, micro air vehicles, unsteady aerodynamics

Page: 27-34

Field: Mechanical Engineering
5. Dynamic Obstacle Avoidance Using Nonlinear Model Predictive Control and Control Barrier Function for Ballbot Systems

Pham Minh Duc, Vu Duc Cuong, Nguyen Thi Thuy Hang, Nguyen Danh Huy , Nguyen Thi Van Anh, Nguyen Tung Lam*

This research presents a tracking control system for a ballbot designed to operate in complex environments filled with both static and dynamic obstacles. The Nonlinear Model Predictive Control (NMPC) framework is formulated to predict the future positions of the ballbot and all surrounding obstacles. This predictive capability is crucial for effective navigation, as it allows the ballbot to anticipate potential collisions in the prediction horizon. The NMPC is integrated with an optimization problem that is enhanced by Control Barrier Function (CBF) constraints. These constraints ensure that the ballbot maintains a safe and consistent distance from every obstacle, thus preventing collisions. Additionally, an Extended State Observer (ESO) is implemented to observe and compensate for uncertain disturbances in the ballbot’s movements, as well as to estimate immeasurable variables that might affect its performance. Various simulation scenarios are conducted to thoroughly test and validate the effectiveness of this approach in achieving precise tracking control and reliable collision avoidance in environments with a large number of obstacles.

Article Code:24010

# Ballbot, control barrier function, model predictive control, obstacles avoidance

Page: 35-42

Field: Engineering Physics
6. Robust Adaptive Path Following Controllers for Autonomous Surface Vehicles with Unknown Disturbances

Quoc Van Tran*, Quang-Hoang Nguyen

This work presents robust path following controllers with disturbance rejection terms for autonomous surface vehicles in the presence of unknown bounded disturbances. The objective is to steer the vehicle to the desired path while its temporal evolution on the path is defined via a path parameter. The disturbance rejection terms are based on sliding mode control utilizing either constant or time-varying gains. To mitigate the chattering effect in the sliding mode controllers, a continuous adaptive control law based on a normalization technique is subsequently developed. Since the control protocols are proposed as control forces based on the nonlinear dynamics of the surface vehicle, Lyapunov stability theory and backstepping control technique are adopted for the control system design and the global stability analysis. Under the proposed controllers, the vehicle is shown to converge to the desired path asymptotically. Simulation results are also provided to support the theoretical analysis.

Article Code:24012

# Autonomous surface vessels, path following, robust adaptive control, sliding mode control

Page: 43-51

Field: Mechanical Engineering
7. Low-Cost Effective Hands-Free Control Devices for Quadriplegia

Son T. Nguyen*, Manh X. Vuong, Tien D. Nguyen

Hands-free control devices are very beneficial for people with quadriplegia. By using these devices, severely disabled people can obtain much more independence and significantly reduce the assistance from relatives. Hands-free devices can be developed based on head movement detection, eye blinking detection, and speech recognition. For many years, these hands-free devices have been very costly and even ineffective for many users. In addition, they cannot adapt to different kinds of users. In this study, low-cost, efficient, hands-free devices have been proposed with the use of inexpensive hardware and software. Firstly, a head-direction-based control system is formed by using an ADXL335 accelerometer and an Arduino Uno board. In this system, intentional head movement can be detected by using a feedforward neural network. An eye-blink-based system can be developed by using a MindWave mobile headset and an Arduino Uno board. Finally, an effective speech recognition system has been developed using an Arduino Nano 33 BLE Sense board with a speech recognition technique that does not require any learning model. Finally, these hands-free control devices are not only effective but also very affordable for various kinds of severely disabled users.

Article Code:24018

# Quadriplegia, hands-free control devices, head movement-based control, eyeblink-based control, speech recognition-based control

Page: 52-59

Field: Electrical Engineering
8. Motion Control of an Electric Power-Assisted Bicycle under the Effects of Operating Conditions

Bui Thu Huyen, Nguyen Ba Hung*

The use of electric bicycle (EB) is considered as a useful solution for reducing the exhaust emissions and dependence of fossil fuels. Along with the development of EBs, studies on their motion characteristics have been receiving more attentions. In this paper, the control of the angular speed of the wheel in an electric power-assisted bicycle (EPAB) is discussed, considering external factors such as slope grade and wind speed. One proposed strategy to optimize vehicle speed is the particle swarm optimization (PSO) algorithm. To achieve this, a simulation model was developed to represent the operation of EPAB under rider control. Based on this operating model, mathematical models including a dynamic model of a bicycle under the driver's control, a dynamic model of an electric motor, and a vehicle speed control model using PSO - based Proportional Integral Derivative (PID) controller are established. The simulation demonstrates that the PSO-based PID controller is superior in terms of control compared to using it without PSO and it works quickly in finding Kp, Ki, Kd to control the angular velocity of the wheel when external conditions change. These simulation results can also serve as useful resources for researchers looking to develop electric-assist bicycles.

Article Code:24027

# EB, PID, PSO, slope grade, wind speed

Page: 60-68

Field: Mechanical Engineering