Advanced Mechanical and Mechatronic Systems

Machine Vision Lab Experiment Based Convolutional Neural Network for Potential Intelligent Manufacturing Applications

Ardian Webi Kirda, Kamil Gatnar, Khairul Muzaka, Nur Arifin Akbar — Mon, 08 Jun 2026 00:00:00 +0000

This study developed a deep learning-based image classification method for the automated assessment of aluminum edge quality after machining processes. The proposed approach classified edge conditions into three categories: Normal Edge, Burr Edge, and Sharp Edge. A Convolutional Neural Network (CNN) based on the VGG16 architecture was employed as the feature extraction backbone, with modifications to the final fully connected layers to accommodate the three-class classification task. The model was trained and evaluated on a dataset of 660 aluminum edge images captured under controlled laboratory conditions at the Robotic Manufacturing Laboratory, University of Brunei Darussalam. The training strategy employed a stratified split with 360 training images, 240 testing images, and 60 validation images. Data augmentation techniques (horizontal flip, rotation ±15°, brightness adjustment) were applied to enhance model generalization. The optimized model achieved an overall classification accuracy of 98.75% on the test set. Precision, recall, and F1-scores for all three classes exceeded 0.97. For practical deployment, the trained model was deployed on an NVIDIA Jetson Nano embedded platform, achieving an average inference time of 47 ms per image at an input resolution of 500×500 pixels (approximately 21 FPS). These results demonstrated the feasibility of real-time edge quality assessment for intelligent manufacturing applications.

A Hybrid Machine Learning Framework for Multi-Limb Human Activity Recognition Using Synchronized Smartphone IMU Sensors: Dataset and Benchmarking

Mon, 20 Apr 2026 00:00:00 +0000

Human Activity Recognition (HAR) using smartphone inertial measurement unit (IMU) sensors has emerged as a transformative technology for health monitoring, fitness tracking, and context-aware computing. However, existing HAR research is constrained by limited data availability, short recording durations, and single-limb sensing configurations. This study addresses these limitations through three principal contributions: (1) introduction of a novel open-access multi-limb HAR dataset featuring synchronized 60-second recordings from hand and ankle positions using tri-axial accelerometer, gyroscope, and magnetometer sensors, publicly available via Mendeley Data repository; (2) systematic benchmarking of classical machine learning classifiers including Random Forest, XGBoost, and Linear Support Vector Classifier under realistic multi-sensor fusion conditions; and (3) comprehensive analysis of model robustness across varying windowing configurations. The dataset comprises recordings from six participants performing six daily activities (walking, stair ascent, stair descent, standing, sitting, lying), totaling over 72 minutes of synchronized multi-sensor data. Experimental evaluation demonstrates that Random Forest achieves superior classification accuracy of 96.13%, significantly outperforming XGBoost (85.22%) and LinearSVC (58.54%). The publicly released dataset and benchmarking results provide a valuable resource for the HAR research community, enabling reproducible experimentation and facilitating advancement in multi-limb activity recognition systems.

Predictive Modeling of Lithium-Ion Battery During Discharging Stage Using CNN-DenseNet and LSTM

Mon, 08 Jun 2026 00:00:00 +0000

Lithium-ion batteries (LIBs) have become essential to renewable energy technologies, enabling the storage of electricity generated from renewable sources. This study presents an application of Long Short-Term Memory (LSTM) method and CNN-DenseNet to predict the lithium-ion battery during the discharging stage to overcome the crucial problem of battery performance degradation in electric vehicles and energy storage systems. To examine the LSTM model, a specially developed battery monitoring system based on an STM32 microcontroller and a Raspberry Pi 4 microcomputer for data acquisition was used. Testing was done on different discharge patterns at different loads (100W, 130W, 180W, 200W, 220W) on 12V/70Ah lead-acid free-maintenance battery with data collected for 1-10 hours discharge cycles. The monitoring system was designed with voltage sensor, current sensor, and temperature sensor to anticipate disturbances during data capture. The proposed LSTM model achieved superior performance, with RMSE = 0.0847, MSE = 0.0505, and MAE = 0.03548, significantly outperforming the CNN-DenseNet approach (RMSE = 0.3333, MSE = 0.4037, MAE = 0.3051) across diverse testing conditions. Although DenseNet performed best under moderate load (100 W), the LSTM architecture excelled under high load (220 W), underscoring the adaptability of the proposed model across varying operating conditions. This study expands the maintenance framework for energy storage systems through an LSTM architecture offering greater accuracy and stability.