IJEIS (Indonesian Journal of Electronics and Instrumentation Systems)
Not a member yet
300 research outputs found
Sort by
Integrasi Robot Lengan Beroda Holonomic dan Pengindraan Visual Menggunakan Yolov5 untuk Pemilah Sampah
Full text linkThe issue of waste management has become a critical topic, particularly in the sorting process, which relies on human labor and poses hygiene risks and inaccuracy. Several technologies have been explored to address this problem, including robotic arms with visual sensing, which are widely used but face challenges such as limited working areas and relatively complex installation processes. This study develops a waste-sorting robot based on a holonomic-wheeled robot integrated with visual sensing and the YOLOv5 algorithm for waste classification. The robot is equipped with a vacuum gripper for waste pickup and placement, as well as sensors for navigation and position control. Tests were conducted on four types of waste: bottles, leaves, metals, and paper. The results demonstrate a classification accuracy rate of 100%, with an average waste placement success rate of 90% for leaves and paper, and 80% for bottles, influenced by the surface characteristics of the waste and the consistency of the robot's positioning. This robotic system offers enhanced efficiency and accuracy compared to manual methods, although there remains room for improvement in the gripping mechanism and synchronization of the robot's movements. Overall, the robot system shows performance with accuracy above 80% with a wider working area than using a robot arm
Pengembangan Prototipe Sistem Peringatan Dini Tabrakan Belakang Pada Truk Berbasis Arduino
Full text link Rear-end collisions with trucks often occur due to the lack of driver awareness in maintaining a safe following distance, as well as the presence of blind spots around the truck, especially at the rear. These blind spots make it difficult for drivers of vehicles too close to the truck to be detected, thus increasing the risk of collision, especially during sudden braking. To address this issue, this study develops a prototype of a rear-end collision early warning system based on Arduino, using TF02-Pro LiDAR and MK421137 speed sensors. This system is designed to detect the distance and speed of vehicles behind in real-time and provide an alarm warning if the distance and speed conditions reach dangerous thresholds. Testing was carried out using simulations with variations in the speed of the following vehicle at 15 km/h, 20 km/h, and 30 km/h. The results showed that the system successfully detected the speed of the following vehicle with an accuracy of 97.75% and an average error of 2.25%. Furthermore, the alarm status was successfully activated based on the predefined distance and speed thresholds. This prototype is expected to enhance road safety by providing effective early warnings to drivers behind trucks
Stunting Classification Model For Toddlers Using SMOTE and Support Vector Machine (SVM) (Case Study: Samalanga Community Health Center)
No full textStunting is a growth disorder that has long-term impacts on child development. This study aims to develop a classification model for determining stunting status in toddlers using the Support Vector Machine (SVM) algorithm, with a case study conducted at the Samalanga Community Health Center. The dataset used consists of 1,205 toddlers. The research stages include preprocessing, data balancing using SMOTE, and parameter tuning using GridSearchCV. The developed model successfully achieved an accuracy of 0.97, an ROC-AUC of 0.96, and an average f1-score of 0.97. These results indicate that the model can accurately distinguish between stunted and non-stunted toddlers. Benchmarking against public datasets shows that the model in this study has a 2% higher accuracy and a 4.7% higher ROC-AUC value compared to previous studies. These findings indicate that the applied pipeline approach is effective in improving classification accuracy. The resulting model has the potential to support fast and accurate stunting classification.
Rancang Bangun Filter EMI (Electromagnetic Interference) Untuk Mereduksi Noise Gangguan Elektromagnetik
Full text linkGangguan elektromagnetik (Electromagnetic Interference/EMI) merupakan permasalahan serius dalam sistem kelistrikan dan elektronika karena dapat menurunkan kinerja perangkat. Untuk itu, diperlukan filter EMI yang mampu mereduksi noise atau gangguan elektromagnetik secara efektif. Penelitian ini bertujuan untuk mengetahui efektivitas berbagai jenis filter EMI dalam mereduksi noise pada sistem kelistrikan. Pengujian dilakukan di CV. AMX UAV Technologies selama sepuluh hari dengan menggunakan empat jenis filter EMI, yaitu PSIM, ALPHA-SL, NEW ALPHA-PSIM, dan WURTH Elektronik. Beban yang digunakan dalam pengujian meliputi kompor listrik, lampu pijar, dan vacuum cleaner. Metode pengujian dilakukan dengan mengukur tegangan ripple sebelum dan sesudah melewati filter menggunakan osiloskop. Hasil menunjukkan bahwa filter PSIM, ALPHA-SL, dan NEW ALPHA-PSIM mampu mereduksi amplitudo noise dari rentang 68Vpp–254Vpp menjadi sekitar 2Vpp. Sementara itu, filter WURTH hanya mereduksi hingga 16Vpp–46Vpp. Hasil ini menunjukkan pentingnya pemilihan filter EMI yang tepat guna meningkatkan kualitas daya listrik dan mengurangi gangguan elektromagnetik. Penelitian ini diharapkan dapat memberikan kontribusi terhadap pengembangan sistem perlindungan EMI yang lebih efisien di berbagai sektor industri
Sistem Klasifikasi Sampah Otomatis Berbasis Deteksi Objek Real-Time Pada Single Board Computer Dengan Algoritma YOLO
Full text linkThe development of an automatic waste classification system based on real-time object detection using the YOLO (You Only Look Once) algorithm on a Raspberry Pi 5 Single Board Computer (SBC) is the main focus of this final project. The main issue addressed is the increasing accumulation of waste, particularly in Indonesia, which requires an effective solution for automatic waste sorting. The system is designed to detect and sort plastic and metal waste in real-time using deep learning and computer vision technologies.This research employs the YOLO11n model, trained on a dataset of plastic and metal waste. The training process involves data augmentation techniques such as rotation and grayscale to enhance dataset variability. The training results show a mean Average Precision (mAP) of 98.44% on testing data. The system is implemented on a Raspberry Pi 5, with the model converted to NCNN format to improve inference speed. Testing results indicate that the system can achieve a speed of 8.90 FPS with a latency of 110 ms, meeting the criteria for a real-time system
Sistem Manajemen Energi Hibrida pada Sumber Energi Baterai-Superkapasitor Berbasis Automatic Switching
Full text linkThis research develops an automatic switching-based hybrid energy management system that integrates batteries and supercapacitors to improve power distribution efficiency in electric vehicles. The system is controlled by a microcontroller that monitors the current in real-time and activates the supercapacitor line when the current exceeds 10 A. The switching circuit uses IR2110 and IRFP4568 MOSFETs. Tests were conducted in three scenarios: battery only, hybrid without load, and hybrid with load. In the no-load condition, the supercapacitor produced a peak current of 18.23 A and an average of 5.73 A, while the battery recorded an average current of 8.52 A. In the loaded condition, the supercapacitor peak current reached 20.87 A, with an average of 10.82 A, while the battery was 11.95 A. The total energy increased from 55022.25 J (battery only) to 92000.76 J (no-load) and 147019.09 J (with load). The efficiency also increased from 1.0% to 2.4% in the hybrid configuration. The system showed a stable energy conversion efficiency of 95% under both hybrid conditions. These results prove that automatic integration of supercapacitors can improve system efficiency and performance without the complexity of control algorithms
ANOTASI OTOMATIS PADA CITRA PERSEPSI AUTONOMOUS CAR MENGGUNAKAN TRANSFER LABEL 3D-TO-2D
Full text linkPemetaan semantik dan panoptik pada citra kendaraan otonom sangat bergantung pada ketersediaan anotasi manual yang memadai. Namun, proses pelabelan manual memerlukan waktu dan biaya tinggi, serta rentan terhadap inkonsistensi antar frame. Penelitian ini mengusulkan pendekatan otomatis berbasis Neural Radiance Fields(NeRF) untuk mentransfer label dari representasi 3D ke citra 2D. Dataset KITTI-360 digunakan dengan 768 citra yang dianotasi secara semi-otomatis menggunakan segmentor PSPNet dan Mask2Former. Model dilatih secara terpisah pada tiga sequence berbeda dan dievaluasi menggunakan metrik Intersection over Union (IoU), mean IoU (mIoU), dan Panoptic Quality (PQ). Hasil menunjukkan model mampu mencapai mIoU sebesar 0.79 (perspektif) dan 0.73 (fisheye), serta PQ sebesar 0.66 dan 0.60. Hasil kualitatif menunjukkan segmentasi yang konsisten pada kelas dominan seperti jalan, bangunan, dan langit, serta kemampuan membedakan instance objek dengan baik. Pendekatan ini terbukti mampu menghasilkan anotasi berkualitas tinggi tanpa ketergantungan pada pelabelan manual. Temuan ini penting untuk mempercepat pembangunan sistem persepsi visual kendaraan otonom dan memperluas cakupan data pelatihan secara efisien
Deteksi Partial Discharge dengan Metode CNN VGG16
Full text linkPartial discharge adalah peristiwa loncatan listrik pada bahan isolasi listrik yang menimbulkan kerusakan pada peralatan listrik. Untuk itu diperlukan suatu metode untuk mendeteksi peristiwa partial discharge. Salah satu metode yang dapat digunakan untuk deteksi partial discharge adalah metode CNN VGG16. CNN akan melakukan pemodelan dari analisa dataset gambar partial discharge VSB lalu menggunakannya untuk mengklasifikasikan data baru sebagai partial discharge atau tidak. Pada penelitian ini akan dianalisa bagaimana pengaruh parameter pemodelan dan pembagian dataset terhadap peforma. Penyesuaian parameter dilakukan dengan memvariasikan nilai learning rate, steps per epoch, dan validation steps untuk melihat nilai terbaik sehingga nantinya nilai terbaik yang akan digunakan. Pembagian dataset dilakukan dengan tiga variasi yaitu pembagian train, validasi, dan test pada dataset pertama dibagi rata, yang kedua didominankan ke train, dan yang ketiga jumlah data noPD terlebih dahulu dikurangi agar seimbang dengan PD kemudian data didominankan juga pada train. Berdasarkan penelitian, terbukti bahwa variasi dataset ketiga yang memiliki peforma terbaik dan menunjukkan bahwa CNN arsitektur VGG16 terbukti mampu untuk mengenali pola dari data sinyal partial discharge dan membuat model yang mampu mengklasifikasi data partial discharge atau tidak dengan akurasi train 95,70%, akurasi validasi 93,12% dan akurasi prediksi data test 92,50% juga dengan nilai MCC sebesar 0,7
Optimizing Biodegradable Waste Conversion with IoT-Based Bioreactor System
Full text linkIn Indonesia, the problem of organic waste continues to grow in parallel with population increase and urbanization. However, the management of organic waste still relies on conventional methods that are often ineffective. Conversely, organic waste holds significant potential to be transformed into organic fertilizer and renewable energy through appropriate technological approaches. This study aims to design and implement an organic waste management system that incorporates automatic sorting and biogas conversion. The system is integrated with an Internet of Things (IoT)-based monitoring system at the 3R Integrated Waste Processing Site (TPST 3R) in Mulyoagung, Malang Regency. The system consists of an anaerobic bioreactor that produces methane gas and liquid fertilizer, an automatic sorting unit equipped with color, IR, and weight sensors, and an IoT dashboard for real-time monitoring of temperature, pH, and gas pressure. The research employed both technical and participatory methods, including community training and processing output evaluation. Results indicate a sorting efficiency of up to 92%, biogas production of approximately 22 m³ per ton of organic waste, and a significant increase in community participation. The system proved to be effective, adaptable, and capable of delivering positive social, economic, and environmental impacts.
Komparasi Performa Model 3D CNN dalam Klasifikasi Demensia Alzheimer pada MRI Otak
Full text linkPenyakit Alzheimer adalah jenis demensia akibat kerusakan pada neuron otak yang memengaruhi memori, bahasa, dan berpikir. Diagnosis manual sering rentan terhadap subjektivitas dan memakan waktu, sehingga diperlukan model otomatis seperti 3D CNN untuk klasifikasi tingkat keparahan Alzheimer. Namun, kompleksitas arsitektur 3D CNN menyebabkan waktu komputasi yang tinggi. Penelitian ini membandingkan tiga arsitektur model 3D CNN yaitu 3D ResNet, 3D ResNeXt + Bi-LSTM, dan 3D CNN + CLSTM untuk menentukan model yang optimal. Dataset yang digunakan diperoleh dari database ADNI. Performa model dievaluasi dengan menggunakan confusion matrix, akurasi, presisi, recall, F1-score dan waktu komputasi.Hasil penelitian menunjukkan bahwa 3D ResNet memiliki akurasi pelatihan tertinggi mencapai 99,54% dan waktu komputasi pelatihan sebesar 57,61 detik/epoch. Model 3D ResNeXt + Bi-LSTM mencapai akurasi pengujian sebesar 99,33% dan waktu inferensi tercepat yaitu 0,0182 detik/sampel, namun waktu komputasi pelatihan terlama yaitu 117,68 detik/epoch. Sementara itu, 3D CNN + CLSTM mencapai akurasi uji sempurna 100% tetapi memiliki waktu inferensi terlama yaitu 0,0268 detik/sampel. Penelitian ini menunjukkan bahwa arsitektur yang sederhana tetap dapat memberikan performa optimal dengan waktu komputasi yang lebih efisien dibandingkan model yang lebih kompleks