Jurnal Nasional Teknik Elektro dan Teknologi Informasi

Genre-Aware and User Guided Music Generation

Daniel Paskah Toti Limbong — 2026-07-20

Music plays a crucial role in human life as a medium for emotional expression and entertainment. However, traditional music composition is time-consuming and requires expert skills, limiting accessibility for non-musicians. Recent advancement in artificial intelligence (AI) such as generative music enables the automatic creation of music, yet many models fail to incorporate explicit user preferences such as genre and composition length. This study proposed, user-guided Transformer XL, a genre-aware and user-guided music generation framework that worked on top of Transformer-XL. The system allowed users to specify desired genre and length, while a built-in genre classifier validated the stylistic accuracy of generated outputs. A genre classification module along with the number of bars was incorporated to ensure the generated outputs reflect the intended musical styles. The model was evaluated through a genre characteristic analysis and expert evaluation, revealing strong consistency in genre-specific features. Extensive experiments using the Lakh MIDI dataset demonstrated strong model performance, achieving an overall accuracy of 78.33%, with notable genre-specific strengths, particularly in jazz with recall 1.00 and classic with recall 0.88. In contrast, subjective evaluation using expert evaluation yielded a more moderate accuracy of 60%, highlighting the model’s ability to capture genre-relevant features even under more stringent, nuanced evaluative standards. Overall, this work demonstrates that integrating user preferences into generative modelling enhances flexibility and musical relevance, offering a robust foundation for future development in adaptive AI-driven composition.

Leak Detection in Pipeline System: A Comparative Study

Dian Mursyitah — 2026-07-20

Leak detection in pipeline systems is a critical challenge for ensuring safety, operational efficiency, and environmental sustainability. Conventional methods often struggle when applied to nonlinear systems subject to noise and uncertainty. This study aimed to perform a comparative analysis of four nonlinear observers, the high-gain observer (HGO), the extended Kalman filter (EKF), the EKF-based nonlinear observer (EKF-NO), and the finite memory observer (FMO), applied to a nonlinear pipeline model under leak scenarios. Each observer was evaluated via simulation for convergence speed, robustness to measurement noise, and sensitivity to initial conditions. Performance was assessed through simulation based on convergence speed, robustness to noise, and sensitivity to initial estimation errors. The results showed that the HGO achieved very fast convergence but amplified measurement noise significantly, thereby reducing detection sensitivity. The classical EKF produced smoother estimates but depended heavily on accurate model initialization and was sensitive to modeling errors. The EKF-NO improved estimation accuracy by accounting for nonlinear dynamics but converges more slowly than the FMO. Quantitative results showed that the FMO detected leaks in approximately 2 s, whereas the EKF required approximately 4–5 s to produce a reliable residual. In conclusion, the FMO achieved the best trade-off between convergence, robustness, and accuracy, producing the most stable and interpretable residuals for leak detection. Its robustness to noise and uncertain initial conditions makes it highly suitable for practical deployment in water, oil, and gas pipeline systems, and the approach shows strong potential for integration into real-time monitoring frameworks.

Recharging Prepaid Energy Meter via Internet Using ESP32-CAM and Telegram Messenger

Yohana Susanthi — 2026-07-30

In Indonesia, the requirement to manually enter electricity tokens into prepaid energy meters poses a significant challenge, particularly when customers are away from home. The inability to recharge remotely poses a risk of power outages and potential material loss. This research aimed to design an Internet-based remote recharging system that automated the physical button-pressing process and monitored the meter’s status remotely. Additionally, the system should be able to detect alarm sounds from the meter and send notifications to the customer when the electricity units are running low. The system consisted of a Scotch yoke mechanism with three servo motors controlled by an ESP32-CAM to move the arm and press the button on the prepaid energy meter. A sound rhythm-based pattern recognition method was used to detect alarm sounds with NodeMCU. Test results indicated a 100% success rate in recharging the prepaid energy meter. The average total time taken for the system to recharge the prepaid energy meter was 65.69 s. The highest accuracy score, 94.5%, was achieved by the system in detecting alarm sounds. The maximum noise level for the system to effectively detect alarm sounds was 55 dB. The implementation of this internet-based system eliminates physical constraints in recharging prepaid energy meters, thereby enhancing user accessibility and convenience.

Context-Aware Lightweight Indonesian Sign Language Recognition Using LSTM and IndoBERT

Amanda Betania Maritza — 2026-08-28

Communication barriers between deaf and hearing individuals remain due to the lack of affordable and computationally efficient assistive technologies, especially for Indonesian Sign Language (Sistem Isyarat Bahasa Indonesia, SIBI) in low-resource educational settings. Existing approaches often rely on computationally intensive architectures or large word-level datasets, making them less suitable for deployment in low-resource environments. This study proposed a novel context-aware SIBI recognition framework that effectively balanced recognition performance and computational efficiency. The system operated primarily at the alphabet level and provided assistive word prediction without requiring large word-level sign datasets. The proposed method used keypoint-based visual feature extraction combined with a long short-term memory (LSTM) network to capture temporal gesture patterns. A Trie-based prefix search was used to generate candidate words, followed by contextual refinement using the Indonesian version of bidirectional encoder representations from transformers (IndoBERT) to produce meaningful word predictions. This design enabled efficient temporal modeling and lightweight contextual processing within a central processing unit (CPU)-friendly architecture suitable for low-resource environments. Experimental results showed that the model achieved an inference accuracy of 93.59%, with a precision of 94.38%, recall of 93.59%, and an F1 score of 92.83%, while maintaining near real-time performance on CPU-only hardware. Statistical evaluation through analysis of variance (ANOVA) showed that the model performance was stable and McNemar’s test demonstrated that the contextual modeling component brought a statistically significant performance improvement. These findings demonstrate that accurate, context-aware, and computationally efficient SIBI recognition can be achieved on low-cost hardware, supporting practical deployment in real-world assistive communication scenarios.

Augmented Reality and Neural Network-Based Speech Recognition for Tahsin Digital AR

Idria Maita — 2026-08-28

Accurate Al-Qur’an tajwid (tahsin) requires repeated practice and direct guidance on makharijul huruf and basic tajwid. During post-pandemic blended learning, State Islamic Higher Education Institutions (Perguruan Tinggi Keagamaan Islam Negeri, PTKIN) students in Riau and Riau Islands often rely on asynchronous materials that provide limited articulation visualization and feedback. This study reported on the development of Tahsin Digital AR, an Android application that combined marker-based augmented reality with audio-visual demonstrations to support letter articulation practice. The study followed an educational research and development design (Borg and Gall) that included needs analysis, prototype development, expert review, and field evaluation. Media validity was assessed by three technology experts and three subject matter experts using a structured questionnaire; user acceptance was measured through a survey of 300 students from three PTKIN. Technology and content validation reached 84.55% and 93.89%, respectively, resulting in an overall feasibility score of 89.22% (highly valid). The survey results indicated a strong need for the application, with 96.0% agreed/strongly agreed that this application needed to be developed, and 92.7% agreed/strongly agreed that this application was practical to use. Most respondents (77.3%) also agreed/strongly agreed that this application was more interesting and effective for learning to read the Qur’an. These findings indicate that the proposed AR media is feasible and well-received as a tahsin support in the PTKIN context. Future work should include experiments on learning outcomes and automatic pronunciation assessment.

Data-Driven Classification of Maintenance Events in the Remanufacturing Industry Using Machine Learning

Lii'zza Aisyah Putri Sulistio — 2026-08-31

Machinery failure occurs when equipment can no longer perform its intended function, either partially or completely, leading to production delays and increased maintenance costs. While machine learning has been used to predict equipment failure, existing approaches tend to focus on predicting failure at the component level or the timing of failure without classifying maintenance types. This study addressed this gap by classifying maintenance events at the machine level for a heavy equipment remanufacturing company in Balikpapan, Indonesia. Historical records from January 2023 to May 2025 were analyzed, comprising 29 monthly performance logs; 2,484 equipment records; and 7,721 maintenance tickets. Maintenance events were categorized as either breakdown (complete loss of function) or corrective (partial functional degradation allowing continued operation). Key features included physical availability (PA), mean time between failures (MTBF) and mean time to repair (MTTR). Four supervised machine learning models, namely support vector machine (SVM), random forest (RF), extreme gradient boosting (XGBoost) and logistic regression (LR), were trained and evaluated using accuracy, precision, recall and F1 score. XGBoost performed best, achieving an accuracy of 65.27%, a precision of 71.17%, a recall of 71.54%, and an F1 score of 71.36%, outperforming RF, LR, and SVM. Although the classification task is challenging due to real-world variability in maintenance conditions, the results demonstrate the effectiveness of ensemble-based models, particularly gradient boosting, for classifying maintenance events. These findings highlight the potential of data-driven analytics to support operational decision-making and maintenance planning in industrial environments.

Enhancing Distribution Transformer Lifetime Through Temperature-Based Load Balancing

Liliana — 2026-08-31

Distribution transformers are essential components in power systems, and their service life is significantly affected by load imbalance and elevated operating temperatures. Load imbalance increases neutral current and hotspot temperature, accelerating insulation aging and reducing operational reliability. This study aimed to extend transformer lifetime by applying a temperature-based load balancing method using actual current data from the Garuda Sakti Substation. The load imbalance level for each feeder was calculated prior to redistributing the loads through power flow simulations using Electrical Transient Analyzer Program (ETAP) software. Oil and winding temperatures were predicted using a linear regression model, and transformer lifetime was estimated according to the Institute of Electrical and Electronics Engineers (IEEE) C57.91-2011 standard by calculating hotspot temperature, relative aging rate, and equivalent aging time. The results showed that the proposed method reduced load imbalance by up to 83.85% and power losses by 56.18%. Winding temperatures decreased between 1.80% and 17.14%, while oil temperatures decreased by 1.80%–17.60%. Consequently, the estimated transformer lifetime increased by up to 3.68% at an ambient temperature of 35°C. In conclusion, the temperature-based load balancing method effectively reduced operating temperature and thermal aging rate, resulting in a measurable increase in the estimated lifetime of the distribution transformer, while also demonstrating its practical applicability as a cost-effective strategy for improving energy efficiency and asset management in modern power distribution systems.

Edge AI–Based Threshold-Free Control for Mushroom Farming Using K-NN on ESP32

Antonius Irianto Sukowati — 2026-08-31

Precise environmental control is critical for the successful cultivation of paddy straw mushrooms (Volvariella volvacea), which requires tightly regulated temperature, humidity, substrate moisture, and air quality. Conventional automation systems typically rely on fixed threshold logic (e.g., ‘if humidity < 80%, activate mist maker’) using isolated parameter triggers and lack contextual awareness of multivariate environmental dynamics. This study aimed to overcome this limitation by developing a threshold-free, context-aware control system that replaced rule-based triggers with multidimensional environmental state classification. To achieve this, a novel edge-artificial intelligence (AI) system that deployed a lightweight k-nearest neighbor (K-NN) classifier directly on an ESP32 microcontroller for real-time, closed-loop actuator control was proposed. The cultivation environment was modeled as four-dimensional discrete state space, yielding 81 expert-defined combinatorial classes that captured nuanced interactions among substrate moisture, air temperature, humidity, and CO₂ levels. Each class was mapped to a specific actuator vector (pump, heater, mist maker, fan), effectively encoding agronomic best practices into a machine-interpretable decision framework. The system achieved 98% classification accuracy, operated with only a 15 KB memory footprint, and completed inference in 12.3 ms, demonstrating that lightweight machine learning could replace rigid rule-based logic in resource-constrained agricultural settings. Beyond improving environmental stability and actuator reliability, this work establishes a new paradigm for knowledge-preserving, edge-intelligent automation in smart agriculture.

Design of C-band Microstrip Antenna Using U-Slot and Defected Ground Structure

Fitri Amillia — 2026-08-31

Microstrip antennas are widely used in wireless communication systems because of their lightweight nature, compact size, and ease of fabrication. However, this type of antenna has inherent limitations, such as narrow bandwidth and low gain. To address these shortcomings, this study proposed a microstrip antenna operating in the C-band frequency range by applying a U-slot technique on the patch and a dumbbell-shaped DGS on the ground plane. The U-slot technique was used to enhance bandwidth, while the dumbbell-shaped defected ground structure (DGS) improved impedance matching and overall antenna characteristics without increasing antenna size. The antenna design process used analytical calculations for both the single-element and 1 × 4 array configurations, followed by three-dimensional full-wave electromagnetic simulations and fabrication on an FR-4 substrate. The antenna was measured using a vector network analyzer (VNA) to obtain the S-parameters. Simulation results showed that the 1× 4 array design achieved a return loss of -46.53 dB, a bandwidth of 376 MHz, a voltage standing wave ratio (VSWR) of 1.01, and a peak gain of 8.49 dBi. Measurement results indicated a return loss of -17.50 dB, a bandwidth of 480.81 MHz, and a VSWR of 1.31 at 5.4875 GHz. These results demonstrate that integrating the U-slot and dumbbell DGS effectively improves antenna performance. Overall, the proposed antenna meets the required specifications and shows potential for weather radar applications as a receiving antenna operating in the C-band front-end system.

YOLOv8-Based Thermal Image Road Segmentation Using Trapezoid Zone Detection for Autonomous Navigation

M Sadam Al Jabbar — 2026-08-31

Autonomous vehicle perception systems predominantly use red-green-blue (RGB) cameras, which experience significant performance degradation under adverse lighting conditions, with detection accuracy declining by 35–45% at night and in inclement weather. This limitation poses substantial safety risks for practical autonomous vehicle deployment. This study developed a robust thermal image-based road segmentation system using a You Only Look Once (YOLO) v8 deep learning architecture integrated with trapezoid zone detection for real-time autonomous navigation capable of consistent all-weather operation. The methodology encompassed five stages, including collection and annotation of thermal images using a FLIR Boson camera across diverse road conditions, two-stage model training employing YOLOv8n-seg architecture with copy-paste augmentation on NVIDIA Jetson Orin AGX, development of a polynomial regression-based distance calibration system, implementation of a six-zone priority-based trapezoid navigation framework, and comprehensive system integration with performance evaluation. The experimental results demonstrated that the proposed system achieved road segmentation average precision (AP) of 98.3% at an intersection over union (IoU) threshold of 0.5, with an overall mean AP (mAP) of 83.3% across six object classes after the second training stage. The copy-paste augmentation strategy improved minority-class (people) detection by 14.0%. The distance calibration system attained an R² score of 1.0 within the measured range. The daytime-trained model successfully detected objects at nighttime without retraining, demonstrating the fundamental advantage of thermal imaging. The system processed at 29.8 frames per second with 18 W power consumption. These findings validate thermal imaging effectiveness for all-weather autonomous navigation and establish a framework for embedded autonomous vehicle deployment.