Maqola

A scalable IoT-integrated neural network approach for real-time fault detection in electric vehicle power trains

G. SatishDepartment of Electrical and Electronics Engineering, SreeDattha Group of Institutions, Sheriguda, Ibrahimpatnam, Telagana 501510, IndiaN. Chitra KiranM.E. Shashi KumarDepartment of Mechanical Engineering, Amrita School of Engineering, Amrita Vishwa Vidyapeetham, Bengaluru, Karnataka 560035, IndiaM. SivaramKrishnanDepartment of Electrical and Electronics Engineering, Karpagam College of Engineering, Coimbatore, IndiaArunkumar MunimathanDepartment of Mechatronics Engineering, Hindusthan College of Engineering and Technology, Coimbatore-641032Ahmed Shakir Al‐HitiFaculty of Engineering Techniques, University of Almaarif, Ramadi 31001, IraqFarrukh BakhritdinovDepartment of Exact Sciences, Kimyo International University in Tashkent, UzbekistanAseel SmeratHourani Center for Applied Scientific Research, Al-Ahliyya Amman University, Amman, 19328, JordanRasoul karimiImam Khomeini Naval Science University of Nowshahr, Nowshahr, Iran

2026en

ABI

Annotatsiya

The performance of electric vehicle (EV) power trains is essential to minimise the unanticipated downtime and costs in the smart mobility age. In this research, an experimentally validated IoT framework is presented for predictive maintenance of EV powertrains for early fault detection using vibration, acoustic and current measurements and an Artificial Neural Network (ANN). A scaled-down EV powertrain, such as a Brushless DC (BLDC) motor, coupling, gearbox and variable load were developed to simulate various scenarios. Common faults, including gear tooth fracture, shaft unbalance, misalignment and bearing wear, were deliberately induced for fault simulation. Temporal and frequency features (RMS, kurtosis, skewness and spectral entropy) of the multi-sensor data were used to train a feedforward artificial neural network (ANN) classifier. The trained model achieved an accuracy of 98.4% under normal conditions and maintained high fault resilience (90.4%) under very noisy conditions (0 dB SNR) when compared to conventional Support Vector Machine and Random Forest classifiers. The entire sensor-to-cloud system implemented with an ESP32 platform (MQTT) and a Node-RED dashboard achieved an average sensing-to-diagnosis latency of 190 ms, which meets the real-time diagnostic needs. Results from a fleet-level simulation analysis indicate that the proposed system can save around 40% unplanned downtime and result in a 10% decrease in maintenance labor demand when representing typical maintenance activities. These projections are based on experimentally proven fault detection performance and should be validated in the field over extended periods of time. The findings demonstrate that the proposed ANN-based IoT system provides a cost-effective, energy-efficient and scalable approach for real-time predictive maintenance of EV powertrains, facilitating its integration into smart and sustainable transport systems.

Hali tarjima qilinmagan

Identifikatorlar

DOI: 10.1016/j.hybadv.2026.100708

Iqtiboslar va manbalar

0 ta iqtibos30 ta foydalanilgan manba