Energy-Efficient Algorithms for Predictive Maintenance of Electric Vehicles
In the realm of modern technology, predictive maintenance stands as a key player in ensuring operational efficiency and sustainability. For predictive maintenance systems, implementing advanced algorithms on embedded devices like the Coral Dev Board is essential for processing and analyzing data efficiently and rapidly. These devices, combined with data visualization platforms like Grafana, offer a comprehensive solution for proactive monitoring and management of systems.

The aim of this project is to evaluate the effectiveness of four advanced machine learning algorithms—CNN, TCN, RandomForest, and XGBoost—in analyzing and predicting the charging patterns of on-road electric vehicles. By leveraging a comprehensive dataset of battery charging data, this study aims to identify the most accurate and efficient algorithm for forecasting battery charging behaviors. This includes predicting the duration of charging sessions, the energy required for optimal charging, and understanding the impact of various factors on charging efficiency.
The implementation of these algorithms was carried out on the Coral Dev Board, a powerful and energy-efficient device ideal for the embedded processing of machine learning algorithms. The AI backend is implemented as a service container that integrates these algorithms to develop predictive models based on the processed data. Chadha Jenzeri also developed a custom graphical interface on Grafana to display the prediction results in real-time. The integration of Prometheus for data collection and storage completes this solution, enabling clear and intuitive visualization of predictions and analyses. The interface allows users to monitor system performance, detect anomalies, and make informed decisions for preventive maintenance.
The primary objective of this work is to identify and select features that significantly impact battery charging behavior. This analysis considers various factors such as charging time, state of charge (SoC), charging power, ambient temperature, and electric vehicle usage patterns to construct a feature set conducive to accurate prediction models.
In conclusion, the implementation of these predictive maintenance algorithms on the Coral Dev Board and their integration with Prometheus and Grafana represents a significant advancement in the field of proactive maintenance. This project, led by Chadha Jenzeri, offers a complete and effective solution for monitoring and managing complex systems, paving the way for more reliable and sustainable systems.
