Diagnostic and Prognostic Analysis Optimization of Field Problems for EV Charging Stations

Abstract

As transportation is one of the most polluting sectors in our daily life, electric vehicle are largely promoted to reduce greenhouse emissions and fuel consumption to protect our environment. The increase of the electric vehicles on the roads raises the demand for the EV charging stations. It is not avoidable that a fault happens on an EV charger or the hardware components of the EV charger degrades during utilization. Therefore, EV chargers need to be serviced to guarantee it continuously delivering electricity to EVs. However, service engineers can only handle a few charger problems every day, which may cause a long downtime for EV chargers. In addition, in such a complex system, it is quite challenging to accurately locate the cause of the faults. Thus, an intelligent diagnostic and prognostic system for EV chargers in highly desired. In this thesis, we investigated to what extent can machine learning help improve prognosis and diagnosis of EV chargers. To address the problem, first, we researched and identified what available sensor data can be utilized to reflect the condition of the chargers. Subsequently, we addressed how to expose the sensor data from the chargers and how to translate these data into a meaningful dataset. Moreover, we pre-processed the dataset in order to feed into the DBSCAN framework and chose five metrics to evaluate the results. In addition, we conducted five experiments to evaluate the influence of fine-tuning the DBSCAN parameters. The results show favorable performance based on all the metrics, such that the proposed method is valid. The performance of the DBCAN algorithm on our dataset is the best when $\epsilon=0.10$ and MinPts=4. The results of this work can be regarded as a baseline for further research on this topic.