Predictive Maintenance and Health Estimation of E-Bikes Using Nearest Neighbor Classification

Small mobility e-bikes are considered as an environmental friendly option for commute over smaller distances. Recently there has been a boom in this sector. These bikes are used locally as well as in the tourist destinations. However, these e-bikes are driven by users with different driving patterns on different terrains. This leads to mechanical degradation in the e-bike health in terms of the braking system, braking pads and cables. In this paper, a k-nearest neighbor (KNN) classification model based health estimation has been implemented for the ebikes. The performance comparison of three different models-Fine KNN, Medium KNN and Weighted KNN is presented in terms of prediction accuracy, prediction speed and training time. The classification model is trained on the actual acceleration, braking lever force and displacement and the bike speed estimation which is based on wheel speeds and bike deceleration sensor data. The trained model is then used to predict the state of health of the e-bike. The state of health consist of the the brake cable tension level, and brake pad degradation level.