Abstract
This study addresses the need for developing new frameworks to monitor and detect sensor failures in connected commercial vehicles (CCV)s. The CCV's sensor health is more important when performance predictions and other communication-related errors (e.g. cyber-physical attacks) can manipulate the sensory network's resiliency. We developed a novel machine learning (ML)-based framework, AutoDetect, to equip the cloud-tied operators with tools for understanding the abnormal sensor data streaming from the vehicle on the cloud level which explains the sensor data errors due to sensor failures only. We developed an innovative autoencoder (AE) neural network algorithm coupled with K-means clustering to create patterns. To learn the relationship between operating samples and features, when streaming sensor data over high-dimensional datasets is collected in the United Kingdom (UK). Different profiles of sensor data are collected under various driving conditions to monitor the ground truth of the sensor's confidence levels in CCVs. The new AutoDetect tracked real-time sensor failures with a minimum accuracy of 90%.