Machine Learning Approach for Water Dew Point Detection in Diesel Vehicles’ After Treatment System

Water droplet formation within the aftertreatment system of diesel vehicles poses a risk to the durability and performance of NOx sensors, necessitating conservative activation strategies that delay sensor operation and compromise emissions monitoring. This thesis investigates a machine learning approach for water dew point detection to enable earlier and safer sensor activation, improving the trade-off between protection and monitoring readiness. A comprehensive dataset was collected from a Euro 6d diesel vehicle equipped with a full aftertreatment system, capturing relevant thermal, pressure, and energy signals under diverse driving conditions. The dew point detection problem is formulated as a supervised regression task, aiming to predict an optimal threshold curve for sensor activation derived from droplet detection events and energy profiles. Various machine learning models, including linear regression, decision trees, and neural networks, are evaluated for their ability to predict the dew point threshold accurately while maintaining robustness across different driving styles and conditions. Feature selection, signal preprocessing, and domain generalization techniques are applied to enhance model performance and generalizability. Results demonstrate that the proposed machine learning approach can predict the dew point threshold with high accuracy, enabling earlier sensor release while mitigating droplet-related risks. This methodology shows potential for integration into diesel engine control units to enhance aftertreatment system efficiency and emissions compliance under real-world driving conditions.