Virtual sensor - AI model training using VOLVO Brake temperatures

Abstract

Accurate prediction of brake disc temperatures in heavy-duty vehicles is essential for ensuring safety, reducing wear and improving braking performance. Excessive heat buildup in the disc can lead to brake fade, accelerated material degradation and increased emissions of harmful wear particles. This thesis focuses on predicting brake disc temperatures using time-series data collected from controlled dynamometer tests. The dataset includes braking signals such as torque, pressure and speed, recorded at high frequency under a wide range of operating conditions. Various machine learning models, including neural networks, were developed to predict brake disc temperatures during individual braking events. This work serves as a foundation for future efforts to extend temperature prediction models to real-world field data and ultimately support the development of intelligent thermal monitoring systems that can reduce brake wear, improve safety and help meet upcoming Euro 7 regulations on particle emissions from braking systems.