محمدرضا اسمي

Real-time estimation of the vertical stiffness of a tire plays a pivotal role in the development of intelligent tire monitoring systems an‎d advanced suspension control strategies. In this thesis, a data-driven framework based on machine learning was designed an‎d eva‎luated with the aim of achieving a highly accurate an‎d practically deployable model under real-world vehicle conditions. To collect operational data under realistic scenarios, a custom-built data acquisition system was developed an‎d mounted on a test vehicle. This system included a tire pressure sensor, infrared distance sensor, accelerometer unit, gyroscope, an‎d a central processing module. The collected field data, along with simulated data generated via CarSim software under various driving scenarios, were used to train an‎d validate the proposed models. On the modeling side, several machine learning algorithms including deep neural networks (DNN), support vector regression (SVR), ran‎dom forest (RF), Gaussian process regression (GPR), an‎d extreme gradient boosting (XGBoost) were implemented. Furthermore, two ensemble models with a stacking architecture were developed: Hybrid Model One (combining RF an‎d GPR) an‎d Hybrid Model Two (combining RF, GPR, an‎d XGBoost). Experimental results demonstrated that Hybrid Model Two significantly outperformed individual models in terms of both accuracy an‎d robustness, particularly under noisy inputs an‎d variable loading conditions. The proposed hybrid framework achieved an estimation accuracy approximately aligned with the direct stiffness values computed from CarSim simulations. Compared to the neural network-based approach in the study by Sun , it improved the coefficient of determination by, an‎d showed a increase over the support vector machine method used in Han. These improvements reflect the superior performance an‎d real-time applicability of the proposed machine learning ensemble in estimating tire stiffness from sensor-based vehicle data. Ultimately, the outcome of this research is a reliable, real-time, an‎d high-precision model that can serve as a foundational layer for the development of next-generation tire health monitoring systems, semi-active suspensions, an‎d advanced vehicle control platforms.

سختي عمودي تاير , سيستم پايش فشار باد تاير , يادگيري ماشين , شناسايي برخط , شبيه‌سازي كارسيم

Tire Vertical Stiffness , Tire Pressure Monitoring System , Machine Learning , Real-Time Identification , CarSim Simulation

Mohammad Reza Esmi

Dr. Massoud Masih Tehrani