Data-Driven Modeling of Tire Stack Compression Based on Regression Algorithms

This thesis aims to identify, validate and test a machine learning algorithm capable of estimating the compression of tire stacks, in order to accurately derive their overall height and that of the individual compressed tires. The developed system must be design to modeling the mechanical behavior of the tire under load, estimating compression based on the physical, geometric, and material characteristics of each element. To this end, several supervised regression models are implemented and trained. These models, starting from the available input data, allow us to predict the compression associated with each tire. As a first step, the dataset construction phase is approached using a rigorous and systematic strategy, including data cleaning, feature normalization, selection of relevant variables, and division into train, validation, and test sets. Model performances are evaluated through a quantitative and qualitative analysis of prediction errors, using appropriate statistical metrics, and through hyperparameter optimization procedures aimed at minimizing errors. Finally, based on the experimental results obtained, the model with the best predictive capabilities is selected, capable of more reliably estimating the physical parameters required to calculate the final heights of the compressed tire stacks.