Abstract:
Road safety and operational efficiency depends on vehicle tire health. Accidents,
increased maintenance cost and poor fuel efficiency are some of the effects of poor
tire condition. However, the traditional tire inspection methods are usually time
consuming and need expert knowledge. In response to this need, this project develops
a Tire Health Monitoring System powered by deep learning techniques. To classify
tire health based on images provided by users, the system uses MobileNetV2, a
lightweight model famous for its computational efficiency.
The system comprises a mobile application for users to upload tire images and enter
tire serial numbers. The trained deep learning model analyzes these images to detect
the tire condition in one of the four categories, excellent, good, poor, and cracked. A
backend system supporting this functionality manages tire data, tracks health trends
and issues notifications to users when maintenance or tire replacement is required.
The dataset of tire images was trained using a model, fine tuned and early stopped to
limit overfitting and generalize the model for different size tires. The evaluation of the
model used in this thesis is presented: evaluation metrics of accuracy, precision, recall,
F1 score, and confusion matrix present accuracy of 94%. The results show that the
system can supply accurate and reliable tire health assessments.
The Tire Health Monitoring System described in this report provides a proactive
approach to tire maintenance, thereby reducing the risk of tire related accidents,
minimizing downtime and optimizing vehicle performance. Additionally, the system
prolongs tire life and enhances fuel efficiency, demonstrating considerable
environmental and cost saving benefits. This project is an example of integrating
machine learning with mobile technology which has the potential to change the way
of vehicle safety and maintenance in the future, and offers a path for the future
development of more advanced solutions in the automotive industry.
