Tire wear progression is a nonlinear and multi-factor degradation phenomenon that directly influences vehicle safety, handling stability, braking performance, rolling resistance, and fleet operational cost. Global accident investigations indicate that accelerated or undetected tread depletion contributes to nearly 30% of highway tire blowouts, highlighting the limitations of conventional wear indicators such as physical tread wear bars, mileage-based service intervals, and periodic manual inspections. These manual and threshold-based approaches fail to capture dynamic driving loads, compound ageing, pressure imbalance effects, or platform-specific wear behaviours, thereby preventing timely intervention in real-world conditions.
This work presents an Indirect Tire Wear Health Monitoring System that employs an advanced Machine Learning + Transfer learning architecture to infer tread wear level and Remaining Useful Life (RUL) without relying on any tire-mounted sensors. The system ingests CAN bus telemetry signals (e.g., wheel torque, longitudinal/lateral accelerations, brake pressure, speed distribution, steering dynamics, thermal exposure) and converts them into high-resolution wear state estimations through a multi-stage feature learning pipeline.
A transfer-learning layer enables model domain adaptation across tire brands, rubber compounds, rim sizes, inflation pressure ranges, and axle-loading variations — reducing retraining cost and ensuring cross-platform.
The pipeline supports both cloud analytics workloads (fleet health dashboards, risk scoring, and advisory scheduling) and real-time embedded inference on in-vehicle microcontrollers for predictive safety intervention. On-road validation experiments demonstrate that the proposed model maintains high correlation to ground truth tread depth measurements, delivering per-tire wear estimation, non-linear RUL curves (in km and %), progressive wear trend modelling, and dynamic replacement advisory logic. The proposed architecture therefore establishes a scalable, sensor-less predictive maintenance framework suitable for OEM, Tier-1, and fleet-operations deployment.