Editorial Type: research-article
 | 
Online Publication Date: 23 Sept 2025

Prediction of the Frictional Power Distribution in the Tire Contact Patch Based on an Empirical Tire Model and an Artificial Neural Network

,
,
, and
Article Category: Research Article
Page Range: 176 – 202
DOI: 10.2346/TST-24-009
Save
Download PDF

ABSTRACT

To predict and prevent uneven tire wear in addition to a reduction of overall tire wear, it is essential to estimate not only the total amount of wear but also how the wear is distributed across the tire width. This requires knowledge of the frictional power distribution in the tire contact patch, which is the basis for calculating tire wear using a wear law. Usually, only 3D structural tire models can generate such distributed contact results. However, they involve high computational costs and cannot be used for comprehensive optimization of a vehicle’s suspension system with respect to tire wear characteristics. Hence, this contribution presents a methodology on how to accelerate the prediction of the frictional power distribution using two components: The structural tire model is replaced by an empirical tire model that on its own is not able to generate distributed contact results. Therefore, an artificial neural network is trained to predict the desired contact results from the kinematic quantities calculated by the empirical tire model. In the initial training phase, both components are fitted to data generated by the original complex tire model. After training, the empirical tire model can replace the structural tire model in vehicle simulations, resulting in significantly shorter calculation times. The simulation results are fed into the artificial neural network, which predicts the frictional power distributions over the tire width with negligible additional effort. Overall, the methodology reduces calculation time for the prediction of tire wear based on virtual test drives to approximately 25% of the time needed when using structural tire models.

Keywords: tire wear; tire model; surrogate model; friction model; frictional power distribution; friction work distribution; machine learning; artificial neural networks
Copyright: 2025
FIG. 1
FIG. 1

Process of predicting frictional power distributions using FTire tire model.

FIG. 2
FIG. 2

Arrangement of belt segments (left), tread strips (middle), and tread blocks/contact elements (right, red dots) in the FTire model.

FIG. 3
FIG. 3

Process with surrogate model.

FIG. 4
FIG. 4

Tire forces and moments (red), velocities (blue), and angles and radii (green).

FIG. 5
FIG. 5

Generation of surrogate model (training).

FIG. 6
FIG. 6

Visualization of Adams Car’s virtual tire test rig.

FIG. 7
FIG. 7

Distribution of samples in the κ–α plane (black dots) and corresponding frictional power at tread strip 11 in the middle of the tire (colored area).

FIG. 8
FIG. 8

Model architecture.

FIG. 9
FIG. 9

Inference overview.

FIG. 10
FIG. 10

Fit of PAC2002 model compared with FTire for pure longitudinal slip, κ = [−2.5%…2.5%], α = 0°, γ = 0°, Fz = 5000 N.

FIG. 11
FIG. 11

Fit of PAC2002 model compared with FTire for pure lateral slip, κ ≈ 0%, α = [−2.5°…2.5°], γ = 0°, Fz = 5000 N.

FIG. 12
FIG. 12

Fit of PAC2002 model compared with FTire for combined slip, κ = [−2.5%…2.5%], α = 2°, γ = 0°, Fz = 5000 N.

FIG. 13
FIG. 13

Longitudinal velocity vlon and lateral acceleration alat of the vehicle during the virtual test drive. Dotted lines mark the transition from city to rural road and from rural road to motorway.

FIG. 14
FIG. 14

Friction work Wf,str,s per tread strip s resulting from virtual test drive generated by the original FTire model with contact output (FTire + .cfo) and predicted by the AI friction model on top of the tire states generated by FTire (FTire + AI).

FIG. 15
FIG. 15

Accumulated friction work Wf,acc generated by the original tire model with contact output (FTire + .cfo) and predicted by the AI friction model (FTire + AI) over time t. Dotted lines mark the transition from city road to rural road and rural road to motorway.

FIG. 16
FIG. 16

Friction work Wf,str,s per tread strip s for full virtual test drive generated by original tire model with contact output (FTire + .cfo) and predicted by surrogate model (PAC + AI).

Contributor Notes

2Corresponding author. Email: lars.muth@uni-paderborn.de
  • Download PDF