Development of mechanistic pavement rolling resistance model considering dynamic loading

Liu, Xiuyu

Permalink

https://hdl.handle.net/2142/115573 Copy

Description

Title

Development of mechanistic pavement rolling resistance model considering dynamic loading

Author(s)

Liu, Xiuyu

Issue Date

2022-04-19

Director of Research (if dissertation) or Advisor (if thesis)

Al-Qadi, Imad L.

Doctoral Committee Chair(s)

Al-Qadi, Imad L.

Committee Member(s)

• Roesler, Jeffery R.
• Spencer, Billie F.
• Hajj, Ramez M.
• Chatti, Karim

Department of Study

Civil & Environmental Eng

Discipline

Civil Engineering

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• rolling resistance
• excess fuel consumption
• dynamic wheel load
• neural network
• transfer learning

Abstract

It is estimated that the transportation sector in the United States is responsible for 29% of greenhouse gas (GHG) emissions; combustion of petroleum-based products from moving vehicles is a major contributor. Various mechanisms in vehicles increase fuel consumption, including exhaust heat, engine friction, aerodynamic drag, and rolling resistance. It has been shown that 5 to 15% of fuel energy alone is consumed in overcoming rolling resistance by passenger cars and light trucks; for heavy trucks, this figure is closer to 15 to 30%. It has long been recognized that pavement-related rolling resistance (RR) caused by pavement-vehicle interaction (PVI) is an important component in the pavement life-cycle assessment (LCA). This study focuses primarily on the development of mechanistic RR models to quantify the excess fuel consumption (EFC) and dynamic wheel load (DWL) caused by pavement roughness. This study begins with the development of a 3D passenger vehicle model and a 3D semi-trailer truck model to simulate vehicle dynamic vibrations. These two models were used to investigate the impact of the road roughness characteristics, dynamic properties, and driving speed on the roughness-induced energy dissipation and the dynamic wheel load. The simulation results confirm the importance of local roughness variance—an indicator of road roughness nonstationarity—in characterizing the impact of road roughness on EFC. Using a 3D full-car model or a 3D truck model could reduce error prediction by up to 11%, compared with a simplified 2D half-car model or 2D half-truck model. The simulation results were used to develop a regression M-RSI model, which has an R2 of 97.7%, which allows integration and implementation into pavement LCA tools to evaluate the use-stage energy and environmental footprints related to road roughness. Because tire contact stresses and contact area directly affect PVI, accurate characterization of tire-pavement contact is important. It is also one of the challenges to pavement structural analysis under heavy truck tire loading. To achieve a rapid transformation from dynamic loading profiles to time-varying 3D contact stresses, two deep learning models were developed. The first model, ContactNet, is a baseline deep learning model that predicts tire contact stresses based on tire type, loading, inflation pressure, and slip ratio. The performance of the deep learning model was evaluated and compared with conventional machine-learning regression models. The second model, ContactGAN, is a generative adversarial neural network model that consists of a Generator and a Discriminator. Results show that the mean absolute error of the ContactNet prediction is 0.91 kPa. This suggests that the ContactNet significantly outperforms the four conventional machine-learning regression methods investigated in this study: polynomial regression, k-nearest neighbours, multi-layer perceptron, and random forests. The results show that the ContactNet model outperforms traditional machine learning algorithms in predicting the 3D tire-pavement contact stresses. The ContactGAN model, based on generative adversarial network (GAN), produces accurate contact stress distributions both qualitatively and quantitatively. It was demonstrated that transfer learning can reduce the prediction MAEs by 83.8% and 92.9% on the cornering tire dataset and the hyper-viscoelastic tire dataset, respectively. Finally, the truck dynamic model, deep learning tire model, and the pavement finite element model were integrated into a vehicle-tire-pavement approach for determining the pavement structure-induced rolling resistance (SRR) under dynamic loading. The distribution of dynamic wheel loads was evaluated using DLC. Simulations showed that the power dissipated within the pavement structure was collected to calculate the corresponding SRR, considering different levels of pavement roughness and vehicle speeds. The effect of dynamic loading on the SRR, due to road roughness, was quantified. At 40 mph, the impact of truck dynamic loading accounts for 4.7%, 7.4%, 10.7%, and 14.0% of SRR for good, fair, poor, and seriously deteriorated pavement, respectively. An Illinois Center for Transportation (ICT) dynamic SRR (D-SRR) model was introduced that considers dynamic loading effect, while provides a quick assessment of SRR.

Graduation Semester

2022-05

Type of Resource

Thesis

Copyright and License Information

Copyright 2022 Xiuyu Liu

Owning Collections