AI prediction models take flight for airfield pavement

An Illinois Center for Transportation team, led by director Imad Al-Qadi, is working to predict the responses of airport pavement under different landing gear configurations more quickly and cost-effectively.

The Federal Aviation Administration project, sponsored by its Airport Technology R&D Branch, is developing machine learning models, a branch of artificial intelligence that learns and predicts trends from data, to reduce the time needed to simulate real-world airfield pavement responses.

“Most of the design procedures or analytical procedures that exist today simplify things, so we’re not able to capture the actual material behavior, the complex loading conditions and the climatic effects,” said Navneet Garg, program manager of FAA’s National Airport Pavement and Materials Research Center and a University of Illinois Urbana-Champaign alumnus.

“If you want to consider all these factors, then you have to go to 3D finite element analysis, and then the analysis becomes very time-consuming and expensive,” he added.

The researchers aimed to reduce this time needed to run airfield pavement models by applying machine learning with physics to FAA data and comparing their results to those of finite element models and field instrumentation responses.

Developing Finite Element Models for Airfields

Al-Qadi’s team first turned to finite element analysis, which breaks an object into hundreds of thousands of smaller elements and models how those smaller elements perform under loads from traffic and the environment.

“By dividing a pavement structure into thousands of connected elements, we simplify and accurately calculate stresses and strains of a complex system,” Al-Qadi said. “Each element allows us to know the responses at any point we’re interested in.”

Their developed models, adapted from ones Al-Qadi designed for highways, predict the responses of airfield pavement to various factors such as the number and spacing of landing gear wheels, tire pressure and aircraft weight.

They verified the accuracy of the models with data from FAA’s accelerated pavement testing machine, the National Airport Pavement Test Vehicle. The machine tests full-scale pavement sections over a construction cycle, from construction to removal, and the sections are equipped with sensors during construction to gather data.

FAA has completed accelerated testing of nine construction cycles since 1999, each with terabytes of data that can be used in FAARFIELD, FAA’s software for airport pavement design.

 “The first step was to organize and reduce the airfield pavement response data and to provide them to FAA in a way that is manageable, easy to get and able for them to use in the future,” Al-Qadi said. “Then, we compared the finite element output to FAA field response measurements, and the results were comparable.”

 

 

Harnessing Physics-Guided Artificial Intelligence

Finite element models, however, are time-consuming, taking around six to eight weeks to run with a supercomputer.

A solution? Apply physics-guided artificial intelligence, an approach that combines the laws of physics with predicting trends, to FAA’s accelerated testing data to feed and predict responses under different conditions.

Al-Qadi and his team also used a database they’ve developed over the past three decades of hundreds of finite element simulations for highway and airfield pavements.

“What takes like six to eight weeks to do now can be done in less than 15 minutes,” Al-Qadi said. “This was a breakthrough on dealing with data, especially because we are working with a huge database with hundreds of thousands of elements and systems, each with their own responses.”

Garg hopes to incorporate the developed AI model into FAA’s existing tools and procedures, such as FAARFIELD and PANDA, an in-development tool for advanced pavement analysis.

 “My main objective is to include or implement these machine learning models into FAA’s pavement design standards,” Garg said. “It’ll make predicting the remaining life of the pavement much more reasonable, so airports will be able to design how to rehabilitate their pavements once they know the remaining life, making it more cost-effective and less time-consuming.”

Al-Qadi credits the success of the work to his graduate students — Qingwen Zhou, Egemen Okte, Aravind Ramakrishnan, Lara Diab and Johann Cardenas — former ICT research scientist Angeli Jayme, and Navneet Garg, the project manager.

“This project tackled a challenging problem,” Al-Qadi said. “But with a clear vision and exceptional students who are not only talented and committed, but also genuinely motivated to solve that problem, we were able to help the FAA identify a solution.”