Machine-learning of instrument responses predicts aircraft impact on airfields

Researchers are looking into ways to sustain well-maintained and functional airport pavement systems, while reducing the approximately $4 billion spent yearly on airport pavement repair and rehabilitation.

One way is to understand pavement response to the combined effect of aircraft landing gear and environmental conditions in the field.

The U.S. has approximately 9,000 paved airports with approximately 650 million square yards of paved runway surfaces. The condition of airport pavements is critical to providing safe mobility to almost 800 million passengers traveling in the U.S. each year.

Traditionally, airfield pavements are analyzed using mechanistic approaches including layered elastic theory or finite element analysis. These approaches must be built on assumptions that simplify or neglect variables in pavement-tire interaction that are too complicated to be represented in mechanistic equations.

Although recent research findings have shown that FE analysis can be extended to include many variables omitted by the conventional pavement analysis approaches, it becomes computationally expensive to adopt in the airport pavement design frameworks.

With improvements in data storage and collection, machine learning is gaining momentum in the field of engineering. ML is the development of algorithms that learn from data through repetition and produce reliable, repeatable decisions and results.

Its ability to capture complicated relations in the data make it a promising candidate to address challenging cases in engineering that would be difficult to solve with traditional mechanistic approaches. Additionally, unlike the mechanistic approaches, it provides computational efficiency because it does not require rebuilding the model (i.e., training) with each new input.

In this study, researchers including doctoral candidate O. Erman Gungor and Professor Imad Al-Qadi, in collaboration with Navneet Garg of Federal Aviation Administration, employed a Support Vector Machine to develop ML models to compute temperature curling strains (concrete pavement slab bending due to environmental loading) within the pavement, and bending strain at the bottom of plain cement concrete slabs under Boeing 777 loading, as shown in Figure 1.

The data was obtained from instruments and sensors installed in Taxiway-Z at the John F. Kennedy International airport pavement during construction. The spacing of sensors installed allowed for response measurement for this type of aircraft gear geometry. Data from an Airbus A380—a double-deck, wide-body, four-engine jet airliner—was also captured.

The results showed that ML-based, data-driven models are promising and capable of predicting pavement responses with high accuracy. Temperatures within the slab (thermal gradient) and tensile strains resulting from PCC slab curling were predicted with very high accuracy.

In addition to predicting the responses with such high accuracy, the ML models also revealed the importance of concrete age on strain and temperature distribution within the slab. Furthermore, air pressure was an important variable for temperature prediction. These two variables (concrete age and air pressure) are currently not considered by the state-of-the-practice airfield pavement design approaches.

The research team also developed ML models to predict the pavement responses to Boeing 777 loading. To obtain the responses, a signal processing framework was developed to extract the peak responses associated with Boeing 777 loading. This signal processing algorithm helps reduce the size of the data by using only the relevant data. After extracting the responses, SVM-based prediction models were generated. These models computed the responses in real-time.

“The ML-based data-driven modeling is a promising alternative to time-consuming conventional pavement analysis approaches,” Al-Qadi said.

Gungor emphasized that the developed ML models were computationally efficient and could capture the variables that otherwise were simplified or neglected by the conventional pavement analysis techniques.

Garg, a UIUC alumnus and program manager at the FAA’s National Airport Pavement and Materials Research Center said, “Incorporation of data-sets from various airports located in different climatic zones may pave the way for fully data-driven pavement design frameworks in the future.”