Using Physics Informed Machine Learning to Bridge the Knowledge-Data Gap for Pavement Modeling
The United States has over 3.9 million miles of roads and the task of maintaining a record of pavement wear and tear for every mile of road is a substantial task. The field of predictive modeling has developed to address the challenge of maintaining road conditions by reducing the uncertainties associated with a road’s ability to handle stress and offering opportunities for timely interventions and fostering sustainable road planning and management. There are various ways of mechanistically assessing the stresses and strains in pavements to prevent future distresses. However, mechanistic analysis is usually time consuming and requires extensive user knowledge. Data based (empirical) methods on the other hand, while requiring little user knowledge, are usually limited to the extent of the data source. The aim of this study is to combine the ease of use of data based methods with the accuracy of physical laws in mechanistic analysis. A physics informed neural network (PINN) is introduced to bridge the gap between these two methods. There different models were examined in this study: a linear regression model, a neural network model, and a PINN model. The performance of the models are combined to investigate their 1) accuracy, 2) generalization ability and 3) interpretability. An efficient choice of a model will improve efficiency in road maintenance and reduce the associated costs for the respective governmental agencies.
Research Area | Presenter | Title | Keywords |
---|---|---|---|
Probability, Statistics, and Machine Learning | Waghe, Shreyas | Fair Machine Learning (0.827586), Machine Learning (0.923077) | |
Computer Science | Berduo, Alan Jesse | Machine learning | |
Computer Science | Shaikh, Aymaan | Machine learning | |
Probability, Statistics, and Machine Learning | Rizvanov, Timur | Machine learning | |
Mathematics and Statistics | Burns, Benjamin | machine learning |