Condition forecasting of road pavements
Existing algorithms used to forecast road pavement condition are known to have their limitations. In order better predict pavement condition, Machine Learning can be used on the enormous amounts of condition data that is currently available. It is also possible to link the data to other datasets that will impact upon the condition of the pavement e.g. traffic, weather, etc. to further improve prediction accuracy.
This project linked tables from the HAPSM network data, pavement condition data and pavement construction data to improve queries, and exploring the available data in order to train a predictive model based on correlating variables.
Improved condition forecasting will enable better understanding of future asset maintenance requirements, and possibly save money on unnecessary maintenance.
Automated analysis of images of road infrastructure
Detecting cracks from forward and downward images is currently performed manually and is a time-consuming task prone to human error. Machine learning offers a way of performing the task automatically. This is not a trivial endeavour, however, and although a method has been developed that gives 95% accuracy, further work is necessary to fully understand the image data and how best to process it efficiently and with greater accuracy.
This project is all about developing a capability to apply machine learning to images. Firstly, choosing a machine learning approach, then understanding the steps of machine learning from creating a training data set, tuning its parameters to optimise the learning without overfitting, and finally, how to apply knowledge to test and improve accuracy.
Rapid rendering of defect from road images will mean that network operators can make a targeted and timely maintenance intervention to reduce the cost of infrastructure repair.
Published Project Report: PPR896