Putting eyes on our roads

29/05/2023

Automatic road fault detection system

Building and maintaining a large network of roads in New Zealand is just one of Downer Group’s longstanding transport services. To keep roads safe, efficient and reliable, Downer Group’s road network solutions are underpinned by asset modelling, pavement engineering, and data collection and analysis. Innovative systems and tools are particularly valuable for predicting work programs.

One such example is a fault detection machine learning solution, built by xDNA, is expected to streamline road quality assessments in New Zealand.

At a glance

• Track and analyse thousands of kilometres of road networks through automation.

• Video analytics is removing people from potentially dangerous on-road situations.

• Data driven insights optimise unplanned and planned maintenance.

• RoadDNA is an exciting linear asset management tool.

Problem

Assessing the quality of roads requires physical inspections and ad hoc user reporting. It’s inefficient and potentially dangerous for both road users and inspectors. Even a simple assessment of potholes – their size, depth and location – can involve road closures and disruption. 

Once an inspection is complete, with photos and assessment notes collected, a worker must then determine what repairs are required, gathering specific instructions about the location and materials. Repairs also need to be triaged based on safety and budget.

The Downer Group sought a safer and more efficient way to keep roads in good condition.

Solution

Downer Group started collecting video footage in New Zealand via vehicle-mounted cameras. xDNA provided a powerful collaborative labelling solution that accelerates the training phase of machine learning model development. Whilst labelled footage started to come in, xDNA’s data scientists developed several machine learning models which were then operationalised on the Neuroverse platform and RoadDNA was born.

Over several months and with input from subject matter experts, the technology was trained to analyse the road footage and build a catalog of faults. It learnt to look for and correct anomalies . It is built to render its findings on common GIS platforms and supports the establishment of road quality indexes. 

Interestingly, the machine learning aspect means RoadDNA can start to track faults and the impact of maintenance programs and weather over time, allowing for continuous road network improvements to be designed based on data driven insights.

The machine learning model was loaded into the Neuroverse data platform, which can cope with a huge volume of data. This means more footage and locations can be added efficiently, allowing for scale.

Outcome

RoadDNA is being trialled over a 6-month period in an area of New Zealand. The model will need continual optimisation, but in time will become smarter and more accurate. It also has the capacity to extract highly reliable reports across location and repair type.

Now that RoadDNA has been built, it is ready to explore more road networks. It’s a platform that can efficiently assess faults in assets, whether they be state road, rail or private networks. It’s a solution accelerator with untapped potential.

‘RoadDNA is an exciting linear asset management tool. It has the potential to make clearer and faster assessments of all types of assets, saving us time and more importantly, keeping people safer when they travel.’ 

About xDNA

xDNA offers an innate understanding of tech solutions for asset-based business. Innovative tools and techniques augment critical frontline worker tasks. Using smart systems, workers can capture, analyse and communicate data to increase efficiency and effectiveness.