(Scroll)
In 2021 a bridge, built 80 years earlier out of reinforced concrete, came in distress.
The old bridge, located in a remote area, was a main connection for traffic between southern and northern parts of Norway.
About 30 sensors had been mounted to monitor its condition, on foundation, columns, bridge deck and joints.
A few days after Easter, the engineer on duty noticed a disturbing development. A physical inspection was undertaken right before the weekend, without indications of anything wrong.
However, readings continued to increase and extensive analytical work during the weekend confirmed the gravity. The situation was rapidly escalating, and a hypothesis was formed what was happening, and where.
An emergency meeting Monday morning led to new inspection. This time the cause was discovered; a main bearing had been crushed, causing loss of all support there.
The inspector immediately used is car to close the lane and called in emergency 24/7 traffic control, an arrangement lasting until the bridge was replaced less than a year later.
© Vegvesenet
A manager from the Norwegian Public Road Administration asked whether the development might have been foreseen. Could the risk of a very serious incident and the excessive cost of emergency action and traffic disruption have been avoided?
Trond Michael Andersen
NPRA Director of Technology,
Presenting the results from the machine learning anomaly detection was difficult. The underlying physics was abstracted into statistics, making it even more difficult to interpret. Explanations were missing, and so also the ability to conclude on the data.
It was the seed for Aidalos. We wanted something that:
And not least, we wanted to a have an efficient assistant that could have assisted in the data fetching and analysis and reporting that weekend. We wanted a specialist on 24/7 duty, patient, diligent, and low cost.
To make this long story short:
We wanted what we now have!