Research bridges big data, infrastructure needs

As Canada’s rapidly aging infrastructure continues to erode, often out of sight of those charged with overseeing it, maintenance budgets and residents from coast to coast to coast face a significant threat from the evitable fallout.

“Most of Canada’s infrastructure, like bridges and dams, were built around the time of the Second World War. They’re near the end of their life cycle,” Engineering professor Ayan Sadhu explained. “Our infrastructure is like a D-plus – which is bad.”

Sadhu’s work, however, looks to gather better data into this critical infrastructure, all in an effort to signal problems sooner, potentially saving millions of dollars and reducing the threat to those living in harm’s way.

Take bridges, for example.

According to a Statistics Canada and Infrastructure Canada report in 2018, only 56 per cent of Canada’s roads, bridges and tunnels were rated in good or very good condition.

Given the fact a vast majority of the country’s 47,000-plus publicly owned bridges are well into middle age, a major infusion of resources will be required in order keep our national transportation system in good health.

Existing bridge-monitoring methods are outdated, relying on prescheduled visual inspections that are highly subjective, vary between maintenance teams and are error-prone in damage detection. Recent advances in sensing technology, however, have provided opportunities for engineers to recognize defects in bridges remotely thanks to sensors.

The problem is those sensors are costly and occasionally inefficient.

“Consider the structure as a human being. Like a patient, we monitor infrastructure using different sensors and use this data to identify the damage,” Sadhu said.

“When we say ‘inspection,’ most things are happening with visual inspection. But some places are not accessible. It’s hard to do inspection. Most of the bridges are pretty busy and, if instrumentation needs to be done, you need closures. No one is happy with that. You need something that is remote and autonomous, where it supplies you with data remotely.

“Usually, sensors are placed in different locations. One of my research focuses is how can we get the information we need using fewer number of sensors.”

Sadhu is looking to improve this method by getting better data with fewer sensors at a lower cost for governments.

For example, close to 70 per cent of the costs of monitoring a structure is in the cables connecting the sensors. Fewer sensors means fewer cables which means lower costs.

All this can be done without a loss in data – in fact, the data might even been stronger.

“It’s not about the data; it’s about the information. That’s the challenge,” he continued. “We use artificial intelligence to extract information. If the information is the same, and we did it with limited sensors, we are actually solving the problem.

“It’s not just the number of sensors but rather the information we get from those sensors.”

Sensors location depends on a number of factors, like temperature, location, traffic volume and environmental conditions. These factors then determine the location and sensitivity needed in the sensors.

Using big data in combination with artificial intelligence, Sadhu has developed signal and imaging processing algorithms to compress the data generated from the sensors to isolate only needed evidence.

“With the image data we get from surface damage, such as cracks, corrosion or fatigue, we can actually train our algorithms on those images to tell us if a crack is due to fatigue, allowing us to determine what, and where, maintenance is needed,” he said.

Sadhu added with the popularity of hybrid construction – the combination of different materials such as wood, concrete and steel – it provides more reason to ensure structure safety going forward.

“You need to see if they’re structurally doing well. This is all about public safety,” he said.

/Public Release. This material comes from the originating organization and may be of a point-in-time nature, edited for clarity, style and length. View in full here.