The world has an estimated 20 million paved miles of road, 60 thousand miles of runways, a million miles of railroad track, 800 million utility poles and 350 billion square feet of paved parking lot globally. Drawing data from various public sources, the total annual maintenance cost of these assets is likely half a trillion dollars.
Add untold other so-called Big Infrastructure buildings: bridges, dams, pipelines, rooftops, and countless other assets, and that big number grows much, much higher. As Big Infrastructure grows in size, volume, and variety so does the demand for its’ constant attention. The ways we give that attention or maintenance need to be rethought, however, and fortunately, machine learning is now helping us with that rethink.
The seemingly mundane but vital task of maintenance has been with us since humans have been building things. Trained inspectors know many things and, among them, the rate at which the thing they’re inspecting is degrading versus the rate it is supposed to degrade. For example, unexpected increases in traffic is one important reason some roads are in such poor condition, as the builders did not, and nor could not, have predicted traffic levels decades in the future.
No two things are built with all the same materials, in the same way or at the same time. In addition, the same assets are not always used the same way either. If that true every pair of Nike running shoes would wear out in exactly the same way and at the same, which they don’t. Again, in our world of roads, a town with only a 100 mile road network may have had dozens of different contractors, using many different materials applied in different ways over several decades.
Since no two things are built with the same materials, in the same way, or at the same time, Big Infrastructure requires significant training. In fact, ‘inspection’ is really only assessment where the viewer is looking for features, anomalies, and problems while also drawing conclusions about its state and possible remedies. Yet, inspection practices have changed very little, and often human viewing is the only method.
Thus, the biggest challenge of all when inspecting Big Infrastructure is size. Infrastructure like pipelines, utility poles, and the like are big, vast, and require a close and considered look. Yet, the devil is always in the detail. For example, those hundreds of thousands of bridges around the world carrying across ravines, rivers and roadways all need someone to go look – closely and carefully – and see how they’re holding up.
RoadBotics has learned a few things about the challenges of inspecting Big Infrastructure. The company uses AI and standard smartphones to inspect road surfaces and other assets accurately and inexpensively for private and public organizations across the US and around the world. What has been learned from this, arguably, is transferable to any inspection regime, particularly those short on people, money, and time:
- Human inspection can be tedious. The more things needing visual inspection and the more varied the things needing to be identified means the higher the chance meaningful details will be missed. Machine learning does not fatigue. Quite the contrary, in fact. It loves lots and lots of data. The more data you feed it, all things being equal, the smarter and more discerning it gets.
- Human inspection can be inconsistent. ‘Baseline’ or initial ground-truthing of Big Infrastructure provides a necessary reference against which the condition of that asset can be compared over time. Regrettably, many inspection regimes lack continuity over time and, as a result, the baseline shifts are assessment and reassessments are made. In the process, valuable information is either lost, mischaracterized or worse. Machine learning is tunable yet tenaciously consistent and completely transparent.
- Human inspection can be hard (and expensive) to scale. Depending on the size of the Big Infrastructure needing attention, it cost prohibitive to adequately scale up the inspection, so one ends up making do with what they have. Machine learning is incredibly scalable.
- Human inspection can be too subjective. No matter how diligent and determined a group of highly trained inspectors may be, they will often render different opinions. Sometimes those differences are subtle and in other cases profound. But those differences will always be there with visual inspection, even when aided by precise data collection tools. Machine learning is data-driven.
Does this mean that machine learning is completely replacing humans in inspection? The answer is no, not now and not in the foreseeable future. Rather, what the industry is seeing is that role of large asset maintenance teams, including inspectors, is not made far more potent and far more interesting. Machine learning doesn’t care much about that, but that is a discussion for another time.
Interested in Smart Infrastructure? Come to our Cities Conference
We’ve been hearing for years about the enormous infrastructure needs facing cities, that cities are “disrupt-able” and ripe for change. But what’s often left out of the conversation is how to capitalize on it.
The Cities Conference is a full-day premium event that will help you make sense of the crucial changes coming to our cities, including transit, waste, water, energy, buildings, and social infrastructure.
It’s July 19th in Chicago and will feature senior level thought leaders from the following sectors: Urban Planning, Architecture, Real Estate, Engineering, Transportation, and Tech.
Learn more here.
This article was written in partnership with RoadBotics.
Roadbotics pitched (and won!) at the BuiltWorlds NYC Startup Challenge in 2018. They are changing roadway management with a combination of the latest in machine learning, simple video cameras, and cloud based technologies. Learn more by visiting their website or by checking out their BuiltWorlds Directory Page.