Bridges are critical connectors to transportation systems. They permit transportation of people, resources and services across water, mountains and other impasses. The dire reality is that over 46,000 bridges in the U.S. are categorized as structurally deficient, or in poor condition.

While these bridges are not deemed unsafe for travel, they are in serious need of servicing, repair or replacement. An aging population of bridges and increasing budgetary needs for repairs is making it more difficult to catch up with the declining quality. There are over 617,000 bridges in the U.S. This is why existing deterioration and preventive maintenance have been becoming a larger focus.

It is more common for bridges to undergo reactive maintenance, but proactive maintenance can help reduce some dangers. Proactive measures aim to anticipate failures and stop them from occurring using condition-based monitoring. Four engineering teams are responsible for the safety and inspection of U.S. bridges: structural engineering, hydraulics and geotechnical engineering, bridge and tunnel safety inspection, and structures management and preservation.

There are many failure modes common for bridges. Extreme environmental loading that is currently being exacerbated by climate change quickly deteriorates bridge structures. High physical loads that are growing year over year are also repeated stressors on these systems. Bridges are vital for economic success in an area, so bridge collapse is not only an obvious detriment to human lives, but also human’s way of life.

For among these reasons, researchers are beginning to use new tools, such as AI, to find bridge defects and effect maintenance and repairs.

Structural health monitoring

Bridges need to constantly be under evaluation due to their frequent heavy loading and challenging environmental factors. Expensive resources, devastating infrastructure loss and the utmost consideration of human lives can be at risk. As many as 5 billion trips are taken over bridges across the United States every day. About 170 million of these trips are over structurally deficient bridges.

Structural health monitoring (SHM) detects damages in real-time to assess conditions through sensors, data processing and evaluation. Various monitoring processes have been used: vibration-based, strain-based and acoustic emission. Vibration-based is among the earliest to have been utilized. Reliable sensors and difficulty understanding the data is a frequent challenge.

SHM involves both physical state and structure function status of a bridge. Diagnostic and prognosis can be conducted to determine defects and estimate length of time to failure. Common challenges to these processes include data analysis and best solution decision-making. It has been established that machine learning algorithms are crucial for pattern recognition and data processing.

In the past, machine learning required supervision in relation to bridge data. A slew of examples in a database of bridge behavior was needed for the machine learning operations to learn and be trained upon. These databases were fed from limited finite element models and lab tests that were devoid of the accuracies and nuance of real-life scenarios. Artificial neural networks help to simulate more real-world applications, but this is often slow to be adopted.

Neural networks and AI advancements for bridges

Australian Catholic University (ACU) has pioneered an advancement for AI and neural network usage in structural health monitoring in the realm of bridge engineering. This university is ranked first in Australia for research quality and 24th globally. They are ranked first in Australia and fifteenth in the world for scientific impact across all disciplines.

Their advancement was tested by a multinational team, led by Niusha Shafiabady, an associate professor for computational intelligence at ACU. She is internationally recognized as an expert in this field. This research was tested on the Chumchup, Gocong, Ongdau and Ongnhieu Bridges in Vietnam.

Their study proved beneficial real-time monitoring of structural health of bridges with neural networks and AI. Neural networks are designed after the human brain and nervous system. They are able to make their own decisions with high autonomy given their ability to mimic human function. They are also capable of learning from experience and adapting. They are built of numerous machine learning algorithms.

The neural networks in ACU’s study operated within the bridge infrastructure as well as with the AI algorithms. This was to make decisions separately. This allows for iterative decision making: one neural network makes a decision before it moves onto another for a subsequent decision. This gives AI the chance to come up with a desirable answer.

The 'loss factor'

The basis of this research was on the loss factor. This is the process of energy dissipation over vibration states. It represents the energy that is lost to heat or friction. It can be likened to the energy dispersed from a trampoline jump; some of the energy does not return to the jumper. This energy loss results from structural deficiencies in bridges. The bridges were tested for vibration patterns with this loss factor consideration.

Energy dissipation was then separated into three groups: structural response, defect indicators and noise interference. Three scenarios were created and modeled in order to test the loss factor. The first scenario involved extremely heavy over-the-limit loading. The second circumstance was for light load and light traffic. The final condition was varied traffic types with a heavy loading. This study utilized multiple real-life applications, a much-needed factor in comparison to historical difficulty in real-life modeling, as aforementioned.

Through multiple varying AI algorithms, the loss factor and data on the structural health of the bridges was obtained and was compared to analyze shifts from each scenario. When the loss coefficient changed, early structural changes could be detected as some form of fatigue or damage.

Preemptive maintenance for bridges

The main use for these neural networks and AI algorithms in relation to a bridge’s structural integrity lies in the ability to conduct effective preemptive maintenance. While the detection of defects in bridges is highly advantageous, a diagnostic framework to help prevent fatal failures could be life-saving. Proper detection of defects and appropriate maintenance can help avoid catastrophes.

These AI-driven maintenance plans are poised to help extend life spans of bridges, minimize costs for repairs and bolster public safety. The structure of the neural networks is capable of improving its own accuracy over time by learning and adapting. This minimizes the need for human training or supervision. This positions bridge infrastructure at a significant juncture where intelligence and self-monitoring can help raise safety and reliability without the need of increased man-power or dollars.

Proactive maintenance heightens reliability, which is imperative for effective transportation networks. Cost savings can be realized, echoing a much-needed benefit to the Federal Highway Administration, Department of Transportation, and Federal Emergency Management Agency. Identification of safety issues prior to their occurrence should be a mandatory feature of bridge design to hopefully reduce unnecessary deaths, infrastructure losses and economic hindrances.

A shift to bridges being repaired more regularly and maintained properly can expand their lifespans. Sustainability can also be recognized since original bridges can be preserved and protected. Resources can be optimized for repairs on an as-needed basis instead of for emergency scenarios. Opting for a proactive response to bridge safety is the best recommendation and neural networks and artificial intelligence are paving the way to safer travel.