Week 2

Hey everyone! I spent this past week transitioning my focus from the condition of the bridge infrastructure in the US to potential technologies that could help fix this. 

After discovering last week about the extensive flaws currently embedded within existing methods for maintaining bridges, some of which will potentially take decades to overcome and thus jeopardize public safety, I believed that it would be a logical next step for me to examine what AI can contribute towards improving access to the data and conducting analyses on the maintenance of bridges. This week was primarily devoted to learning about foundational concepts (machine learning, data modelling, sensor networks, and AI-enabled structural health monitoring systems). I wanted to make sure I had a solid grasp on all of these before diving into the research literature.

After laying out the basic concepts, I moved into the various academic journals and case studies examining how artificial intelligence is currently being applied to infrastructure testing. Through machine learning algorithms, AI can create very sophisticated computer algorithms for analyzing and predicting potential problems based on large sets of real-time data, coming from a variety of sensors that have been placed within bridge structures. These are meant to continuously monitor changes in conditions (e.g., stress levels, vibrations, etc.) and will be used to develop predictive systems for predicting when these changes will require some type of maintenance work to be done before anything has become seriously damaged. 

One study I came across applied a machine learning model directly to National Bridge Inventory data and reached a prediction accuracy of 94.23%, while cutting the number of bridges scheduled for inspection by nearly 30%! In another study done in Taiwan, researchers analyzed over 27,000 records of past inspections and concluded that a Random Forest type of model (a machine learning algorithm that makes predictions by building a large number of decision trees and combining their results) would return a prediction accuracy of 88% in predicting which individual components of a bridge were likely to be deteriorating. This outperforms many more traditional types of algorithms in the process, and results in the ability to apply a data-driven approach to prioritizing maintenance on a bridge, thereby allowing for much greater efficiencies to be identified in the inspection and maintenance of bridges.

From everything I’ve seen throughout this research, AI will not take the place of Engineers (I hope), but rather give them more accurate and detailed information. The best way to do this is through a “human in the loop” framework: that’s when outputs from a model are combined with field observations and engineering judgement. That distinction is critical, and it’s going to form the framework for this project. There is no question about whether the technology will be there. It’s whether we can utilize this technology responsibly and effectively as we develop actual infrastructure systems.

Work Cited

Contreras-Nieto, Cristian, et al. “Bridge Maintenance Prioritization Using Ordinal Optimization and Machine Learning.” Automation in Construction, vol. 104, 2019, pp. 243–256.

Jiang, Yi, et al. “Predicting Bridge Deterioration Using Machine Learning: A Case Study on the National Bridge Inventory.” Journal of Infrastructure Systems, ASCE.

Sagi, Omer, and Lior Rokach. “Explainable Decision Forest: Transforming a Random Forest into an Interpretable Tree.” Information Fusion, vol. 61, 2020, pp. 124–138.