AI and Deterioration Modeling

Status: Candidate/ Concept
Timeframe: Near-term
Potential Funding: NCHRP Full Program


AI is coming – are we ready? With the MAP-21/FAST Act legislation, and the renewed emphasis on Transportation Asset Management Plans (TAMPs), projections made by management systems will come under increasing scrutiny as agency executive leadership is asked to make large scale funding decisions based on these projections. This scrutiny as well as the inherent complications in predictive modeling of asset deterioration, presents an opportunity for the use of Artificial Intelligence (AI) in this type of analysis.

AI is becoming ubiquitous in the realm of automation and pattern recognition and shows promise in improving predictive modeling for infrastructure managed by highway agencies. Because data collected over time is especially valuable for deterioration modeling, it is very important for agencies to start collecting the right data, and putting in place the right quality control, as early as possible so that this data is ready for immediate use as more research into AI techniques for predictive modeling is conducted.


This research project would aim to develop a Primer or Guidance document to help agencies tasked with managing infrastructure (including pavement and bridges) to assess their current data, data collection processes, and data needs to best position them to be able to take advantage of burgeoning artificial intelligence techniques to develop increasingly accurate predictive models regarding their infrastructure.

Research Plan:

The quality of data is extremely important – “garbage in, garbage out” - and quality of data in terms of accuracy and precision is already getting much needed attention. However, while many agencies are actively improving collection of accurate and more data, collection the right quality data for accurate and precise prediction requires an additional level of scrutiny.

Collection of more accurate and precise data will undoubtably increase the accuracy of predictions, accurate predictive modeling also relies on understanding the underlying variables that affect the predictions. For example, variables that might affect the structural deterioration (for instance in the next time period) of an infrastructure element such as a pavement management section, might include:
- Structure information such as layer thicknesses and materials
- Environmental conditions such as temperature means and variation, rainfall etc.
- Load information such as traffic and truck traffic
- Current condition such as current cracking, rutting and roughness information
- Current condition such as layer properties and structural strength
- Information on previous maintenance, rehabilitation and reconstruction actions

Similar attributes would be considered significant variables for deterioration prediction in bridges, and this would also apply to many other non-bridge, non-pavement types of infrastructure assets.

Statistical analysis of this type of data for predictive analysis purposes is not new and Analysis of Variance (ANOVA) techniques have been used in this area for decades. However, with the advent of automated data collection techniques and with the quantity of available data growing at a considerable rate (so called ‘Big Data’), various types of AI such as artificial neural networks (ANNs) and deep learning techniques, are beginning to supersede some of these traditional statistical techniques. The ‘training’ portions of these techniques will require accurate and repeatable data as well as information on significant variables.

In addition, one the most valuable aspects of AI is the ability these types of techniques to continuously learn and improve. In this respect, it is again very important for agencies to understand how this learning could be accomplished, not just initially but continuously over time, using processes that involve continuous updates (e.g. through crowd sourcing). Agencies would therefore benefit considerably by having guidance available to help them set up their data capture and governance techniques to best benefit from AI modeling, training and continuous learning in the future.

Ideally, an agency would collect data that has the necessary attributes to facilitate an AI analysis and have processes in place that would allow continuous learning such that predictive modeling for the agency would continue to be trained and improved as the AI continued to learn. The current reality is such that condition data that is being collected may not be easily utilized in an AI analysis. The consequence is that the complicated decision-making process that highway agency executives depend upon may not be producing the level of accuracy in condition and funding projections that is required to make funding decisions in their investment strategies.

Project Budget:

Topics: Analytical Tools and Models, Data Collection Tools/Technologies, Management Information Systems
Subjects: Inventory and Condition Assessment, Life Cycle Costing/Economic Analysis