Browse Projects > Detailed View

Extracting Information out of Big Data for Pavement Management


Recently there has been growing interest in the analysis of very large databases (so called “big data”) using advanced statistical methods and “computational intelligence” tools. These approaches include machine learning techniques such as artificial neural networks and other pattern recognition methodologies, evolutionary optimization algorithms, and fuzzy logic models. Examples of the application of these tools are now all around us: facial recognition in Facebook photographs, intelligent search results and recommendations, targeted advertising and the many other “smart” things our computers, phones and even watches do. These techniques differ from traditional statistical approaches in that they focus on learning patterns and correlations from the data, without real regard for causative relations, and do not focus on individual variables. Therefore, engineers typically consider these tools as an opaque “black-box” approach to determining outcomes.

While many pavement mangers might feel overwhelmed by their existing data collection activities, pavement management system (PMS) data collection has yet to truly enter the realm of big data. However, with the introduction of 3D pavement imaging technology, continuous deflection measurement devices, and the possibility of crowd-sourced data from road users and connected vehicles (especially self-driving cars), the era of big data in PMS is rather imminent. Traditional PMS software typically summarize a few critical performance indicators on large management segments, and use simple (and transparent) performance equations and tools such as decision trees. Big data analytical methods have the potential to impact all of areas of pavement management. Examples of these impacts might be:

(1) using machine learning to categorize cracking from surface images;

(2) identify pavement distress from vehicle mount accelerometers;

(3) determining optimal treatments directly from surface images;

(4) predicting future performance at a much more granular level than management segments; and

(5) using evolutionary optimization heuristics replacing conventional exhaustive methods for optimization of treatment projects

The downside of the learning approaches is that they could require existing examples for training, and thus are tied to replicating the existing practice, which itself could be a sub-optimal approach. In addition, neither learning nor evolutionary methods do not provide a traditional and transparent presentation of how various outcomes were obtained, so it may be difficult to build confidence in the results without proper education of pavement engineers in computational intelligence theories.


The goal of this research would be to encourage and enable the use of these techniques within the pavement management community. This would be achieved through the following specific objectives:

Obtain suitable example data from existing collection efforts (e.g. 3D surface profiles for a full network, decision making examples from historical PMS data)

Apply a variety of big data analytical techniques to this data to determine ease-of-use, effectiveness, applicability, etc.

Report on how the various techniques performed in relation to extracting useful information out of PMS data, and

Set future research direction in how pavement management systems and software might need to be altered to make use of these techniques.


Existing pavement management systems tend to be rigid in their approach to decision making, and the outcome of the PMS optimization is often heavily dependent on only a handful of critical variables, such as a trigger level for IRI. Big data analytical techniques have the potential to enable PMS systems that are more resilient to missing data, can capture field variability in pavement properties, less dependent on hard trigger levels, have better performance predictions and, overall, produce better and more cost effective treatment plans.


The required tasks for this study have been tentatively identified as follows:

1. Conduct a literature review on the applications of computational intelligence in extracting useful information out of Big Data for pavement management

2. Identify analysis needs for extracting information out of big data such as error handling, addressing missing data, pattern recognition, optimization, quality control, etc.

3. Investigate application of computational intelligence techniques in filling the analysis gaps identified in the previous task

4. Investigate visualization tools that can assist pavement engineers in understanding the problem-solving capabilities of computational intelligence tools

5. Conduct a pilot study to demonstrate two or more of the identified applications through a real agency PMS database

6. Prepare guidelines for educating agency staff on such applications and implementing such computational tools in agency PMS

7. Identify Future Research Needs


The implementation of this research would be in form of guideline document(s) detailing how big data analytics might be implemented in pavement management – intended to educate researchers, vendors and practitioners to the possibilities and spur further research and development, and a report detailing how future research efforts might be focused to obtain maximum benefit from not only these techniques, but also the big data collection efforts required to drive the process.

Sponsoring Committee:AKT10, Pavement Management Systems
Research Period:12 - 24 months
Research Priority:High
Date Posted:04/06/2016
Date Modified:04/18/2016
Index Terms:Pavement management systems, Big data, Data mining, Learning (Artificial intelligence), Optimization,
Cosponsoring Committees: 
Data and Information Technology
Maintenance and Preservation
Planning and Forecasting

Please click here if you wish to share information or are aware of any research underway that addresses issues in this research needs statement. The information may be helpful to the sponsoring committee in keeping the statement up-to-date.