The Vision

Our vision is that AsphaltMine becomes a perpetual asphalt community project that is continuously collecting new data, building new tools and updating the existing ones.

The Objective

The project’s objective is to enable predicting laboratory test results from results at a lower scale. For example, predicting the rutting test result from the component material test results and their proportions.

As the first step, we have built a structured web database for uploading asphalt rutting test results and all the mix component material test results. You can see the structure here.

To ensure that the database can be used across different European countries, we built it according to the EN standards.

Depending on the requirements of each partner, an individual data sharing agreement will be prepared to ensure your data is handled in the way you prefer.

What results are we collecting?

In the first stage of the project, we would like to collect wheel tracking test results (all methods), and indirect tensile strength results results along with the respective mixture designs and component material test results.

The more results we collect, the better the prediction tool will be. We are open to your suggestions for other test methods we should include in the database.

Who is participating?

The data will be contributed by various institutions, including private companies, research labs and universities. See the current participants and join AsphaltMine here.

How will the data be shared?

The web-based platform allows to fully customize the access rights.

If you not wish to share the results publicly, they will remain undisclosed forever.

Undisclosed results will only be used for building the prediction tool but will not be shared with public.

See more information about the access of the data and the machine learning model below.

It is important that only reliable results are collected.
Unsatisfactory test results (samples that don’t pass the requirements) should also be included in the dataset. You can choose not to disclose them to the public.

How will the database work?

The database will be built using a platform called “Configurable Data Curation System“. This system is built for data curation, it is free and open-source.

The asphalt test results will be stored in a XML tree-like format and organized in a way that allows simple filtering. Downloading and converting back to a Excel format will also be possible.

How to upload the data?

The user can upload the results in two ways:

1. Using a web form.
2. For standardized test result protocols (pdf or Excel files), we can develop a plug-in for automatic uploading. This has to be done on a case-by-case basis.

Where is the data stored?

By default, the data will be stored on a server in Switzerland, but the database functionality allows remote storage if needed.

We will prepare a contract where the data handling is described.

How will we build the machine-learning tool?

Once we have collected enough data in the database, we will train machine-learning models that allow predicting laboratory test results.

Initially, several algorithms will be employed until we identify the one with the best accuracy. We might also augment the tool with established physical materials science relationships.

What is there is not enough data?

This is a project that has not been done before so we expect difficulties and limited data availability is among the top concerns. Here are some of the challenges we expect and how we plan to address them.

After creating the machine learning tool, we will make it available through this website. Let’s call it the “Test Result Prediction Tool”.

You will have to enter information about the mixture design. Then on a click of a button the prediction tool will calculate the expected test result, for example the rutting test result.

What can the prediction tool be used for?

Initially, a rutting test result prediction tool will be developed using the collected results. The tool can be used to:

• Predict the test results from the mix recipe.
• Reduce the testing costs and save time since multiple recipes can be virtually tested before committing to a physical lab test.
• Optimize the mixture recipe to reduce production costs and improve performance.

Prediction tools for other test methods will be added at later stages.

How will the tool be accessed?

We will develop a web-based platform for using any tools we develop. It will be available through this very AsphaltMine website. We will make sure the tools have a user-friendly interface.

Who gets the right to use the tool?

The prediction tool access is fully customizable. Below is our initial offer for the data and prediction tool sharing principle for two prediction tool versions:

1. Public access prediction tool built using the open access data. The tool will be freely available on the AsphaltMine website.
2. Contributor access prediction tool built using the entire data, including undisclosed data. All the contributors whose data is used for training the machine-learning model will have access to the “Contributor Access” tool. After 3 years, the RV tool will become available for public access.

Who is sponsoring the work?

The initial work for the creation of the database creation is funded through the Swiss National Science foundation project No.213163 which will last for 4 years (link). It is part of a collaborative project with University of Antwerp and Vienna University of Technology.

I am open to jointly exploring other funding opportunities for further work on the idea.

Who is doing the work?

Martins Zaumanis is the project principal investigator. Scientist at Empa, Switzerland.

Mohammad Abbas will do the main work on the database and the creation of machine learning model. Postdoc at Empa, Switzerland.

Marko Oreskovic is volunteering his knowledge and energy to help the project succeed. Assistant professor at the University of Belgrade, Serbia.

Partners of the SNSF research project from the Antwerp University and TU Wien are contributing their expertise and resources to the building and population of the database.