Artificial Intelligence in Geo-Engineering

The past: A failure to communicate

In the past, geophysicists have struggled with communicating the benefits of geophysical investigations for geotechnical engineering projects. Though the benefits of cheap, non-invasive site investigations may be obvious to them, convincing project owners to try more innovative site investigation methods has been challenging.

Many potential clients only had confidence in direct measurements acquired from expensive boreholes and distrusted indirect and less precise geophysical surveys. Even for engineers that had experience with geophysical surveys in their projects, efficient tools for communicating results were lacking. Maps, cross-sections, and PDF reports were too cumbersome for many clients to use in their planning and design workflow. Only specialized geotechnical engineers have been able to extract the true value from complex geophysical models.

Innovations to overcome roadblocks

At EMerald Geomodelling, machine learning techniques have helped to overcome this skepticism. By combining the strengths of geophysical and geotechnical data in a quantitative manner, the resulting interpretation products are both better quality and also produced at much lower cost.

Industry at large has started using 3D software as an important planning tool. Gone are the days that geophysicists can just deliver static reports. EMerald adapted by delivering 3D models that could be imported directly into engineers’ planning software. This ensures that results of geophysical investigations are considered when making major engineering decisions.

Together, these changes have helped EMerald break down the barriers separating engineers and geophysicists, thus building bridges between siloed disciplines within applied earth science.

Future challenges: Finding consensus and finding clarity

There are two main challenges on the horizon for the geotechnical industry: a lack of standard formats for 3D subsurface data, and a poor understanding of the limitations of AI.  

EMerald usually delivers models in a specific format for each client. Modifying workflows to accommodate each new requested format is time-consuming. In the wide world of geotechnical engineering, there’s little consensus on how to structure these data sets. This hinders communication between organizations.  BIM has become a standard for structural engineers working above the ground. Thus far, it seems to be a promising direction for subsurface work.

Lastly, though AI tools have delivered impressive results, experienced analysts with discipline-specific knowledge are still key for quality control. This is because we still only have limited tools for auditing how AI models come to a conclusion. As machine learning is applied in more novel and complicated problems, need to be more wary and critical about the output.

‍