Welcome to the first CORE.AI Newsletter!

The CORE.AI Newsletter is a quarterly update on the happenings within Thornton Tomasetti’s Artificial Intelligence and Machine Learning initiative.

Introducing CORE.AI

CORE studio is excited to welcome Dr. Badri Hiriyur to the team as our incoming Director of Artificial Intelligence & Machine Learning and will be leading the AI initiative.

Artificial Intelligence (AI) and Machine Learning (ML) are popular buzzwords these days because they describe a family of algorithms that have enabled various cool technological achievements such as robots, self-driving cars, Alexa, and AlphaGO. CORE Studio has also been working on AI / ML technologies – with a specific focus on AEC applications – for some time now. We have consolidated these efforts, included an external CORE lab R&D project and brought in additional resources to launch a new initiative – CORE.AI.

CORE Studio welcomes Dr. Badri Hiriyur to join its ranks in his new role as Director of Artificial Intelligence & Machine Learning. Dr. Hiriyur comes to the studio from the Applied Science practice of Thornton Tomasetti, where he spent many years developing software – used by the US Navy – for high-performance computational fluid dynamics. Dr. Hiriyur also has a strong background in machine learning, deep learning, and robotics and is the creator of T2D2. Dr. Hiriyur has a masters degree (2003) from Johns Hopkins and a Ph.D. (2012) from Columbia University.


What IS AI/ML?

Artificial Intelligence and Machine Learning are arguably the hottest buzzwords in the world right now and most people have no idea what these words mean (p.s. robots are not taking over the world). Artificial Intelligence / Machine Learning is a family of algorithms that enable machines (computers, robots, cars, etc.) to process various kinds of information (visual, audio, data) and make intelligent decisions that help them optimally achieve predefined objectives. In contrast to traditional software programs, which also process information and enable decisionmaking, these algorithms learn to make their decisions by looking at prior data/decisions (supervised learning) instead of being fed explicit rules that govern their decision-making process.

For example, ML algorithms can learn to map various combinations of user-defined parameters (commonly termed features) to a set of output labels (classification) or to numerical output values (regression). ML algorithms also help gain insights into groupings of data (unsupervised learning) and self-improvement over time based on a policy of rewards and penalties associated with its decisions (reinforcement learning).

Deep Learning

Neural networks are a class of machine learning algorithms that use programmatic circuits connecting a series of input parameters (or input data) to output quantities through one or more layers of nodes. Each of these nodes performs a weighted combination of its inputs, transforms the result with a function (activation) and passes it on to the next node. The key to neural networks mapping inputs to reasonable outputs lie in finding the appropriate connection weights. During the supervised learning process, a technique called loss back-propagation is used.

Gill, J. K. (2017, July 21). Automatic Log Analysis using Deep learning and AI. Retrieved March 19, 2019, from https://w w w .xenonstack.com/blog/log-analytics-deep-machine-learning-ai/

Among the many types of machine learning algorithms, deep neural networks (with many layers of hidden nodes between the input and output quantities) have gained increasing popularity in recent times because of the performance gains they made on many classes of problems (e.g. in computer vision, natural language processing etc) that were once considered hard for algorithms to crack. Another reason for their popularity is because they have the ability to work with raw data (e.g. pixels, audio wave-forms) and automatically identify the relevant features that matter, instead of having to pre-process data to extract important features.

CORE.AI Projects

T2D2©: Thornton Tomasetti Damage Detector

What is it?
T2D2 is an application that processes images or video feed from structural inspections to automatically detect visible damage and defects. It includes a geo-localization module which maps detected damage to associated locations on the structure. T2D2 is deployed in two forms: T2D2 Mobile is an app on a mobile phone or tablet, which packages the TensorFlow inference models and enables hand-held inspections on-site. T2D2 Web provides an online inference service that processes camera image feeds and returns JSON/XML metadata describing all the corresponding damage detections. This can be used for drone-based inspections when combined with the location, altitude and orientation data obtained from drone localization and presented to users on a dashboard.

What’s under the hood?
T2D2 uses deep convolution neural networks for classification and bounding box detection which are implemented in TensorFlow – a framework for deep learning developed by Google. These neural networks are trained on a vast database of annotated images collected from prior structural and facade inspections.

Development Status
The T2D2 dev team is currently working on expanding the training data sets to cover various structure and damage types, in addition to polishing up the user interfaces to enable a smooth inspection workflow. A beta release of T2D2 Mobile is expected to be rolled out to select users within Thornton Tomasetti near the end of Summer 2019 and a pilot inspection study using the T2D2 web service is expected later in the year.

Application Areas
T2D2 has broad applications across the Thornton Tomasetti practice portfolio and can be deployed to enhance inspection efficiency for various types of structures, such as buildings, bridges, tunnels, nuclear reactors, and petrochemical facilities.

Asterisk (alpha)

What is it?
Asterisk is a web application for rapid structural optioneering in early conceptual design. Simply upload a mass and core model from the Rhino client and adjust a few parameters, and Asterisk returns a concept-level structural design in a matter of seconds. Users can create iterations, explore and filter design spaces, and compare the performance metrics of select sets. Asterisk is a collaborative platform in which project teams can create, review and present design spaces across multiple mediums.

The History
CORE studio began researching Machine Learning in 2015 and published the first white paper at IASS 2015. Click the link below to read it.

Performance Measures from Architectural Massing Using Machine Learning

With Asterisk, You Can:

  • Iterate through a series of user-defined parameters, such as program, bay spacing, and material, to build up design space. Upload masses from Rhino, iterate and download wire-frames back into a modeling workflow.
  • Explore multiple structural iterations in filterable design space with the integrated Design Explorer interface. Set limits to results, like weight and cost or bay spacing and floor-to-floor height inputs, and get back the iterations that meet project criteria as they evolve.
  • Compare user-selected sets of iterations to better understand their relative performance. See trade-offs between options in a comparative matrix that visually highlights the top-performing metrics in each category.

What’s under the hood?
Asterisk is built from a foundation of CORE platforms, including Spectacles and Design Explorer, CORE studio’s web application suite. Asterisk achieves its speed by leveraging CORE studio’s predictive models. These models have been trained on data generated to Thornton Tomasetti design standards and pulled from the expertise of our engineers and from their workflows.

Asterisk is the first platform to apply the ongoing research at Thornton Tomasetti to machine learning and its implementation in AEC.

CORE studio + Applied Science

The Applied Science practice of Thornton Tomasetti has a long history of using a variety of machine learning algorithms on its projects to provide solutions to clients in the defense, life sciences, and energy sectors. CORE Studio supports these projects through the continued involvement of Dr. Badri Hiriyur in addition to providing development resources for related web and desktop applications.


 Want to get in touch? Curious to know more? REACH OUT!