Engineering Perspectives: Optimizing Optimization

Syska Hennessy has been doing a lot of work lately on computational approaches to AEC problems, and optimization has been at the forefront of our minds. Perhaps because of this focus, many of us get a nails-on-the-chalkboard type of feeling when we hear the industry’s often casual use of the word “optimization.” For example, the following kind of statement is really common:

“We optimized the 3D model.”

Uh, what? Such a simple statement leads to so many questions. What exactly was optimized? Was it a prescribed approach or was optimization found “organically”? Was the optimization done by a team or by a person or computer application?

To give us all a foundation of how Syska Hennessy thinks we should think about optimization, we wanted to propose a simple definition of optimization for AEC so we can all use the same terminology.

What is optimization?

To define optimization, we’re going to start with how we define it at Syska Hennessy, then break down this definition to explain the importance of each part. Here goes:

Syska Hennessy’s Definition of optimization:

an iterative process of selecting a solution with a measurably superior outcome.

Here are three parts that we want to dig into:

1) Optimization is an iterative process.

Optimization is an iterative process because it is different from just answering a question where there’s a clear answer. In high school math, it was so satisfying when we did a problem and could definitively say, “The answer is 42.” In computational terms, a non-iterative process is usually a solver, a process that results in one answer without iteration. The objective of optimization is different from that of a solver because optimization centers on achieving a series of solutions, which ideally keep on getting better.

The iterative nature of optimization is important to highlight because of Syska Hennessy’s growing experience with machine learning. With ML, we’re able to do more iterations, faster, cheaper, and better. The characteristics of the optimization process are defined by the cost, speed, and value change per iteration.

2) Optimization involves selecting a solution.

Optimization must involve more than one solution. If a team can only present one solution, there’s no way it can be optimized because it cannot be selected against another outcome. This concept is important for AEC because we typically take time to develop solutions and, for this reason, we rarely fully combine multiple outcomes into a full solution. This practice, however, will change as our industry embraces workflow improvements, like algorithmic design automation.

In the future, Syska Hennessy believes we will be able to present more solutions at higher levels of detail for our projects and that optimization will be a core skill for our teams.

3) Optimization results in a measurably superior outcome.

This gets at the heart of the question, “What are you optimizing?” There’s an art to applying mathematics to the real world, and finding a way to define a measurably superior outcome is a big part of the optimization challenge. AEC industry problems rarely boil down to situations where the value or cost is described by a single, simple equation. But optimization must include evaluation criteria to compare different outcomes. For engineers, constructability describes this challenge. An engineer can optimize a system to be perfectly efficient, but also completely unbuildable. The “art” of optimization is then determining ways to add value or otherwise control for poor solutions, like outcomes with poor constructability.

Now that we’ve gone through the definitions, it’s time for predictions. Syska Hennessy knows optimization is going to become more important in the very near future. We hope our thoughts on the definition of optimization are helpful. We’ll be following up with more information on how our investments, research, and internal process improvements relate to optimization.

Stay tuned!

Written By Robert Ioanna