As part of Autodesk University 2021, I collaborated with Sean Turner, director of innovation, on DfMA for MEP with Generative Design and BIM Automation. We’ve continued to evolve and refine our toolset for DfMA, Generative Design, and BIM automation. Having an evolutionary mindset is critical to successfully deploying phased releases of design tools.

We’re used to working with refined applications (e.g., Outlook, Chrome, Revit). We forget that version one of the tool is going to be different than version 20. Our Excel calculators and AutoCAD LISP routines were developed over decades. We should expect an accelerated development cycle with newer tools, but not perfection. Concept development (DfMA components) must be paired with traditional product development (automation for DfMA).

Evolution of DfMA concepts

The initial duct module has remained the same with an option for a 2.5′ and 5′ module. We’ve furthered the DfMA mindset to other parts of the duct layout. The 8″ runouts are now limited to 1′, 3′, and 5′ lengths while the runout elbows were limited to 45° and 90°. The optimized length of the runout was a simplified geometry problem to get the end of the runout within 5′ of the air terminal for a flex duct connection. This changed the optimization problem from one triangle to a series of smaller triangles with varying length sides and angles. The problem is not limited to all angles and all lengths but rather a small set of each. This provides guardrails for the algorithm with a balance of flexibility and purpose. We did not have to invent the components mentioned as we did with the initial duct module. Instead, we researched and used common building blocks already utilized in the industry and incorporated them into the layout algorithm. With the modularized components modeled, they can then easily be counted on schedules. The design intent is communicated and enhanced without creating risk.

Evolution of automation solutions

In a previous article on BIM Automation, I highlighted automation that focused on design coordination and calculation. Our library of automation has grown and we recently went through an effort to catalogue and classify all of our automation. The categories we used at a high level were administrative, calculation, and design (layout or coordination). The results showed that we have overwhelmingly focused on automating administrative tasks. This makes sense as there is just as much opportunity for efficiency on the operations side of projects as the engineering side of projects.

There is a distinction to be made for automation, automated modeling, and generative design. Automation is not just about efficiency, but also includes accuracy and quality of life. Reducing manual input and transactions results in more accurate data entry and transfer. Removing or reducing monotonous tasks leads to a better quality of life and is part of our overall designer experience.

Administrative automation examples include revision management, routine model management, view and sheet setup. Automated modeling is applying the results of computation or simply scripting modeling of components. Automation can encompass that but also extend beyond modeling. Our effort around the initial duct module was primarily manual layout with a generous amount of automation, and a touch of generative design in the mix. While this was interesting enough to talk about, it was really just a great proof of concept of what our ultimate vision was. This was the beginning of our shift from automation to computational thinking, computational analysis, and computational programming. We intend to shift our automation focus to more calculation and design in the next year.

Evolution of generative design understanding

The first focus was using out of the box Autodesk Generative Design, which was released with Revit 2021. The software includes several examples as well as a primer course. While we found the primer course beneficial for teaching the fundamentals of generative design concepts, we also found that the software was difficult to adapt. The samples work. However, it’s not as simple as swapping out your specific files and problem types with the solution types that they provide. We wanted to find a way to design a solution that would be more reliable and efficient. It was helpful to use the mindset of separating our problem types into the generative method categories of optimization, randomization, and cross-product.

Optimization is a generative method that involves finding the optimal solution to a problem by minimizing or maximizing a specific objective function. This involves adjusting design variables to find the solution that best meets the design constraints.

Randomization is a generative method that involves exploring a range of possible solutions by randomly sampling design variables within a defined range. This method can be useful for identifying a wide range of potential solutions and can be combined with optimization to further refine the results.

Cross-product is a generative method that involves combining multiple design options to create new options. This method can be used to explore the potential interactions between different design elements and can be useful for identifying novel solutions that might not be obvious through other generative methods.

Each of these generative methods can be used separately or in combination to generate design options and solve design problems. The choice of which method to use will depend on the specific requirements of the design problem and the desired outcomes.

Beyond the method type was the type of algorithm, of which there are several. Our focus was primarily around three categories. Evolutionary algorithms, genetic algorithms, and simulated annealing are all optimization algorithms that are inspired by natural processes and can be used to find solutions to problems that are difficult or impossible to solve using traditional optimization techniques. However, they work in different ways and are used to solve different types of problems.

Evolutionary algorithms are a type of optimization algorithm that are inspired by the process of natural evolution. They work by simulating the process of natural selection. Evolutionary algorithms can be used to solve a wide range of optimization problems, including problems with continuous or discrete variables, problems with multiple objectives, and problems with non-differentiable objectives.

Genetic algorithms are also inspired by natural evolution. Like evolutionary algorithms, they work by simulating the process of natural selection. However, genetic algorithms are specifically designed to solve problems with discrete variables and are often used to solve optimization problems in which the search space is very large, or the problem is highly constrained.

Simulated annealing is a type of optimization algorithm that is inspired by the process of annealing in metallurgy, where a material is heated and then cooled slowly to reduce its defects and increase its structural purity. Simulated annealing works by randomly perturbing the current solution and then accepting or rejecting the perturbation based on an acceptance probability that is determined by the difference between the current solution and the perturbed solution, as well as a temperature parameter that decreases over time. Simulated annealing is often used to solve optimization problems with continuous variables and is particularly useful for solving problems that have many local minima. Check out our friends at EvolveLab, who do an excellent job at explaining the connection between all of this and the design application.

That’s part of our story. To implement this class of generative design process, we tried and had minimal success in implementing tools that made an impact on our day-to-day design and engineering. We needed help. The decision to hire external parties with expertise in generative design in the AEC industry was not taken lightly. Dustin Schaefer, chief technical officer, recently sat on a panel regarding this topic. Traditionally, all of our development has taken place in-house and held close to the vest. In the end, we decided to not outsource but rather partner with an industry expert. This is what we hope our clients would do with us and we are happy to work with EvolveLAB.

Conclusion

This is the current version of our story. As I mentioned, we’re always evolving. The next set of challenges lies with not only evolving the complexity, quantity, and quality of our toolset, but also incorporating other corporate initiatives like sustainability and client experience. Stay tuned for more info.