Module 8 Questions:
Part 1 - Study One Talk - Option A:
The speakers throughout the video are trying to solve a major challenge that happens in the early-stages of building design, which is obtaining energy performance feedback early enough to actually influence design decisions. Traditional design processes prevent there to be an informed design from the building performance perspective because an energy model cannot be accurately created until the design is close to finished, at which point the data from the energy model would be useless since the design cannot be changed that much. Additionally, energy models are technically difficult to set up and they are slow to keep up with rapid iterations that happen during the conceptual design phase. To help address this problem, the speakers developed a workflow that uses Dynamo, Generative Design, Autodesk Insight, and machine learning. They were able to use generative design to automatically create thousands of building variations. From there, they could run an energy simulation on each of the designs and then use the resulting dataset to train a machine-learning model that was used to predict the Energy Use Intensity (EUI) of the building. Their overall goal was to create a surrogate model that was able to provide rapid performance feedback without requiring a full energy model every time the design was changed.
One thing that surprised me was how big of a synthetic data set they were actually able to create. One of the biggest roadblocks in this field of study is that architectural plans are not readily available because companies won’t publicly publish them. So, instead of relying on real project data, they were able to generate their own dataset that consisted of different building taxonomies, unique geometries, data samples, and climate zones. I had always guessed that machine learning applications in AEC would always struggle and be limited by the lack of available data, but this project showed me that generative design can actually create the data that is needed to train the models. One thing that I would push back on is whether or not this prediction model would actually be able to fully replace a simulation based analysis of the building. While the presentation reported a prediction error of 1.91%, which is insanely small given the conditions, I would still feel the need to have a detailed energy simulation in order to make final design decisions, especially on projects that are very focused on building performance.
Out of the information that was shared, I would like to try using a similar workflow during early stages of design to evaluate HVAC and energy performance decisions. As an architectural engineer who is very interested in sustainable design, I really see value in receiving almost instant feedback on how changes to the design of the building can impact the energy use. This kind of tool would allow me to work closely with the architect in early stages to help shape the building for optimal performance. However, I would be a little sceptical about totally replacing engineering analysis with this tool. Personally, I would need to have confidence that the data being computed by the tool could be verified against detailed simulation results before any major design decisions were made.
Part 2 - Reflect on the Quarter:
Throughout the quarter, I encountered many points of friction during my modeling process. The first and main one that I can think of came when I was creating a model with a lot of moving parts. With these kinds of models, I learned that limiting your inputs by correctly setting the min and max was vital in order to not crash the program every time I tried to run it. This was a very important lesson to learn because it helped me drastically reduce the amount of time I was spending on each assignment. Speaking of time management, I sometimes underestimated how long debugging my Dynamo graph would take as well as how long it took to align my model with the design intent once the basic graph was set up. Especially in the later assignments, setting up my evaluation criteria and running iterations until I was satisfied often took a long time. Another challenge that I encountered during specifically Assignment 7 was finding the design alternatives that would give me the most informative data. I went through many iterations where my model was working but the results weren’t super helpful. So, I had to really spend the time to adjust the inputs, review the outputs, and decide which design alternatives I wanted to actually investigate.
An AI-augmented version of this workflow could look like working with a design assistant who you can feed your project goals to and it would help you reach your starting point. Then, after each iteration, it would learn from its mistakes and automatically adjust the parameters to reach your final outcome much faster. While I think this AI-augmented workflow could be helpful in reducing the time it took me to complete my assignments, I don’t think I would personally want to utilize it. I found that most of my learning and true understanding came after I spent the time to debug and align my model properly. I think having these points of friction, particularly in the early stages of learning a new platform, are critical to grow.
Part 3 - Scout the Frontier:
1. cove.tool
Cove.tool is an AI-platform that is able to perform building performance analysis using automation and machine learning in order to help engineers and architects evaluate energy use, carbon emissions, daylighting, and HVAC performance. This allows its users to compare design alternatives and receive rapid feedback at early stages of design without running a manual detailed simulation every time. This platform is interesting because traditionally building performance analysis occurs later in the design process once major decisions have already been made. This usually means that the findings will only help to inform mitigation measures instead of impactful design decisions. Cove.tool helps to bring the analysis to the front of the project by providing quick performance predictions and making automated comparisons. This helps to make informed decisions regarding sustainability, energy efficiency, and building systems, from early stages in the design. This tool could have been really beneficial to have or try out during Assignment 7 where we evaluated different design decisions. Since we made simple geometric models, it would have been able to give us detailed information that would have helped us to understand the tradeoffs between different design alternatives even in the early stages of design.
2. Hypar
Hypar is a cloud-based platform that automates the early, conceptual design of buildings. It does this by converting building program requirements and other constraints into 3D floor plan options in a short amount of time. This is interesting because it drastically reduces the amount of time spent working on the time consuming conceptual design phase of a project. Unlike traditional modeling software where designers have to manually model their geometry, Hypar allows users to define the rules for the building and let the computer generate solutions. This shifts the engineering efforts from modeling to system design and decision-making. Since we weren’t designing floor plans during this course, I don’t believe that it would have been super helpful to the work that we were doing. However, if we could use some of the principles of Hypar to help us automate the geometry by implementing only our parameters, it would have been a large time saver that could have allowed us to focus more on evaluating the performance and tradeoffs of our design instead of spending so much time building our actual models.
3. Augmenta
Augmenta uses AI to automatically generate building system layouts, currently only electrical systems but they are working on developing the same technology for mechanical and plumbing systems as well. It does this by having the user upload their architectural Revit model and defining some inputs and parameters regarding the system. It then analyzes the building and finds clash free and optimal routes for each building system. At the same time, it compares many solutions and determines which is the best one. This program is interesting because it helps engineers to automate portions of their actual design process and does the modeling for them, which from experience is a tedious and long process. As someone who is interested in mechanical systems, I found this tool the most exciting. A lot of mechanical design involves duct sizing and routing which is a boring and time consuming task. Augmenta allows MEP engineers to focus more on system performance, sustainability, and design optimization. Similarly to Hypar, this tool would not have been super helpful during this class with the work that we were doing. However, it would have been interesting to see if we could have implemented it into Assignment 7 where we were weighing different design alternatives. We could have taken the information that Augmenta gave and tried to optimize the system for different things such as minimal duct length or least amount of elbows.