Module 8 Questions:
Part 1
I chose Option A - The technical pipeline. The speakers are talking about the lack of quality datasets in usage for generative design workflow. Particularly, they are talking about estimating the building performance in regards to energy performance and how to use said generative modeling to create multiple models and possibilities using machine learning that output accurate predictions. They propose the use of parametric modeling through Dynamo & Revit to build these datasets using the generative design features. These datasets are then used to train models in machine learning to give an EUI prediction, with relatively high accuracy of slightly under 2%.
One of the aspects I found slightly confusing was the part about how the different target taxonomies had to avoid overlap between the classifications. While I understand that this is used to reduce bias, I also think it could unintentionally omit certain designs that don’t fall neatly into a specific category. This could possibly be solved by having an “other” taxonomy, but this could also potentially create bias and connections between geometries that would otherwise be unrelated.
I think the process of generating datasets through parametric design would be really interesting, particularly with bid sizing of designs. If the tool was able to accurately estimate the sizing of structural members for preliminary designs, or provide a range of guidelines, this could be super helpful and reduce added costs from errors during bid time. There would also need to be a method to backcheck the validity of the models generated through the parametric design process to ensure that there weren’t any compounded errors or improper assumptions that the software was making. Plus, this tool could help with sustainability and reducing carbon emissions, as seemed to be a focus of the talk and reason for generating such a tool to make more economical decisions during the design phase of a building.
Part 2
I would hit some friction in the design ideation phase of the different models. I often found it difficult to visualize something that I had the tools to generate into a model. AI could potentially help with this in the sense of transforming a sketched idea into a more palatable format that can be implemented into a model. Or, to go a step further, an AI workflow could involve presenting a sketch and then it would build a preliminary model based on that sketch. The only downside of this is that the sketch would likely need to be very detailed or else the model would be basically useless. It often took longer to debug than expected, especially with figuring out previews and learning when to run the model or how to best debug the model to limit run times and only provide the necessary cases for testing. Some repetitive work could be automated through the use of things like custom nodes, but in general, there was some logic aspects that were a bit repetitive throughout the weeks. Therefore, automation could come in the form, particularly for buildings, in extruding surfaces and generating solids with an inputted sketch or line profile.
Specifically, the error codes could potentially be clearer and provide more specific guidance on the bugs. This could be aided by an AI agent guiding the inputs and outputs and providing comments and questions about the intended purpose of the logic. This would then ensure the provided workflow followed the correct logic while also running, as sometimes you can debug a piece just for the outputs to be wrong due to incorrect logic. Yet, having said all that, the friction of debugging does teach a lot about the software and really challenges you, so I would not want the augmentation, at least from a beginner’s perspective. For large-scale implementation with programmers who have a strong understanding of the software, I believe the AI usage could be helpful and save a lot of time.
Part 3
Company 1: Trane Technologies Smart Buildings
This company has an integrated workplace management system which seeks to manage the assets in a building, like the lighting and usage rates within a building. While smart buildings have been a growing concept, this company combines the individual components of a building, like the HVAC, plumbing, and elevators, into one managed system. Due to the large quantities of data that have to be constantly assessed and optimized, AI is necessary for such a large-scale application and to allow the various components to work together. For example, this system could adjust the AC in a system based on the number of people in a room by sensing the body heat of individuals or coordinating room usage and calendars through the AI agent. While this tool doesn’t involve the modeling side of the built environment, it could enable designers to be more innovative with their designs and worry less about energy usage at the start due to the saving measures that could be implemented during the lifespan of the building.
Company 2: TestFit
This company allows for the generation of various basic built environments given a specific site. This allows for quick real estate planning and feasibility as well as fast idea generation that can be visualized in real time through the use of AI. This is interesting as it helps for planners to determine the constraints set by a site location early on. It also helps to determine the various ways in which land can be used and which is more aesthetically pleasing or optimal for various use cases. This allows for optimization in design early on as well as it helps to visualize various design components to ensure they fit well into the planning of the surrounding area. This tool could have potentially helped with idea generation by seeing the various forms of buildings on various sites, even though it would not have been able to help with the exact implementation of then building the model following the initial idea.
Tool 3: Buildots
This tool is an AI monitoring tool that tracks and compares the progress of a construction site with that outlined in the BIM model as well as set schedules. It can also help to forecast the progress of work which in turn helps to predict bottlenecks and potential timing issues or physical clashes based on variance in the built product as compared to the modeled design. This is interesting as it can help improve visibility for designers on the physical progress of their design as well as allowing for faster times to fix conflicts and issues. By pointing out these issues earlier, the AI tool decreases delays and also ensures the building is built to specification, leading to less potential safety hazards during the building’s lifecycle. This is again a tool more for application in the built environment than in the modeling process, but this could be really helpful in ensuring designs are built to the specifications and allowing for real-time design updates and changes.