By Brett Young
May 1, 2017
There's a lot of hype around many of the current technology trends. At BuildingSP, we want to be sure that we stay above the hype when we're talking about our technology roadmap. As we talk about our product with customers, we want to make sure that they understand our vision for future development. This is especially important because ClashMEP, our real-time clash detection solution, is priced using a subscription model where current users get the benefit of future upgrades. This post describes our thinking and near-term approach to development around clash management and machine learning. By describing it here, our current and future customers can understand future capabilities and how we are going to achieve them, cutting through the hype.
BuildingSP's development roadmap is going to solve a big problem in the practice of clash detection: coordination meetings. We see three problems with the vast majority of clash coordination meetings.
- There is too much manual work that is required to prepare for a meeting. This includes combining files, reviewing clashes, comparing current clash results to previous clash results, and creating issue logs.
- There's significant wasted time in coordination meetings. Coordination meetings are expensive meetings with many highly skilled, highly paid staff members participating. If you're reviewing clashes in between two firms, the other firm at the meeting isn't involved and is wasting time.
- The meetings are structured incorrectly. Coordination meetings for clash usually follow a summary of a series of clash tests from traditional clash detection software packages. For example, you'll look at all the plumbing clashes with electrical, followed by all the plumbing clashes with fire sprinklers. This structure is illogical because it lacks focus and is pedantic.
BuildingSP will change clash coordination meetings by introducing the concept of clash prioritization. This is important because it will allow teams to focus on these high-priority issues during meetings without wasting effort sifting through unimportant clashes. At BuildingSP, we believe that the importance of clashes follows a normal distribution, as described by this graph:
On the left-hand side of the chart, you have clashes that are of low importance and that can be ignored. In the middle are the majority of clashes. These clashes are important, but are also fairly easy to correct. We characterize these as "procedural" clashes. They typically only involve one moving system type, don't involve any cost issues, and don't require design changes. The problem with current practice is that we often struggle to identify the clashes that are most important. These are the clashes on the upper end of the normal distribution. These clashes are complicated, involve cost issues, impact the functionality of the systems, and require meetings.
To prioritize clashes, BuildingSP is using three approaches – clash heuristics, computational BIM enabled by ClashMEP, and machine learning – all of which contribute to a clash priority "score." We'll tackle each of these elements as we describe how we score the priority of a clash.
To start, let's describe heuristics. Here's a definition of heuristic:
A heuristic technique (/hjᵿˈrɪstᵻk/; Ancient Greek: εὑρίσκω, "find" or "discover"), often called simply a heuristic, is any approach to problem solving, learning, or discovery that employs a practical method not guaranteed to be optimal or perfect, but sufficient for the immediate goals.
When talking about a heuristic technique for clash prioritization, we simply mimic our intuition and experience to create a starting framework. For example, we know that a clash between a duct and piece of the structure of a building is a high-priority clash because the steel can't be easily modified, ducts are a high-value system, and ducts take up a lot of building volume, which means that other systems are likely to be impacted. In a heuristic way, this clash would have a high score. The opposite of this would be a clash between two plumbing parts, such as a coupling on a pipe. The fact that they are clashing is unimportant and more than likely the result of an error in the plumbing part family. Therefore, this clash would have a low heuristic score. We can use our expertise to create rules and methods around all of our intuitive beliefs about clash priorities to come up with a scoring method that may not be optimal or perfect, but is somewhat reasonable.
To the heuristic component, we add some computational BIM techniques that are enabled by ClashMEP. ClashMEP is integrated into the modeling environment, rather than as a separate platform. This means that ClashMEP has access to more metadata and information than a clash detection platform that relies on exported or uploaded files. One example of a computational BIM technique for scoring clash prioritization is the measurement of congestion. If a clash happens in an area of the building that is highly congested, it's also likely to be high priority because congestion means it will be hard to fix. Using computational BIM approaches, we can measure the congestion around a clash in order to further refine our clash priority score. Congestion is one of only several computational BIM approaches we're using at BuildingSP to inform our clash priorities.
Machine learning is the final component we use to prioritize clashes. It is helpful for adding user preference and refining all the previous scoring components. In the ClashMEP user interface, there will be a method to increase the priority of a clash. This provides a feedback loop that can be used to change and modify the scoring. This is a known method with many other parallels in machine learning. For example, your Netflix movie viewing history is used to refine the standard heuristic methods to recommend the next movie you'll watch. This is a classic example of machine learning in action and ClashMEP uses some of the same procedures.
To see how machine learning informs clash prioritization in practice, check out this workflow to view the feedback and learning loop:
By looking at the workflow and explanation of our methods, I hope it's clear that we have an approach that is based on our experience with similarities to other proven methods in other domains. The overall goal is to forever change and improve how we coordinate projects using advanced computational methods, rather than the blood, sweat, and toil that we currently put in through our manual methods. At BuildingSP, machine learning is a key component of how we're improving the process. However, it is not the only way.
We'll talk more in future posts about how collaboration methods borrowed from software development and manufacturing, which are enabled by ClashMEP, will further revolutionize our industry.
There's more information available on LinkedIn.