This is the third in a series of blogs about Multivista’s industry-leading Automated Firestopping Assessment.
- How did the story of automating firestopping assessment begin?
- What were our clients’ concerns during the prototyping phase, and how did we address them?
How did the story of automating firestopping assessment begin?
Multivista’s Firestopping Assessment journey began when one of the world’s top general contractors expressed an interest in documenting the firestopping measures for a project in Scotland, UK.
Alan Bryant, owner of Multivista’s Glasgow franchise, responded to the request and began documenting that project’s firestopping measures using our 360 Photo reality capture technology.
Alan Bryant, Multivista Scotland’s Sales Director
As the project progressed, Alan and his team accumulated a wealth of knowledge on how to properly document firestopping measures, how to find potential issues, and how to link them to Building Information Modeling (BIM).
Using what they learned, Multivista Scotland continued to provide this service for more than 40 other mega projects worth more than £1.2 billion.
Alan and his team in Scotland
After a year of continuous manual firestopping assessment work with various construction owners, project managers, and contractors, Alan and Multivista began exploring the idea of using the hundreds of thousands of collected images as the foundation for a machine-learning algorithm that could automatically detect problematic firestopping issues and unsealed penetrations.
Furthermore, the algorithm could provide project teams with a punch list item for each detected issue, as well as the location and a photograph of the issue. Everyone agreed that this automation would save contractors a lot of time and significantly reduce rework.
Collaborating with parent company and global information technology powerhouse, Hexagon, Multivista worked with artificial intelligence experts to develop the world’s first automated firestopping assessment process by combining our specialist-captured 360-degree jobsite images with human-based firestopping penetration identification knowledge.
The team in charge of this AI technology used a variety of image recognition methods and created a number of computer models, which are capable of detecting penetrations through fire-rated walls and recognizing and flagging any that may be incomplete, poorly done, or problematic.
After 6+ months of training and re-training, the Multivista Firestopping Assessment AI now functions as well as or better than a human technician. After analyzing hundreds of thousands of images during the learning process, the system runs continuously on massive servers and employs an unsupervised learning algorithm that has five stages: initialization, sampling, matching, updating, and continuation. Which is to say Multivista’s Firestopping AI learns and becomes more accurate with every job it performs.
What were our clients’ concerns during the prototyping phase, and how did we address them?
We worked around the clock and communicated with construction contractors and project managers all over the world to understand their concerns and get their feedback on firestopping and the issues around it. Here are the top five questions and concerns we’ve received, as well as how we responded to them.
Q1. Manual visual quality control is always costly and time-consuming. In this case, what would the automated solution provide?
Multivista documentation specialists have received extensive training in capturing images of firestopping measures based on the company’s best practises accumulated over the last 18 years. The 360-degree digital images captured are fed into a highly advanced deficiency detection algorithm trained by our team of 20+ subject matter experts and drawn from the world’s largest construction documentation image database. The trained AI program can recognize unsealed firestopping penetrations at fire-rated walls.
Q2. There is a lack of integration from a manual QC report to actionable punch lists for quick and complete resolution of deficiencies. What would you do?
For each detected penetration, our system will automatically create a task in Multivista’s Task Manager. Each Task will be assigned to the applicable personnel responsible, with a due date, and linked to a photograph of the deficiency, and its location on the floor plans for easy remediation.
Q3. Finding the exact location of quality control issues on a floor plan can be difficult, time-consuming, and error-prone. Is there a way to speed up this process and maintain accuracy?
Multivista Documentation Software (MDS) has the ability to annotate and comment on captured firestopping images and all flagged photos are linked to floor plans and historical images of the same location from different dates.
Q4. Because not all data is captured by the standard manual QC, determining and holding parties accountable for deficiencies can be challenging. What steps would you take to ensure that all data is documented?
Our documentation specialists are constantly coordinating with the project manager and checking the project log to ensure the ideal time to visit the project site and capture the information needed without interfering with the work of the subcontractors. As a result, firestopping measures documented by Multivista specialists are correlated with the project log, which shows who is responsible for the installation of services in each area of the project.
Q5. New solutions imply a change in existing construction workflows, and such technologies would necessitate extensive training to operate. Is this going to be the case with this new technology?
Without human intervention, the solution detects unsealed penetrations with an accuracy of up to +95 percent. All images and the punch list of potential issues are housed within the same Multivista Documentation System (MDS) software our clients have been using successfully since 2003.