Model Maintenance

3D representation of a building
Dashboard 1
Dashboard 6
Dashboard 2
Dashboard 4
Dashboard 5
Dashboard 3
Dashboard 8
Dashboard 7
3D representation Dashboard

Inspiration

We were inspired by the ability of immersive 3D models to visualize data in a new way. We wanted to create a system that any user can immediately visualize the state of the building that they are working in.

What it does

Model Maintenance has several pages including a notification system, directed preventative maintenance in a pipeline system, as well as an ever changing 3D model that represents which portions of the building needs the most attention. This is accomplished through a machine learning model that takes in the current state of an elevator/hvac system and finds the current state of a given machine. Since the model can be served at a quick cadence, system problems can be found quickly. This allows maintenance works to focus on the tasks that matter the most. When everything is going well the system will still encourage the most necessary preventative maintenance. All of these notifications happen through an integrated SMS alerting system.

How we built it

We built this system in 2 pieces. The front end displays the website that holds all of the information required for the workers to do their job. The backend deploys the model and processes the sensor information for a given system through our machine learning model. It can then determine simple reasons for potential failures such as door failure, vertical adjustments needed, cable service or coolant leaks.

Challenges we ran into

The main challenge we ran into was a lack of maintenance related data for both HVAC's and elevators, and displaying an immersive 3D model for our users that represented a real building. The data challenge was solved by an old dataset on kaggle that we had to clean and refine. This challenge would be further mitigated in the future as we can collect time series data to actively refine the model. In the case of the 3D model we had to use blender to create the model and export a file that was usable by Javascript.

Accomplishments that we're proud of

We have several accomplishment that we are proud of working through this project. This is our first project where we trained a unique model to fit our specific use case and the first time we have worked with 3D models. In regards to our frontend we are proud of the reactivity and pipeline page that correspond to service tickets that are needed in our example building.

What we learned

We have learned how to stream data in a constant fashion, and being able to check the state of the system at any point during the day to ensure that we are always staying on top of the required maintenance for the companies that use our product.

What's next for Model Maintenance

We plan on better supporting a single company being able to manage multiple buildings at the same time as well as being able to assign tasks to a companies maintenance workers. This would allow us to better take advantage of our SMS alerting system on a worker to worker basis