Inspired by many a destroyed DC motor, we wanted to create a solution to monitor the various behaviors of motors and give them the necessary intelligence to prevent catastrophic results due to overloading, overheating, etc. We believe this has great value on an industrial scale, where a broken DC motor could cost thousands in halted assembly lines and repairs.
What it does
MotorSkills offers a Machine Learning approach to modernizing factories to allow for reduced maintenance costs from damaged/destroyed motors, automated diagnostics, and at-a-glance malfunction corrections. Our package of hardware and software allows for easy implementation of a distributed and scaleable motor monitoring solution for industries small and large.
How we built it
MotorSkills skims data from motors (e.g. in an assembly line) including their amperage and revolutions per minute (RPM), and passes it to AutoML to do trend analysis and classification of sensor inputs. These results are translated on a front-end UI for visualization and are sent via push notifications to alert employees of any drastic issues.
Challenges we ran into
Designing a Machine-Learning model to track and provide suggestions for how to resolve classic DC motor issues (including stalls, load spikes, etc.) required a unique solution where we turned time-series sensor readings into miniature images that could be ran through an image classifier to detect for anomalies in motor behavior.
Accomplishments that we're proud of
Creating training data for our ML model using our own hardware was a tactile way to learn more about the interaction between our training data and our model.
What we learned
We learned how to connect (almost all) the GCP suite together in one cohesive, useful hardware application. We were also excited to implement serverless function calls and parallel clustering of our Node.JS backend to support Bluetooth communications inline with AutoML.
What's next for MotorSkills
We'd like to look into adding more sensors to our system to make our motors smarter and more reactive to their environment (such as monitoring for overheating to prevent catastrophic motor degradation). In the end, we want to take an eighteenth century technology into the modern, ML-powered era.