I’m trying to take data from different sensors in a machine and using it to run an self supervised learning model on the edge which will tell me about the next service date for the machine along with the root cause of the problem with the machine eg. bearing worn off
I have the following questions:
Can I use my own Machine learning model on 2nd step of edge impulse instead of the models provided by the platform?
Also, Is it possible to take data from 25 different sensors in real time using edge impulse?
It would help me a lot if someone could answer these questions.
For the machine learning model, you’ll need to train the model through edge impulse itself, but you can use any custom model architecture implementable within Keras. You’ll just need to enter expert mode as described here:
For the multiple sensor types - this is fully supported with the only limitation being all sensors will need to be collected at the same sample rate for now. We have a few different methods of getting data from sensors in real time.
The first (and simplest) is using the data forwarder. This is suitable for low-frequency sensor input and is documented below. It is basically just a comma delimited stream of sensor values sent over a serial connection. You then use our command line tools to identify what sensors the values correspond to.
The more complex method involves using our ingestion sdk to create a sample in the edge impulse data format on the device, and then send it via serial or any other on-device supported protocol. The advantage here is this supports any sample rate, as well as creating multiple samples with potentially different labels from the same device without user interaction.
Yes @daschwar is working as a User Success Engineer at Edge Impulse.
We do offer customer support with our enterprise offers. Can you please contact us at email@example.com explaining your project, one of our awesome sales executive will contact you