Project has been developed to run a ML model in an Arduino Portenta H7 to detect operation conditions of a fan by measuring its noise

Fan is small sized and powered by USB (low power consumption)

Samples have been taken considering four different conditions: Fan not running Background noise Fan running Normally Soft failure Severe failure

Failures were forced by placing objects interfering rotation of the fan.

ML model will be run on the edge in an Arduino Portenta H7. Arduino will measure environmental noise with its integrated microphone and will run the model to detect fan's condition. Output of the model will be flashing in-built LED to categorize operation conditions:

Blue LED: Fan not running Green LED: Fan running normally Orange LED: Fan running with soft failure Red LED: Fan running with severe failure

To train the ML model I've taken 450 samples for each condition. Samples have a length of 2 seconds each. Thus, each category samples sum 15 minutes (15 minutes * 60 seconds/minute = 900 seconds)

Model has many known biases as: Samples were taken from the same distance to the fan. Modifying microphone distance to the fan might have an impact in models performance Background samples were taken in a room without any source of noise. People chatting or music being played in the same room than the fan could impact models performance Samples were taken without any source of interference between the fan and the microphone. Objects standing in the way could have an impact on microphones measurements Fan failure could make different noises where many of them could have not been captured

To improve models performance and flexibility a wider amount of samples and in different conditions should be considered including the ones mentioned in the biases.