I am applying a 1-D convolutional network on current that drive an electromechanical component for diagnosing faults occurring within the system. A (state space equations) is available for when the component is healthy. It describes the relation between the current signals and the mechanics within the component. So it is possible to fit the to the current signals to the and obtain residual signals instead. My goal is to investigate whether using these residual signals results in a better classification.

My question is: is anyone familiar with work where machine / learning is applied on signals that have been fitted to a domain-specific model first? I have been looking in the health monitoring field without much success. This was the best work I could find (using a parameter found by a Kalman filter as additional input to the NN).

But I could imagine the same problem statement occurring in other fields where physical signals are used, like climate modeling, quantum chemistry, building physics,…

Source link
thanks you RSS link
( https://www.reddit.com/r//comments/8wbk1z/d_deep_learning_on_sensor_signals_for_which_a/)


Please enter your comment!
Please enter your name here