Best neural network for certain type of pattern analysis?

I'm work开发者_StackOverflowing on a system that will send telemetry data on machine operation back to a central server for analysis. One of the machine parameters we're measuring is motor current drawn vs time. After an operation is finished we plan to send back an array of currents vs time to the server. A successful operation would have a pattern like a trapezoid, problematic operations would have a pattern completely different, more like a large spike in values. Can anyone recommend a type of neural network that would be good at classifying these 1D vectors of current values into a pass/fail type output?

Thanks, Fred

Maybe taking the FFT and passing it through a radial basis function neural network will do the trick. It seems like the features you are looking for are periodic features which will be captured by the FFT, and RBF can do the learning.

Many types of neural network might be used to solve this problem, but I imagine that a relatively simple scoring function might work as well and be much easier to implement. If you can identify the likely locations of the beginning and ending of your trapezoid, I suggest trying something like average "absolute difference from a trapezoidal template shape" as a measure of machine performance.