XPost: comp.ai.neural-nets, comp.ai.genetic, comp.ai.edu   
   From: rohanar@ecg-inc.com   
      
   Would a neural network be useful for the following application?  If   
   so, what type of network, and how?  What language / software should it   
   be implemented with?   
      
   I have sensors that simultaneously detect deflection (vibrational   
   data) on 3 pieces of interconnected rotating machinery [M1,M2,M3].   
   The raw data includes 3 sets of ~2000 points representing deflection   
   -- this is the time-domain data.  Performing FFT calculations on this   
   data shows spikes at certain frequencies -- this is the   
   frequency-domain data.  In "good" data, there are spikes at a few   
   frequencies, and all 3 machines have roughly the same deflection.  We   
   gather and store this data several times a day.   
      
   We want to detect faults in the machinery based on this data over   
   time.  Each type of fault is indicated by some change in the   
   time-domain data and/or the frequency-domain data.  For example, we   
   expect a spike at 100HZ; if that spike is higher or lower than some   
   variable "normal" on a the data for M1, then a Skid has occurred on   
   that machine.  If that spike is higher or lower then some variable   
   "normal" on all 3 sets of data, then it's something else entirely --   
   it's a fault that effects the casing that houses M1, M2, and M3.   
      
   I have about 20 of these faults to detect.  (Other faults are detected   
   using changes in standard deviation, mean, locating steps changes,   
   etc)   
      
   Also, these "normals" are going to be variable based on load (10% to   
   100%) on the machines.  There is not necessarily a linear relationship   
   between load and deflection either.   
      
   Any ideas would be greatly appreciated.   
      
   [ comp.ai is moderated.  To submit, just post and be patient, or if ]   
   [ that fails mail your article to , and ]   
   [ ask your news administrator to fix the problems with your system. ]   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)