Research Article

Neural Networks Applied to the Wave-Induced Fatigue Analysis of Steel Risers

Algorithm 1

ANN training procedure.
For each loading case response do
 nset = 1
 While [nt <= MaxTries and nset <= QttSet] do
  Train ANN
  Apply ANN
  if [ <= Dmean and <= Dstd and <= Dmax and FlagSp = 1] do
   TestFlag = 1
  else do
   TestFlag = 0
  end
  if [TestFlag = 1] do
   nt = 1000, nset = 1000 (successful training!)
  elseif [TestFlag = 0 and nt < MaxTries and nset <= QttSet] do
   nt = nt +1
  elseif [TesteFlag = 0 and nt = MaxTries and nset < QttSet] do
   nt = 1
   nset = nset + 1
  elseif [TesteFlag = 0 and nt = MaxTries and nset = QttSet] do
   nt = 1000, nset = 1000 (unsuccessful training!)
 end
end
nt = current number of attempt per trial set
nset = current trial set number
MaxTries = maximum number of attempt per trial set
QttSet = trial sets quantity
Dmean = acceptable difference for mean (5%)
Dstd = acceptable difference for standard deviation (8% for axial tension, 6% for moments)
Dmax = acceptable difference for maximum observed value (300% for axial tension, 200% for moments)
FlagSp = Energy spectrum flag, equal to 1 if no significant energy is observed in high frequencies, for ANN response
= mean values of the training and predicted time series
= standard deviation values of the training and predicted time series
= observed maximum values of the training and predicted time series