Prediction of Residual Stresses Using Partial Least Squares Regression on Barkhausen Noise Signals

Residual stresses are quantitatively predicted based on the Barkhausen noise measurement with a partial least squares regression model. The measurements are taken from two sets of case-hardened samples. The benefits of using certain feature elimination strategies prior model identification are also studied. The elimination methods applied are correlation-based elimination, uninformative variable elimination and successive projections algorithm. The results show that the best predictions are usually obtained when the successive projections algorithm is applied. The prediction accuracy of the best models found shows that partial least squares models can be successfully used for prediction of material properties based on the Barkhausen noise measurement.