A data-based modelling scheme for estimating residual stress from Barkhausen noise measurements

Abstract

Barkhausen noise measurement is an intriguing technique for the non-destructive testing of material properties. It is applicable to ferromagnetic materials and shown to be sensitive to many material properties. Typically, some features are calculated from the Barkhausen noise signal and compared to the material properties. However, the results are mainly qualitative. In this study, quantitative prediction of residual stress based on the Barkhausen noise measurement is studied. The studied material is case-hardened steel 18CrNiMo7-6 (EN 10084). The development of the prediction model is divided into feature generation and selection, followed by model identification and validation. In feature generation, the aim is to generate features as robust as possible, while in feature selection the most suitable features are selected to be used in the prediction model. In the model identification step, a simple linear regression model is identified. The model is validated in the final modelling step to ensure that it can also be used for future predictions. The identified model gives rather accurate predictions, indicating that the proposed modelling approach is applicable for predicting material properties from the Barkhausen noise signal.