Hole-drilling method eccentricity error correction using a convolutional neural network

Abstract

Hole eccentricity is an important error source when residual stress is measured via the hole-drilling method. The conventional ways to correct eccentricity error for hole-drilling residual stress measurement rely on complicated mathematical processes and are difficult to use. To overcome this shortcoming, this paper proposes a method that uses a convolutional neural network to correct for the hole-drilling method eccentricity error. First, the hole-drilling method measurement process in uniform biaxial stress field is simulated via the finite element method. The influence of the eccentric distance, eccentric angle, and stress ratio on the strain measurement error is discussed. Then, a convolutional neural network is trained using simulated data and the hole-drilling method strain measurement error is predicted for arbitrary eccentricity conditions. Finally, the residual stress is corrected by introducing the strain error into its equation. The simulated residual stresses of ten eccentric measurement points in predefined stress fields are corrected using this procedure to conducted numerical tests. The maximum error of simulated stresses decreased from 30.46% to −4.67% after correction. Therefore, the hole eccentricity has a significant influence on the residual stress measurement accuracy of hole-drilling method. The proposed correction method can effectively eliminate this error.