Speaker
Description
For the determination of the hardness penetration depth in thermally hardened steel parts we present a non-destructive, contactless methodology based on laser-ultrasound and supervised machine learning. Ultrasound pulses are excited at the surface by a pulsed laser source. These propagate into the sample and are preferentially backscattered at the interface between hardened layer and core due to the difference in grain size. The backscattered waves are subsequently detected at the sample surface by a second laser and a two-wave mixing interferometer that is capable of measuring on rough surfaces. The backscattered acoustic waves carry the information of the extent of the hardened layer.
We demonstrate the method on three industrial grade samples with different microstructural peculiarities accounting for typical difficulties arising in the industrial production process. Due to the inherent fast data acquisition in our laser ultrasonics setup, we are able to perform lateral scans along the sample and use the additional spatial information to apply a supervised machine learning approach that provides us with the sub-surface lateral and axial contour of the hardened layer. We require no additional calibration step for the data evaluation which contrasts with the usual time-domain evaluation methods and a major improvement regarding industrial needs.
In conclusion, a spatio-temporal measurement method based on laser ultrasound is successfully applied to industrial samples with hardened surface layers. The subsurface spatial profile of the hardness penetration depth can be determined by a supervised machine learning approach without additional calibration step. Our current findings also show the need for more sophisticated measurement schemes in the presence of additional scatterers in the hardened layer. An improved fundamental understanding of the spatio-temporal scattering in heterogeneous microstructures for quantitative evaluation will be required for further improvements of this technique. Furthermore, the method needs to be extended to thinner layers that are currently masked by the initial opto-acoustic crosstalk of the laser excitation mechanism.