A technique combining laser spot thermography and neural network for surface crack detection in laser engineered net shaping

Abstract

Laser engineered net shaping (LENS) is one of the core technologies commonly used in additive manufacturing. Improper optimization of the LENS process parameters may lead to defects, such as porosity and cracks, in the manufactured parts. Cracks, for instance, can adversely affect the mechanical properties of the parts, which may render them unserviceable. Therefore, it is necessary to develop online nondestructive testing (NDT) methods to monitor and detect the defects. As an important NDT method, laser spot thermography (LST) has drawn great attention due to its characteristics of non-contact and high-sensitivity. In this study, we design a laser spot scanning system, which can exert a multi-mode point heat source. A new technique combined with laser spot thermography and neural network is developed to characterize the crack parameters in LENS components. The temperature and its gradient at the crack location are extracted and used as inputs for training the neural network. The trained neural network builds a relationship between the input (temperature signal) and output (crack width) data. After verifying the reliability of the neural network by conducting numerical simulations and experiments with artificial cracks, the widths of real cracks in LENS samples are evaluated. An average absolute error of 2.0 μm (and the maximum absolute error was 6.1 μm) is obtained for LENS samples having crack widths in the range 3–68 μm. We conclude that the LST-NN technique can be used to detect the location and width of cracks with reasonable accuracy, and has the potential for further applications in the online monitoring of the LENS manufacturing process.