Predictive models based on RSM and ANN for roughness and wettability achieved by laser texturing of S275 carbon steel alloy

Abstract

Laser texturing is increasingly gaining attention in the field of metal alloys due to its ability to improve surface properties, particularly in steel alloys. However, the input parameters of the technology must be carefully controlled to achieve the desired surface roughness. Roughness is critical to the activation of the surface before further bonding operations, and it is often assessed using several parameters such as Ra, Rt, Rz, and Rv. This surface activation affects the properties of the metal alloy in terms of wettability, which has been evaluated by the deposition of ethylene glycol droplets through a contact angle. This allowed a direct relationship to be established between the final roughness, the wettability of the surface and the texturing parameters of the alloy. This raises the interest of being able to predict the behaviour in terms of roughness and wettability for future applications in improving the behaviour of metallic alloys. In this research, a comparative analysis between Response Surface Models (RSM) and predictive models based on Artificial Neural Networks (ANN) has been conducted. The model based on neural networks was able to predict all the output variables with a fit greater than 90%., improving that obtained by RSM. The model obtained by ANN allows a greater adaptability to the variation of results obtained, reaching deviations close to 0.2 µm. The influence of input parameters, in particular power and scanning speed, on the achieved roughness and surface wettability has been figured out by contact angle measurements. This increases its surface activation in terms of wettability. Superhydrophilic surfaces were achieved by setting the power to 20 W and scanning speed to ten mm/s. In contrast, a power of 5 W and a scanning speed of 100 mm/s reduced the roughness values.