Detecting Steel Surface Properties using Machine Learning

In the production of steel coils, the temper rolling (finishing) process imprints a variety of surface and texture properties into the resulting steel. The temper rolling and resulting surface properties have a large impact on what the steel can be used for. It impacts the steel's press performance later when it is shaped and impacts how it looks when painted. Therefore, it is important that the temper rolling process is set correctly and measured to ensure desirable properties. However, surface measurements are taken at the final stage of production, using tried and tested, but slow devices. These devices are not capable of measuring the steel's surface at the speed at which it is produced. This means that it takes a long time to discover whether the steel that is being produced has the correct texture or if the rollers needs adjusting.
In this project we aim to define a method of machine learning to detect and correct erroneous surface measurements to allow for inline measurement sensors to be used with confidence and speed during the rolling process. We also aim to define a methodology for classifying texture properties based on the 2D line profiles and 3D depth data from the steel surface.