Machine Learning applied to prediction of shape defects in round cross section rolled bars

During the rolling of round bars, monitoring status of main mill components is fundamental to detect in advance eventual malfunctions that could lead to product shape and quality problems and to allow the implementation of suitable and immediate countermeasures. In this work, the problem of ovality, that occurs when the section of the bar is not perfectly circular, is addressed. To this aim, a system for the collection and the management of plant data was set up based on a hardware and software architecture capable of managing in real time a large amount of information. Further, advanced machine learning techniques are used to predict ovality from vibration signals measured by a set of sensors installed in suitable positions throughout the rolling mill. The resulting analysis allowed the identification of the most critical conditions for ovality occurrence and the optimization of the mentioned HW/ SW architecture.