Modellazione e simulazione di un sistema di produzione di vetro mediante algoritmi di Machine Learning. = Modelling and simulation of a glass production system with Machine Learning algorithms.

The present study aims to model and simulate the behavior of a flat glass annealing lehr, by means of data analysis algorithms. In the framework of “Industry 4.0”, the availability of more production data and the need to find correlations between process variables, enhance the use of Machine Learning to generate models that are able to predict a process or equipment outputs, given certain inputs. The target of this work is to create a regression model that predict the permanent stress of a glass sheet (output) generated in the annealing phase of the controlled cooling. The inputs considered are the setting of the machine and other working variables such as production parameters, initial and boundary conditions. The pipeline of this analysis include a first preprocessing phase: the data are evaluated and treated in terms of features engineering, distribution, outliers and scaling. Once the dataset is prepared, it can be used for the training: in this phase the Support Vector Machine (SVM) algorithm operates on a subset of samples to generate the regression model, then is tested on another subset to validate the results. In order to improve the model accuracy, a tuning of the parameters and its Cross validation are performed. The purpose of the model in real production, is the identification of eventual deviation in the quality output, before the quality check is effectively done and even with a higher frequency; furthermore it can be used to simulate different equipment settings without doing online test.