Support Vector machine and Monte Carlo simulation for robust optimization of industrial processes

Goods-producing industries continuously search to improve the quality of final products. The main approach is to identify a correlation between the process settings and the quality of the final product. In this work, a three steps robust approach is presented to improve an industrial process. The first step consists in using a Support Vector machines Regression (SVR) method to build a model of the considered process. It is based on the historic process data defined by an output (a criterion on the product quality) and multiple inputs (various production line settings). Then an optimization step based on an iterative descent method is done on the obtained model to identify interesting settings. Finally the set of settings found is validated by a Monte Carlo simulation approach used to simulate and test settings close to the one found on the optimization step. The proposed regression and optimization methods are compared to existing methods from the literature on a fluidized bed combustion boiler in the context of paper industry. The experiments confirm the efficiency of our approach.