Search

In many practical situations, it is important to evaluate the relationships between the factors that compose an industrial process and their effects on one or more response variables that are of interest to an enterprise. The main contribution of this present study is to propose a new conceptual hybrid framework based on the DMAIC (Define, Measure, Analyze, Improve, and Control) methodological structure, to optimize complex experimental problems with multiple responses. This procedure combines Response Surface Methodology, with the Desirability (D), Modified Desirability (MD), Compromise Programming (CP) functions, with Generalized Reduced Gradient (GRG) and Evolutionary Algorithms (EA). We made real application to a glass lamination process case study to describe how to use the proposed framework. The procedure allowed several configurations to be tested involving the D, MD and CP functions, adopting the GRD and EV, to optimize the studied industrial process. The best configuration was defined by a practical confirmation experiment and validated by company engineers and experts. As examples of the advantages of adopting the proposed framework in the glass lamination problems, the best solutions resulted in a 49.86% increase in grinding wheel shelf life, corresponding to a 927kg reduction of steel-use per year, and a 41.7% reduction in dressing stone consumption, saving 17,200 stones per year. [ABSTRACT FROM AUTHOR]

The response surface methodology (RSM), which uses a quadratic empirical function as an approximation to the original function and allows the identification of relationships between independent variables xi and dependent variables ys associated with multiple responses, stands out. The main contribution of the present study is to propose an innovative procedure for the optimization of experimental problems with multiple responses, which considers the insertion of uncertainties in the coefficients of the obtained empirical functions in order to adequately represent real situations. This new procedure, which combines RSM with the finite element (FE) method and the Monte Carlo simulation optimization (OvMCS), was applied to a real stamping process of a Brazilian multinational automotive company. For RSM with multiple responses, were compared the results obtained using the agglutination methods: compromise programming, desirability function (DF), and the modified desirability function (MDF). The functions were optimized by applying the generalized reduced gradient (GRG) algorithm, which is a classic procedure widely adopted in this type of experimental problem, without the uncertainty in the coefficients of independent factors. The advantages offered by this innovative procedure are presented and discussed, as well as the statistical validation of its results. It can be highlighted, for example, that the proposed procedure reduces, and sometimes eliminates, the need for additional confirmation experiments, as well as a better adjustment of factor values and response variable values when comparing to the results of RSM with classic multiple responses. The new proposed procedure added relevant and useful information to the managers responsible for the studied stamping process. Moreover, the proposed procedure facilitates the improvement of the process, with lower associated costs. [ABSTRACT FROM AUTHOR]

In the last years, the use of metamodel-based simulation optimization techniques to solve industrial problems stood out as a promising research field, mainly due to the advance of machine learning techniques. The number of metamodeling studies has grown considerably in recent years, but many academics and practitioners still have doubts about which metamodels to choose for their projects. In this way, some studies have compared the effectiveness of metamodeling algorithms. However, they have just analyzed the performance of one or more metrics separately; i.e., they did not analyze the overall efficiency of these metamodels. Basing the metamodels' choice only on one or more metrics empirically might generate biases, causing distortions in decision-making. Therefore, we propose using the multi-criteria data envelopment analysis (MCDEA) model to systematically compare some of the main machine learning algorithms (support vector machine, artificial neural network, gradient-boosted trees, random forest, and Gaussian process). To evaluate the proposed approach, we developed discrete-event simulation models of three real case studies to obtain their input and output data. Moreover, we used machine learning algorithms to train and optimize the metamodels and, finally, new-MCDEA was adopted to compare the metamodels' efficiency considering the associated error, fitting, training and prediction times, and response, among other metrics. Different from traditional comparison approaches, where different algorithms could be chosen depending on the decision-maker bias, the proposed work allowed a good balance between all metrics, and for all cases, the metamodels based on gradient-boosted trees were considered the most efficient. [ABSTRACT FROM AUTHOR]