A simulator based on artificial neural networks and NSGA-II for prediction and optimization of the grinding process of superalloys with high performance grinding wheels

Computational intelligence (CI) has been applied to grinding processes in order to perform more and more efficient operations. Among the techniques that compose the area of CI, it is possible to highlight the artificial neural networks (ANN) and the multiobjective optimization algorithms, such as the NSGA-II. These tools are useful to the computational modeling of several processes. However, until now, the hybrid use of these techniques has not been explored for external cylindrical grinding of superalloys. Filling this gap, this work had the objective of developing a methodology that uses a multilayer perceptron (MLP) ANN associated to the NSGA-II, in the form of an objective function. The method optimizes the grinding process of superalloys with the use of two grinding wheels, one conventional and another cBN superabrasive. The methodology consisted of three phases: the first one was of data collection through experiments for the construction of a dataset, containing input configurations of a grinding process related to output parameters that determine the quality of the machining. These parameters, in addition, are also objectives to be minimized when this problem is mathematically modeled. The second one dealt with the training and validation of an MLP ANN as the simulator of this grinding process, and finally, the third, dealt with the generation of optimized solutions by means of the NSGA-II associated with the MLP ANN already trained. Regarding the results, it could be demonstrated that the simulation does not present evidence of statistical differences when compared to real data. As far as optimization is concerned, the non-dominated solutions presented values, in several scenarios, consistent with the literature of the area. In this sense, this work brings as its main contribution a hybrid, scalable computational method that can be used as a tool for decision making. It is also concluded that it can be useful in the planning of economically efficient operations, since the simulation does not involve logistic costs. (c) 2020 CIRP.