The Optimization of the Zinc Electroplating Bath Using Machine Learning and Genetic Algorithms (NSGA-II)

Abstract

In this study, our aim is to predict the compositions of zinc electroplating bath using machine learning method and optimize the organic additives with NSGA-II (Nondominated Sorting Genetic Algorithm) optimization algorithm. Mask RCNN was utilized to classify the coated plates according to their appearance. The names of classes were defined as “Full Bright”, “Full Fail”, “HCD Fail” and “LCD Fail”. The intersection over union (IoU) values of the Mask RCNN model were determined in the range of 93–97%. Machine learning algorithms, MLP, SVR, XGB, GP, RF, were trained using the classification of the coated panels whose classes were detected by the Mask RCNN. In the machine learning training, the additives in the electrodeposition bath were specified as input and the classes of the coated panels as output. From the trained models, RF gave the highest F1 scores for all the classes. The F1 scores of RF model for “Full Bright”, “Full Fail”, “HCD Fail” and “LCD Fail” are 0.95, 0.91, 1 and 0.80 respectively. Genetic algorithm (NSGA-II) was used to optimize the compositions of the bath. The trained RF models for all the classes were utilized as the objective function. The ranges of organic additives, which should be used for all the classes in the electrodeposition bath, were determined.