Adhesion strength is one of the most significant quality characteristics for coating performance. Heat treatment and sanding process parameters affect the adhesion strength. The aim of this study was to predict the adhesion strength using machine learning and optimization algorithms. Process factors were selected such as temperature, time, cutting speed, feed rate and grit size while coating performance index was selected as adhesion strength. Adhesion strength values of the specimens were determined by employing pull-off adhesion-type equipment. Firstly, central composite design with analysis of variance was used to create the experimental design and to determine the effective factors. Moreover, the main effect plot was used to determine the values of effective factors. Then, support vector machine (SVR) and extreme learning machine (ELM) were used to predict the adhesion strength. Finally, genetic algorithm was applied to optimize the parameters of SVM and ELM in order to improve the prediction accuracy. The proposed hybrid SVR-GA and ELM-GA approaches were compared with linear regression (LR), SVR and ELM. Experimental results showed that the proposed SVR-GA and ELM-GA approaches outperformed the LR, SVR and ELM in terms of prediction accuracy.

粘合强度是涂层性能最重要的质量特征之一。热处理和打磨工艺参数会影响粘合强度。这项研究的目的是使用机器学习和优化算法来预测粘合强度。选择工艺因素，例如温度，时间，切削速度，进给速度和粒度，同时选择涂层性能指标作为附着强度。样品的粘合强度值通过采用剥离粘合型设备测定。首先，采用带有方差分析的中心组合设计来创建实验设计并确定有效因素。此外，主要影响图用于确定有效因子的值。然后，支持向量机（SVR）和极限学习机（ELM）用于预测粘附强度。最后，采用遗传算法对SVM和ELM参数进行优化，以提高预测精度。将提出的混合SVR-GA和ELM-GA方法与线性回归（LR），SVR和ELM进行了比较。实验结果表明，提出的SVR-GA和ELM-GA方法在预测精度方面优于LR，SVR和ELM。