Proceedings of 1 st Ahmad Dahlan International Conference on Mathematics and Mathematics Education Universitas Ahmad Dahlan, Yogyakarta, 13-14 October 2017 148 Development of Dual Response Approach using Artificial Intelligence for Robust Parameter Design Tritiya A. R. Arungpadang 1 , Benny L. Maluegha 2 , and Lily S. Patras 3 1,2 Department of Mechanical Engineering, Sam Ratulangi University, Manado, Indonesia 3 Department of Electrical Engineering, Sam Ratulangi University, Manado, Indonesia E-mail : tritiya_arungpadang@unsrat.ac.id Abstract. Prediction process of parameters in robust design is very important. If the prediction results is fairly precise then the quality improvement process will economize time and reduce cost. Dual response approach based on response surface methodology has widely investigated. Separately estimating mean and variance responses, dual response approach may take advantages of optimization modeling for finding optimum setting of input factors. A sufficient number of experimentations are required to improve the precision of estimations. This research recommended an alternative dual response approach without performing experiments. A hybrid neural network-genetic algorithm has been applied to model relationships between responses and input factors. Mean and variance responses conform to output nodes while input factors are used for input nodes. Using empirical process data, process parameter can be predicted without performing real experimentations. A genetic algorithm has been applied to obtain the input factors optimum setting. An example has been studied to demonstrate the procedures and applicability of the proposed approach. 1. Introduction Robust design is an important engineering design methods for quality improvement. It can reduce manufacturing cost and design cycle time. Robust design is a cost-effective methodology for determining the optimal settings of control factors that make the product performance more sensitive to the effects of noise factors [1]. Robust parameter design (RPD) is a method to determine the optimal conditions of input variables so that give the optimal response. RPD introduced by Taguchi. He suggested the use of orthogonal array and SN (signal-to-noise) ratio using the inner array to factors beyond the control and outer factors for noise factor. The combination of the best design parameters are determined by minimizing the SN ratio. Besides the average value endeavored to achieve the desired target by identifying adjustment factors [2]. Offering a more statistically sound and efficient approach to data analysis, a dual response approach to RPD by combining Taguchi's philosophy and response surface methodology (RSM) was suggested [3]. A dual response approach may offer considerable modeling flexibility by giving an estimate of process mean and standard deviation at any point of interest. Separately modeling response functions for process mean and variance, an engineer can gain insights on the relationship between input variables and responses [4-7]. A second-order polynomial model is mostly assumed for predicting response functions in RSM and it is often the case that the behaviour of mean or variance responses may not be described well by a reasonable second-order polynomial model. In practice, the fitting and predictive performances of low-order polynomial models and especially of the process variance response are very poor when the relationship between the input factors and the quality characteristic of the process is highly nonlinear and noisy [8]. For example, there is disputed that the prediction of a pharmaceutical response based on a second-order polynomial equation often produces in the poor estimation of optimal drug formulation [9].