Using Taguchi and neural network approaches in the optimum design of product development process

In product design, accuracy of the product information greatly affects design quality. The Taguchi method simplifies the analysis method and provides an effective product design approach by confirming the variable characteristics and determining the optimum combination of characteristics. However, determination of the levels of the variable characteristics in the Taguchi method relies on human experience and might not achieve the optimum situation. The research objective is to use evolutionary neural networks into a robust product design to help designers search for an optimum combination of variable characteristic values for a given product design problem. In the design procedure, the data resulting from the experimental design in the Taguchi method are forwarded to the back-propagation network training process and genetic algorithm simulation to predict the most suitable combination of variable characteristic values. The recommended combination of variable characteristic values is represented in 3D form. A design case of a lat bar for pull-down fitness stations is used to demonstrate the applicability of the design procedures. Note that the signal-to-noise ratios are derived from experiments that measure the muscle responses using an electromyography (EMG) apparatus. The results indicate that the proposed procedures could enhance the efficiency of product design efforts.