Parameter design is intended to conduct an experiment on multiple control factors and to distinguish the control factors that may affect the variation from the control factors that may affect the average value.
5.2 Conventional Two-stage design
In the case that there are four conditions, optimum conditions can be determined as follows:
If a certain target value is given, optimum conditions are determined one by one based on multiple conditions – a conventionally used single-factor experiment. In optimizing conditions using this single-factor experiment, a design condition A is to be optimized first and other conditions are fixed. This design condition A is drifted through about three levels and a condition A1 that is close to a target value can be obtained. A condition B is then drifted with the condition A1 fixed and a condition B2 can be obtained. With the condition A1 and B2 fixed, a condition C1 and a condition D3 can be obtained. Target conditions are extracted from this process and allowable tolerances of each individual condition are determined to reduce the variation.
If obtained values are still outside the target value after conditions are processed in steps described above, some adjustments must be made:
Using tighter allowable tolerances for A, B, C and D Increasing the level of parts
Using high-priced material, etc.
Using this conventionally used approach that is called the two-stage design, conditions are first processed to fit results in set target values, then adjustments are made to decrease the variation.
5.3 Two-stage design for Parameter Engineering
Parameter design is based on the use of nonlinearity. In using robust design to optimize conditions, variation instead of target values is first taken up as an object to which nonlinearity effects are to be applied. Because variation is the first point to consider, A3, C3 and D2 are chosen for A, C and D, respectively. Such a factor as B does not vary widely and therefore remain unaffected under any given condition.
In the first step where variation must be minimized, a combination of A3, C3 and D2 is chosen. This combination is expected to reduce the variation in characteristic values. Although variation can be reduced, it is still far off target values.
To bring the variation close to target values, average values must be adjusted to target values. In this case, it is necessary to use such a factor as B since the use of B allows average values to fluctuate while variation remains fixed. If B3 is chosen, average values can be brought down and desired characteristic values can be obtained.
Because variation is first minimized using parameter design, a combination of conditions can be found that can keep the variation in characteristic values to a minimum though low-priced parts or parts with a wide range of variation are used. After variation is minimized, conditions are adjusted to bring them close to average values.