An experiment is a series of tests. It is the process of us intentionally changing the setting values of input variables in a system in order to observe the reason of output value changes and trends when carrying out these tests. And the design of experiments (DOE) is a systematic method that arranges the combination of conditions in an experiment through statistical methodologies to get reasonable experimental data more efficiently and to increase the reliability of analytical results. Let’s say we are interested to know which material can product a product with the highest hardness. In this experiment, the input variables are the “materials” whereas the output values are “hardness of a product”. The output values that we are interested in are called “raw responses”, the input variables that may affect the responses are called “factors”, and the setting values of a factor are called “levels”. The goal of DOE is to design good test conditions so that the actual factor can be identified from the various potential factors and a level that yields the response closest to the expected value can further be identified. Whether it is to shorten the development cycles of new products or to improve the manufacturing process in order to enhance the quality of products, experiments play a crucial role. Efficient experiment are developed through the design of experiments to explore key parameters that affect product characteristics, realizing the goals of wise cost management, process variation reduction, and yield rate improvement.

Full Factorial Design

The full factorial design includes all combinations of factors. This DOE can get the most sufficient data. It can accurately estimate the main effect of all experimental factors and estimate the highest number of interaction effects between factors. However, with excess number of experiments, the full factorial design oftentimes doubles the cost.

Regular 2-level Fractional Factorial Design

All factors in the regular 2-level fractional factorial design are 2-level, as is suggested in the name of the DOE. The number of experiments can be greatly reduced by half, by one fourth, and by one eighth. Therefore, it is suitable to be used as screening experiments at the initial phase of experiments. The purpose of which is to screen significant factors from massive amounts of factors to reduce the scope of experiments for the next phase. However, one issue is that not all interaction effects can be estimated.

The Taguchi Method is a DOE widely applied in the Industries for its robust parameter design. Its feature is that it reduces the number of experiments and costs while taking into account other uncontrollable factors’ effect on product characteristics (such as environmental factors). The Taguchi Method is helpful for identifying a method that stabilizes the quality of products and is insensitive to noises along the production process.

BBD（Box-Behnken Design）

The Box-Behnken Design (BBD) is an experimental design for response surface methodology. If quadratic relations are found between key factors and responses, the factorial design is no longer sufficient and the response surface design is required. BBD is also a well-known design among response surface methodology with the benefit of small number of experiments.

CCD（Central Composite Design）

The Central Composite Design (CCD) is an experimental design for response surface methodology. If quadratic relations are found between key factors and responses, the factorial design is no longer sufficient and the response surface design is required. CCD is a very efficient design suitable for second-order models.

Orthogonal Design

The orthogonal design provides systematic and efficient ways to change factors, yielding useful statistical information with smaller number of experiments, making it a good compromise between experimental costs and accuracy. However, since not all combinations of levels can be executed, it is possible that the optimal combination is included in the test group; in addition, not all interaction effects can be estimated.

D-Optimal Design

The D-optimal Design aims at minimizing the volume of the regression coefficient vector combined with the trust region, thereby maximizing the reliability of the regression coefficient. Its characteristics are that the number of experiments can be chosen by users themselves and the relations between factors can be limited; for instance, during grinding, “the time of grinding- Factor A” and “the force of grinding- Factor B” cannot exceed a certain value; otherwise the object grinded may break.

We provide comprehensive statistical analysis methodologies. You can carry out all the statistical analyses included in the DOE after completing the experiments and reporting relevant data, and the analyses can assist you in identifying the optimal setting values for factors.

Effects Analysis analyzes the effect of a certain factor on the response after the value level changes. It also observes if interaction effects exisit between factors through interaction plots. Analysis of Variance (ANOVA) divides the total variations in an experiment into multiple sources of variations to further explore the importance of experimental factors. Regression Analysis describes the relations between factors and responses and builds them into models. The models can further make predictions and optimize the manufacturing processes. Composite Contour Charts of response surface methodology overlap contour charts of various responses, allowing users to have a clearer observation of the trends of responses. However, since the charts are that of 2D, users can only observe the relations of two factors and responses each time and other factors need to be fixed at a set value. The Response Variable Prediction Function of response surface methodology, on the other hand, is a mothod that is able to measure multiple response variables at the same time. It transforms multiple response variables into a value through desirability functions and find the optimal setting values of factors by mazimizing such value.