This guidesheet describes the important steps for planning industrial experiments. These steps are crucial in order to ensure that the experiments will produce information that can be used to improve a given system (i.e. a manufacturing or a design process). Improperly, planned experiments, when carried out, can lead to experimental results that are inaccurate and conclusions that are misleading.

Steps 1 to 5 represent the major steps in the planning process. Steps 6, and 8 represent the execution, analysis, and conclusion and recommendation steps. These 3 steps have been included in order to illustrate the activities that take place after the planning phase.

The use of a team approach for planning experiments will enable individuals with various expertise and knowledge to work together in order to achieve the objective(s) of the experiment. The team should made up of relevant personnel such as the line engineer, supervisor, technicians, and operating personnel. Operating personnel often have much insight regarding the process and the likely sources of the process variability, and therefore should not be overlooked. The inclusion  of the relevant manager can help ensure that the experiment has the necessary support of management in terms of obtaining the required resources (material, equipment, time, etc.) for carrying out the experiment.

Once the team has been formed, it is necessary for the members to develop a clear idea of what the objective(s) of the experiment should be. The participation of all members is needed in order to accomplish this. The use of historical data such as data from previous experiments as well as data that is being collected on a routine basis (daily yield data, customer complaint data, etc.) can be used to help the team determine the objective(s). Such data is also useful for steps 3 and 4 of the planning process.

The response is the outcome of the system which is to be improved.  Usually, an important measure of product quality or process efficiency or both such as yield, tensile strength, percent shrinkage, assembly time, cycle time, cost, etc. is used as the response. More than one product quality or process efficiency measure can be used (i.e. multiple responses).

A response can be either be continuous (such as tensile strength, percent shrinkage, concentration, etc.) or discrete (such as good or bad, pass or fail, meets specification or otherwise, etc.). Continuous responses are generally preferred since they contain more information regarding the outcome such as knowing if the tensile strength is within specification or not.

It is important that the measurement system that is used for measuring a continuous response is reliable. Methods for verifying the reliability of measurement systems can be found in AIAG (1990) and Peace (1993). the inspection criterion that is used for obtaining the discrete response should also be clearly specified.

This step requires the members to choose the factors which are to be manipulated in the experiment and their setting (henceforth referred to as 'levels'). As in the case of the response, a factor can either be of a continuous nature (i.e. quantitative factors are temperature, time, and concentration. Examples of qualitative factors are vendor, position, and instruments. Typically the step is the most difficult and time-consuming part of the planning process. The following guidelines are intended to facilitate this step.

1. Firstly, a list of potential factors for the experiment should then be grouped into 'control' factors and 'noise' factors. Control factors are factors that cannot be easily set or are impossible to set to a constant value during the entire experiment. Examples of noise factors are tool wear, user conditions, batch number, environmental conditions, etc.

2. The next step is to review the factors on the diagram, and to determine which additional factors should be incorporated and which factors should be deleted.

3. The factors should then be grouped into 'control' factors and 'noise' factors. Control factors are factors that can be easily set and maintained. Noise factors are factors that cannot be easily set or maintained. Noise factors are factors that cannot be easily set or are impossible to set to a constant value during the entire experiment. Examples of noise factors are tool wear, user conditions, batch number, environmental conditions, etc.
Coleman and Montgomery (1993: 6) mentions a third category of factors, known as 'held-constant' factors. The latter are control factors whose effects are not of particular interest. Such factors are held at a particular value during the experiment.

4 Having grouped the factors into the above three categories, the team members must decide on the number of levels for the control and noise factors, and to also specify the values for the levels. At this step, the members may find it more appropriate to re-categorize some of the control factors as noise factors.

5. Trial runs should be conducted to determine which values of the factor levels will result in bad product or infeasible operating conditions. Such values should be avoided in the experiment. As for the number of levels, at least two levels are needed. In the case of 'variable screening', two levels set at the extreme boundaries of the feasible operating range is adequate.

6. Whenever possible, the members should discuss which pares of interactions are likely to be important, and therefore should be investigated in the experiment. Coleman and Montgomery (1993): 10) suggests that the process of elimination and inclusion be used for this purpose. This can be done by posing the questions "Are there any interactions that must be estimated clear of main effects?"; "main effect" here refer to the effects of individual factors.

During this step, the team members should determine the details of the individual experimental runs, the order of performing the runs, and the number of replications for the experiment (i.e. the number of times the experiment is to be repeated). In order to reduce the effect of unknown factors on the experimental results, it is recommended that the individual experimental runs be carried out in a random order (i.e. 'full randomization'). However, problems of logistic and cost may prevent this from being carried out, in which case, alternatives to the use of full randomization has to be looked into. Similarly, problems of logistic and cost may prevent a large number of replications.

At this point in time, the team members must choose an appropriate experimental design for the experiment. This step requires knowledge of experimental designs and knowledge gathered from steps 2 to 4. Decisions pertaining to randomization and replication will also affect the choice of experimental design.

The amount of time required for carrying out the experiment as well as the schedule for the experiment should also determined.

A co-ordinator should be appointed among the team members to ensure that the experiment is carried out according to plan, and to highlight cases of deviations to other team members. All personnel involved in the actual conduct of the experiment (operators, testers, inspectors, etc.) should be properly informed as to how the experiment is to carried out and how the responses are to be obtained and recorded.

The experimental results can be analyzed using graphical and more formal statistical methods. The choice of experimental design will determine the choice of formal statistical methods. The latter requires knowledge of formal statistical methods.

The use of available commercial softwares such as Microsoft Excel and Minitab can aid in the analysis of the experimental results. The use of graphical methods can aid team members in the explanation of the experimental results to upper management and non-technical personnel.

Based on the information obtained from the analysis of the experimental results, the team members should then draw conclusions and recommend an appropriate course of action. The latter can include confirmation runs and the planning of subsequent experiments.

Though it is often said that no experiment goes exactly as planned, careful attention to detail during the planning  and  execution  stages  of  the  experiment  are met. No doubt, the combined  expertise  and  co-operation of the various team members is an important pre-requisite for the careful planning and execution of the experiment.

[1] AIAG (Automotive Industry Action Group) (1990) Measurement Reference Manual, Troy , Michigan .

[2] Coleman, D. E. & Montgomery, D. C. (1993) A Systematic Approach to Planning for a Designed Industrial Experiment (with discussion) , Technometrics, February, Vol. 35 No. 1, pp. 1-27.

[3] Peace, G. S. (1993) Taguchi Methods; A Hands-On Approach, Addison-Wesley Publishing Company, Massachusetts .

[4] Wu, C. F. Jeff & Hamada, M. (2000) Experiments; Planning, Analysis, and Parameter Design Optimization, John Wiley & Sons, inc., U. S. A
.