This was part of an overall deployment of Six Sigma methods for problem-solving and design. For this particular project, we used the D²C² design method.

Define

The team re-examined the needs of both internal and external customers by building a Quality Function Deployment table linking customer desires with product attributes. The team leader stated that this document fundamentally changed how he would approach new product design in the future, and that for all the technical wizardry that was to come, the QFD was the most important step. The QFD identified a number of product characteristics, including some not previously considered, and assigned relative importance to each. Measures were found that correlated with these characteristics and were incorporated into the testing criteria.

Design

The product composition had already been estimated and tested in the lab before the re-chartering of the project as DFSS. Although the QFD did not suggest a change this composition, the new knowledge gained during the Six Sigma Black Belt class led the project leader to consider using experimental design to optimize the composition. After considering the likely interactions between eight components, a highly fractionated saturated design was constructed to test the effect on seven critical output variables (identified in the QFD) including the performance of the product after exposure to head and humidity. An L₁₆ design was found that could test all of the effects of interest. An appropriate sample size was calculated, and the experiment was performed in the lab at bench scale.

Design projects commonly have a number of different outputs that must be optimized and balanced, and each output had a different weight, again based on the QFD. In order to pick the best combination of product characteristics, a polynomial model of the response surface was generated which would predict the seven output characteristics for any input composition. A numerical method was used to weight the different outputs, and an “Overall Quality Factor” was calculated, which was used as an indication of customer-experienced quality. This number was manipulated using a solver function to find the theoretical best and worst compositions (neither of which were actually run as part of the original experiment). The polynomial model was used to estimate the product characteristics of these two compositions, as well as the original nominal composition. These compositions were then made in the lab and tested to validate the model. The actual results were found to agree very closely with the predictions for the original experiment, including the fact that the original composition would have failed in production as the summer months approached. The theoretical optimum exceeded all expectations by not only resisting the decreased simulated summer performance in all the critical measures, but exhibiting an increase for some measures in the presence of heat and moisture!

The optimum composition was tested on production equipment, and was again found to conform very closely to estimates from the polynomial model. Also during this trial, the team worked closely with production to insure that production had no problems in manufacturing the product.

Commercialize

Commercial production commenced once the product had been qualified on production equipment. During the initial production runs, the DFSS team observed manufacturing and aided in fine-tuning the product so that it ran smoothly during production ramp-up. Additionally, technical experts, including the design chemist and team leader, were present during the first customer uses. The customers were ecstatic, as the product exceeded all critical customer requirements. The internal feedback was likewise high praise, as the product maintained its performance through the summer months, unlike its predecessor product.

Control

The performance of the product continues to be monitored, and the output characteristics are, if anything, better than the predictions as time passes. Calculations have not yet been done on the avoided cost had the original formulation been used as planned. These costs include direct scrap and rework costs, overtime, and additional machine burden rates during the summer months when the original formula would have had difficulties meeting product specifications. During its first year of production, the new polymer product earned $2.8 million and continues to expand its market.

The customers were ecstatic about the final product, especially the Asian manufacturers since the new product actually increased in performance after being exposed to heat and humidity. The product made a big impression in the market, since no other product met and exceeding the market requirements like this one. In addition, our client saved money because they no longer had to schedule overtime during the summer months to make product (most of which was scrapped) in order to fulfill orders.

Below, you can see a screenshot of the actual experimental results being used to optimize four quality measures (these measures and setting names have been changed to protect the intellectual property of the client). As the settings are changed, the average and spread of the measures are predicted and can be optimized to reduce the overall cost to the manufacturer and customer. This is the best setting, and as is common in the real world, there is no optimum setting that gets everything on target. Using the Taguchi Loss Function and realistic costs for each of the criteria, we can find the best possible settings to use in production. A similar approach was used for other discrete measures of quality.