A Real World Application of Design of Experiments

Black Belt students are exposed to Design of Experiments (DOE) in their classroom training, but in my experience very few of them actually apply the technique while completing a project to earn their Black Belt certificate.  One of my students recently completed a project that made excellent use of a Designed Experiment, and his company gave me permission to share the results of his work in this article.

The process that needed improvement was the filling of 50 pound bags with a powdered product.  The current process performance was determined by selecting a sample of four bags in a row every two hours for a period of two weeks and creating Xbar and R control charts.

The mean weight of a bag before improvement was 50.57 pounds, with an upper control limit of 50.77 and a lower control limit of 50.36, and the process was not stable as evidenced by the following control chart.   The suspected cause of special cause variation was the density of the product being packed.

The target for after improvement was a stable process with a mean bag weight of 50.20 pounds, a lower specification limit of 50.00 pounds, and an upper specification limit of 50.4 pounds.  The financial impact of reducing the amount of overweight was determined to be $50M per year for every .10 pounds of reduction, making the potential impact of this project as much as $200M per year.   In addition, the Voice of the Customer indicated that customers were unhappy with inconsistent bag weights.  The brand of the company was suffering because the customer viewed the inability of the company to provide consistent bag weights as “dishonest”.

A process capability study prior to improvement showed that the process was clearly not capable of meeting the target specification.

The project began in May 2016 and was concluded by October 2016.  One of the issues identified early in the project was a lack of detailed knowledge regarding how the packaging equipment was programmed and how all of the components of the equipment communicated with each other to achieve a target mean weight.  Subject matter experts from the equipment supplier were added to the team to address this deficiency.

Prior to conducting the experiment it was verified that the scales being used to weigh bags had been calibrated by the scale supplier.  The scales were also being checked weekly by the packaging operators using NIST traceable certified weights.

A long list of factors that could potentially impact the weight of the bag was evaluated.  After much team discussion these factors were sorted into three categories:  Hold constant, noise and factors to experiment with.  There were eleven hold constant factors, all related to various equipment settings.  There were two noise factors, ambient temperature and ambient humidity.

The three factors that were chosen for experimentation were the coarse fill weight setting, the final weight setting and the Waversaver setting.  The Waversaver is a combination of hardware and software that allows accurate weight readings despite vibration that is caused by various factors in the process.  The Waversaver may be set to average multiple weights of the same bag over time frames varying from 60ms to 2.5 seconds.

Each factor was evaluated at two levels.  The levels for the coarse fill weight were 49.4 pounds and 49.6 pounds; the levels for final weight setting were 50.0 pounds and 50.2 pounds; and the levels for the Waversaver setting were 1.0 seconds and 3.5 seconds.

A full factorial experiment was conducted, and the resulting data was analyzed using Minitab.  The overall R squared value of 82% indicated that the model was a good fit as far as predicting outcomes.  The Half Normal Plot of Standardized Effects chart showed that both the final weight setting and the Waversaver setting have a significant effect on bag weight.

The Response Optimizer tool in Minitab was used to determine optimal settings.  The results are as follows:

A confirmation run was performed with the recommended optimal settings of 49.6 pounds for coarse fill, 50.20 pounds for final fill and 2.11 seconds for the Waversaver setting.  Ten bags in a row were weighed each hour for 25 hours of production time.  The confirmation run resulted in a mean bag weight of 50.186 pounds with very little variation around the mean.  The recommended setting were documented in standardized work instructions, all involved personnel were trained in the new procedures and the improved process was implemented in production.  A subsequent process capability study conducted during October 2016 showed a process mean of 50.2024 pounds and a Cpk of 1.48, indicating that the process is now highly capable of meeting the target requirements.

The following Xbar and s control charts, created using the same data, showed that the process after improvement is stable.  In addition, the fact that there is very little variation around the mean and that the lower control limit is 50.16 pounds indicates that further improvement is possible.

The level of improvement was clearly significant from a practical standpoint.  A two-sample t-test was conducted and confirmed that the improvement was also of statistical significance.

Your comments or questions about this article are welcome, as are suggestions for future articles.  Feel free to contact me by email at roger@keyperformance.com.

About the author:  Mr. Roger C. Ellis is an industrial engineer by training and profession.  He is a Six Sigma Master Black Belt with over 49 years of business experience in a wide range of fields.  Mr. Ellis develops and instructs Six Sigma professional certification courses for Key Performance LLC.   For a more detailed biography, please refer to www.keyperformance.com.