Management

Announcing a Productivity Breakthrough Using Deming & Taguchi Methods in New Product Launches

Using the framework of the System of Profound Knowledge, as established by Dr. Deming, along with Taguchi methods, increases overall success in higher quality and faster cycle times.

3D Printing — *All Images Source: Dr. Mustafa Shraim and Brandon Petrie, OU 3-D Lab Research Assistant*

SUMMARY-RECOMMENDATIONS:

Look to the great thinkers, Dr. W. Edwards Deming and Dr. Genichi Taguchi, for inspiration and guidance.
Dr. Mustafa Shraim, Professor of Engineering Technology & Management at Ohio University, Brandon Petrie, OU 3-D Lab Research Assistant, and Kelly Allan, Trustee of the Board of the W. Edwards Deming Institute, combined the teachings of Deming and Taguchi to achieve new product launches that saved time and money.
Using the framework of the System of Profound Knowledge, as established by Dr. Deming, along with Taguchi methods, increases overall success in higher quality and faster cycle times.
A key is to apply the Study step in the Plan-Do-Study-Act cycle that Deming advocated to study the best settings derived from the Taguchi Design of Experiment, and then test and verify the best settings using control charts.
Prof. Shraim is pleased to share the details of the methods and processes used with readers of Quality.

OVERVIEW & QUICK HISTORY

“Interestingly, the most challenging aspect of accomplishing the productivity breakthrough is not the tools themselves but thinking differently about how to optimize the process for new product launches,” says Prof. Shraim. “Tools matter, of course, but the sequencing of using them also matters. The traditional approach improving only one factor at a time has been helpful to manufacturers through the years, but the results it achieves just can’t compete with the approaches that have evolved since then for product launches.”

The Traditional Approach to New Product Launches

Typically, different approaches were used over the course of dozens and dozens of production launch single-factor iterations to figure out, eventually, the right mix of variables and their settings. The single-factor approach can take weeks of very expensive production equipment and time -- and lots of wasted materials -- to figure out a “recipe.”

Taguchi Methods

Of course, the single-factor approach has been superseded in many cases by the application of Taguchi methods related to multi-factor Design of Experiment (DoE). The marked improvement of results from the multi-factor DoEs over the iterative single factor is well known.

What’s New with the Study at Ohio University

The study at Ohio University demonstrated additional benefits for new product launches by combining Dr. Deming’s insights with those of Dr. Taguchi. For example, our approach very intentionally created and optimized the launch recipe for the new product by applying Dr. Genichi Taguchi’s Methods related to Design of Experiment (DoE) model within a framework provided by Dr. W. Edwards Deming’s System of Profound Knowledge, and specifically the Plan-Do-Study-Act learning cycle that Dr. Deming advocated.

The Experiment in Brief

The challenge was to produce a 6-sided die cube in the 3D printing lab at Ohio University. The characteristics of interest included:

Surface quality of the object represented by a composite quality score with a score of 10 being highest quality.
Dimension of one side of the cube in millimeters
Weight in grams
Cycle time in minutes

A significant aspect of the experiment was to mitigate the usual painful trade-off between fast cycle time and high quality. In this case, cycle time and quality characteristics were optimized very effectively for the best outcomes.

THE FRAMEWORK for the Productivity Breakthrough:

Combining the amazing genius of Dr. Taguchi and Dr. Deming provided an essential framework for effectiveness and efficiency for the new product launch.

Deming’s System of Profound Knowledge (SoPK) contains four elements. We applied them in a variety of ways, including by asking questions. Here are a few that kept us efficient and effective:

Understanding Variation: What is the aim, overall? What is the aim regarding variation?
Appreciation for a System: What are the elements we need to balance? How to minimize tradeoffs?
Psychology: What will be the temptations to tamper with the process?
Theory of Knowledge: How do we design the PDSA cycles? What are our operational definitions of speed, cycle time, quality and so on?

The Plan-Do-Study-Act cycle was especially helpful in deploying Taguchi’s Methods. The following is a summary of how this approach was carried out:

Plan

Decide on the AIM we are trying to accomplish. For example, develop an optimal manufacturing process for a new 3-D printed object.
Decide on what characteristics or response(s) and how they will be evaluated, by whom, and by what method. For example, surface quality, weight, critical dimension(s), cycle time, etc.
Study the process and determine the following:
- Experimental factors: these are the factors that will be experimented with by manipulating the different levels according to the experiment. For example, extruder temperature,print speed, etc.
- Fixed factors or parameters: these are the factors that are just held constant throughout the experiment. For example, infill pattern was held constant throughout the experimental runs.
- Noise factors are the ones that we are not going to control because they are either too hard to control or too expensive to control. Nonetheless, we would like to optimize the process across their fluctuation during production. For example, raw material changing between original (virgin) or recycled (regrind).
Select the appropriate Taguchi experimental design

The “Plan” stage in the 3D printing example
The aim of this study is to find the best 3D printing settings that allow recycled PETG print material to match the performance of original PETG The process for conducting the study can be summarized as follows: 3D-Printer Material: PETG (Original vs. Recycled) Experiment: Taguchi’s L₁₆ Orthogonal Array Analysis: Means & Signal-to-Noise Ratios Printed Object: Sing Die Cube Responses (using original (O) and recycled (R) material): Quality (composite Score): Includes surface quality, edges, clarity of printed numbers, etc. Cycle Time (sec) Dimension (mm) Weight (g) *Taguchi’s L16 Design (A & B at 4 levels, C-F at 2 levels)*
3D Printer	Printed Object (not to scale)

The “Plan” stage in the 3D printing example

The aim of this study is to find the best 3D printing settings that allow recycled PETG print material to match the performance of original PETG

The process for conducting the study can be summarized as follows:

3D-Printer
Material: PETG (Original vs. Recycled)
Experiment: Taguchi’s L₁₆ Orthogonal Array
Analysis: Means & Signal-to-Noise Ratios
Printed Object: Sing Die Cube

Responses (using original (O) and recycled (R) material):

Quality (composite Score): Includes surface quality, edges, clarity of printed numbers, etc.
Cycle Time (sec)
Dimension (mm)
Weight (g)

Taguchi’s L16 Design (A & B at 4 levels, C-F at 2 levels)

3D Printer

Printed Object

(not to scale)

Do

Run the experiment as planned and collect data for the characteristics of interest
Analyze the experiment using factorial plots and signal-to-noise (S/N) ratio plots

The “Do” stage in the 3D printing example
Printing: Each of the 16 combinations was run twice according to its setup, once using original and once using recycled material. For example, the first run (ID #1) was run with settings of Extruder Temp=235 ^oC, Outlines Print Speed= 30 mm/s , Infill%=15, Layer Height=0.15 mm, # of Shells=3, and Travel Speed=50 mm/s Note: This could also be replicated for each type of material if time and other resources allow. Evaluation Each printed object was evaluated for Quality, Cycle Time, Dimension, and Weight as described under the PLAN stage Data was entered in the statistical software using Taguchi’s Experiments and analyzed to: (1) Increase Quality Score, (2) Reduce Cycle Time, (3) Optimize Dimension to 17.0 mm, and (4) Optimize Weight to 3.5 grams. Plots of Response Means as well as Signal to Noise (S/N) Ratios were generated for each Response (Quality, Cycle Time, Dimension, and Weight

Study

Based on the analysis, decide on the initial optimum conditions (settings for the experimental factors that minimize the variation across the levels of the imposed noise factor(s).
Implement optimum conditions from the Taguchi experiment(s) and utilize a control chart for each of the important responses (characteristics).
Identify and properly remove special causes of variation.

The “Study” stage in the 3D printing example
Interpretation: For the Mean plots, the aim is to increase Quality Score, reduce Cycle Time Score, Achieve targets of Weight =3.5 g and Dimension of 17.0 mm. For reduction in variation, we aim to increase the S/N ratio for all characteristics For competing objectives (e.g., when reducing cycle time compromises Quality score), trade-offs must be made. Based on Mean and S/N plots, best settings were determined (considering trade-offs between competing objectives). For characteristics with nominal-is-best (namely Dimension and Weight), we select the factor that has the least S/N ratio difference between its levels and locate a setting closest to target value on the Mean plots. This was determined to be Factor C: Infill%. Best Settings: Factor A: Extruder Temp @ 250 ^oC Factor B: Print Speed @ 50 mm/s Factor C: Infill % @ 28% (moved for optimization) Factor D: Layer Height @ 0.4 mm Factor E: Number of Shells @ 5 Factor F: Travel Speed @ 50 mm/s Implement Best Settings: Monitor with control charts. The example shown below is an Individual and Moving Range (I-MR) chart for Cycle Time, but a control chart should be used for each of the other characteristics of interest (Quality Score, Dimension, and Weight) Identify and remove special causes of variation as they occur. In the control chart below, one out-of-control point was identified in the chart below (Point #10). Special Cause: 3D printer took longer time to heat up since it was printed at 7 am and window was open in lab prior night - and thus inflated the Cycle Time (CT). Countermeasures were applied to ensure consistent ambient temperature and subsequent points indicate more consistency.

The “Study” stage in the 3D printing example

Interpretation:

For the Mean plots, the aim is to increase Quality Score, reduce Cycle Time Score, Achieve targets of Weight =3.5 g and Dimension of 17.0 mm.
For reduction in variation, we aim to increase the S/N ratio for all characteristics
For competing objectives (e.g., when reducing cycle time compromises Quality score), trade-offs must be made.
Based on Mean and S/N plots, best settings were determined (considering trade-offs between competing objectives).
For characteristics with nominal-is-best (namely Dimension and Weight), we select the factor that has the least S/N ratio difference between its levels and locate a setting closest to target value on the Mean plots. This was determined to be Factor C: Infill%.

Best Settings:

Factor A: Extruder Temp @ 250 ^oC
Factor B: Print Speed @ 50 mm/s
Factor C: Infill % @ 28% (moved for optimization)
Factor D: Layer Height @ 0.4 mm
Factor E: Number of Shells @ 5
Factor F: Travel Speed @ 50 mm/s

Implement Best Settings:

Monitor with control charts. The example shown below is an Individual and Moving Range (I-MR) chart for Cycle Time, but a control chart should be used for each of the other characteristics of interest (Quality Score, Dimension, and Weight)
Identify and remove special causes of variation as they occur. In the control chart below, one out-of-control point was identified in the chart below (Point #10).

Special Cause: 3D printer took longer time to heat up since it was printed at 7 am and window was open in lab prior night - and thus inflated the Cycle Time (CT). Countermeasures were applied to ensure consistent ambient temperature and subsequent points indicate more consistency.

Act

If needed, further optimize the process once stable and predictable – free of special causes. Specifically, reduce common-cause variation and/or move the process towards target using additional PDSAs.
When moving the process towards target for any characteristic:
- Utilize an input factor or parameter that has no direct interaction with any other input factor
- Watch for impact on other characteristics and implementing trade-offs when necessary (e.g., cycle time vs. surface quality)

IN CONCLUSION

Dr. Deming and Dr. Taguchi provide us with great guidance –if we are willing to learn from them.
Quality techniques like Design of Experiments, and SPC can work well together, especially when sequenced effectively –being guided by the teachings of the masters.
Just doing the DoE is not enough. Even if you get your best values in the best conditions, you still might have special causes entering, over time. As you implement the SPC, you want to watch for special causes.
Once you have true stability, reliability and predictability, then you want to start thinking about further improvement.

Related References:

Deming, W. E., The New Economics, 3^rd ed., The MIT Press, Cambridge, MA, 2018.
Moen, R., and Norman, C., The history of the PDCA Cycle, Proceedings of the 7th ANQ Congress, Tokyo, 2009.
Ross, P. J., Taguchi techniques for quality engineering, 2nd ed., McGraw Hill, New York, NY, 1996.
Scholtes, P., Joiner, B., and Streibel, B., The Team Handbook, Oriel Inc., Madison, WI, 2003.
Taguchi, G., Chowdhury, S., and Taguchi, S., Robust Engineering, McGraw Hill, New York, NY, 2000.
Wheeler, D., and Chambers, D., Understanding Statistical Process Control, 3^rd ed., SPC Press, Knoxville, TN, 2010.

LEARN MORE:

Looking for a reprint of this article?
From high-res PDFs to custom plaques, order your copy today!

Dr. Mustafa Shraim is a professor of Engineering Technology and Management at Ohio University. If you need more details or have questions, please email Professor Shraim at [email protected].

Kelly Allan is a trustee of the board of the W. Edwards Deming Institute.