Applying Design For Six Sigma

Design for Six Sigma. How to identify, assess and mitigate variation throughout the product development process

Design For Six Sigma (DFSS) What is DFSS?

A specific product design methodology where Customer Requirements dictate the critical parameters (CTQs) (CTQ = Critical To Quality) and the variability of the critical parameters are optimised for predictive product performance, manufacturability and reliability’ DFSS is an enhancement to your product development process, not a replacement for it.

Relative Cost/Difficulty to Correct Quality Problem

DFSS Methodology : 4 step
Short description

DFSS - Main activities & Tools

DFSS Methodology What is the definition of Critical-to-Quality?

The selected few, measurable key characteristics of a specific part/process/specification that must be in statistical control to guarantee customer satisfaction. Therefore, they must be measured and analysed on an on-going basis.

Defining the Transfer Function: Exact Definition

Approximating the Function

• Regression (Passive Data Collection)
• DOE - Planned changes in KPIV’s in order to observe corresponding changes in KPOV’s

DFSS Scorecard as final output of the process

• We capture the Product CTQ’s (Y) in the top level DFSS scorecard
• Now we set a target and tolerances on the KPIV’s (inputs) and record them in our input variable scorecard
• For sure some inputs come from our suppliers. We have to track those as well
• Through the transfer function we predict the defects per million of our product CTQ’s
• Predicting scrap and rework (defect) levels allows us to predict costs (business decisions)

In conclusion . . .

Example of DFSS for a subsystem

(Application of Robust Design, Tolerance Design, MonteCarlo simulation)

Design phase

We are faced with the task of designing a pump capable of delivering a constant flow rate of 10 l/min. Customer requires a pump with ‘6 sigma’ performance with a flow rate between 9 and 11 l/min.

DATA for DESIGN and MANUFACTURING PROCESSES

MonteCarlo Simulation

How much variation in Y (output) is created by variation in X (inputs) and the system function Y = f (X)

• Random values of X are generated and applied to function Y = f (X) to predict the variation of Y
• Transfer function Y = f(X) can be non-linear
• X’s can be any distribution
• May require a lot of trials (~1000) and results are not repeatable

Optimise phase

The result achieved in the ‘solution 1’ is not in line with the goal: a pump with six sigma performance. Too much variation in the output : in fact Cpk is 0.79 (Six sigma means Cpk = 1.5)

It may be possible to reduce the output’s variation by adjusting the average of the inputs (if there is non linear relationship between inputs and output)
Adjusting the averages can make the pump less sensitive or more ROBUST to the variations of the inputs. Since it is generally less costly to adjust targets than tighten tolerances, this step has the highest priority.
This was done in the ‘solution 2’ where practically we decrease the radius and we increase the stroke lenght.
We got an improvement in the output variation. Cpk is 1.30.

If the desired performance has not been achieved, the next step would be to tighten tolerances.

The question is which tolerances and by how much?
In deciding which tolerances to tighten, you should consider both the cost and the effect of change.

CONCLUSIONS

• We designed a pump with six sigma performance and we were able to predict in each step (solution 1, solution 2, solution 3) the number of defective products (in ppm). It is a business decision in which step to stop the optimization;
• Knowing the transfer function Y = f(X) between input and ouput we were able to simulate the pump performance without building prototypes;
• With MonteCarlo simulation the inputs are not a single value, but a distribution: these combinations of input variables are used to calculate a large number of outputs for which a distribution is generated;
• Among the 4 variables, only ‘N’ is a ‘critical characteristic’ for the pump and for it we decided the tolerances and we must include it in our control plan. For the other 3 we can use ‘economic’ tolerances.