Case Study 2

Separating the Wheat From the Chaff

 Situation:
When manufacturing process are monitored automatically, they take readings of quality attributes at fixed time intervals. Often these intervals are quite short so that at the end of the run, you have a considerable amount of data. However, when so much data is collected, often it is ‘noisy’ and, as a result, it is hard to draw any conclusions about the underlying process.

Problem:
The following study involves a rubber extrusion process. An extruded gasket material that is rectangular is extruded at a fast rate. The dimensions are monitored by a controller mechanism that makes adjustments during the extrusion in order to keep the height and width on target. The controller tended to over control the process, thereby causing wide swings in the width of the extrusion.

Solution:
Figure 1 shows an SPC chart of 2000 width measurements collected on the process.

Figure 1

SPC (Individuals) Chart of Extrusion Width

 

As one can see, there are quite a few data points above and below the limits, suggesting a lack of control. Also, there appears to be some undulation in the data, although it is hard to see and is buried in the wide swing of the process. If there is some periodicity or undulation, it is obscured by the ‘noise’ in the data.

So how can we determine what is going on with better clarity? One approach is to view the data as a weighted average rather than in it’s raw form. Weighted averages have the property that they smooth out the data and reduce the variability caused by noise in the data. I prefer the exponentially weighted moving average because I can easily adjust the amount of smoothing.

By trial and error, I found the best degree of smoothing and the resulting smoothed data is shown in Figure 2:

 

Figure 2

EWMA Chart of Extrusion Width

l = .05

Notice that the noise in the graph is substantially reduced and, as a result of the smoothing, and an underlying pattern can be seen. The graph has periodicity that is fairly constant in magnitude and frequency across the plot.

As it turns out, the engineering team was able to identify the cause of the periodicity. It was due to the hopper feed mechanism on the extruder that was cycling at about a rate that about corresponds to the peaks and valleys of the EWMA graph. After some adjustments were made, the process smoothed out nicely and attained a better state of control. The width of the extruded rubber was much more consistent along the length of the extrusion.

Details and further information available on request.
Back to Examples of Case Studies
Back to D.S. Schwarz Home page
For how-to-contact information