SPC setup

Hi,

We are producing audio headset. At the end of production line we performed a quality test which consists of measuring the 'Total Harmonic Distortion' (THD) of each headset. The test consist of measuring the 'THD ratio' over a range of frequencies (for e.g. from 500 - 4500 Hz).

I would like to setup a SPC on the THD. Since I do not have one value per headset but instead a series of results I wonder which SPC chart will be applicable in this situation? Based on historical results it is observed that there are only a few frequency ranges where we have seen failures. We have only an USL on the THD and the lower the THD the better it is.

Thanks

Don't you put a "pass/fail" result on the test?

Two dimensional characteristics can be challenging. Is there a specific point that is very important? Either at a specific point in the profile or perhaps a peak value regardless of where in the profile it occurs? This is the approach I have taken (very successful) when confronted with such a. Measurement.

On a graph like this, I could see the peak THD Ratio as one and the frequency of the XX percentile of THD Ratio (i.e., breakpoint of the drop-off) as the second.

Thanks everyone for your comments and actually i have learned something new about the process. The specs levels are not same across all the frequencies but it changes throughout the range (see attached picture). The straight red line is the specs limits. And each other line represents one part.

I can think of following two possibilities

i). Split the results into smaller frequency ranges which has the same specs limits.
ii). And then plot the peak value from each range into a separate XMR chart.


a. Plot the difference between peak value and the spces limits on a single XMR chart.

I think plotting the average value of each range into a X-Bar S chart will hide the variation.

Are there any other possibilities then the above two?

Thanks in advance

japayson said:

Don't you put a "pass/fail" result on the test?

Click to expand...

Yes, i can do that as well but since i collect variable data so i would prefer to make a variable chart .

Bev D said:

Two dimensional characteristics can be challenging. Is there a specific point that is very important? Either at a specific point in the profile or perhaps a peak value regardless of where in the profile it occurs? This is the approach I have taken (very successful) when confronted with such a. Measurement.

Click to expand...

Yes, it is the peak value in each result which is of interest so smaller the values the better it is (low noise)

Since you test every unit an I, MR chart of the peak in a specification zone (do not combine zones) is teh best approach. You will ‘see’ all of the variation and you can add the specification limit as a reference. (keep the UCL/LCLs just add teh spec limit for 3 limits)

I would strongly suggest you use "try and error" as your test method. Thus, I would set-up different SPC charts:
e.g. for the low freq. range
1. the average value per frequency range,
2. the max. value per frequency range,
for the mid freq. range
1. the curvature of a polynomial fit,
2. the linear slope of such a fit,
and for the for the high freq range
1. the max frequency,
2. the slope of the poly fit.
I would use many more options, but these are just the obvious ones.

After setting up these plots, I would use historic data to select the best plots. Remember: The goal should be to detect systematic trends, which you are able to "counter-act". It the counter-action and the resulting stabilisation of the production output, which brings value to the analysis. Thus, the primary goal should be to correlate the charts with critical input parameters of your process.