Hello All,

I want to find the most important factors that impact the mechanical property of a moulded part.
the data were taken by an intern (back to school now) and he obviously did not follow any DOE technique.
I have now:
6 columns of data (injection speed, back pressure, etc...) and he has modified one factor at a time. The changes in the factors cover the possible range of the machine (so more then low, high variations).

Ex:
temperature Inj Pres Inj speed Rot speed .. Mech stdev-mech
120.0 35 8 30 ..... ...
120.0 40 8 30
108.9 40 8 30
114.4 40 8 30
120.0 40 8 30
125.6 40 8 30
120.0 40 8 30 etc . ...

...... etc.. 29 rows

Can I use the data?
Any suggestions on how?

Thank you

This data does not fit DOE or Factor Analysis.

For DOE, you need to design the DOE in advance, collect the data per the design and analyze. OFAT data does not meet this criteria.

Factor analysis is used to find correlations between a large number of variables to reduce the possibilities to a few factors.

Your best chance would be to use regression analysis to analyze this data. It is very flexible and works on historical data better than DOE.

Thank you.
I did try to use regression but in this case it was not very efficient and the error level was huge. Finally I used the Response Surface Design under the DOE menu in Minitab. It performed above my expectations. The only thing is that I could not use the available standard deviation for the various response. Maybe I should type in all the different data? (5 to 10 samples were tested per test combination).
The response surface design found the optimum parameters and it could also show me the ones that were not so important. I still have to figure out the best options when I use this tool.... more learning! ;-] Some things never change!
So I would appreciate hints on using the Surface design tool of Minitab.

Thank you

Just remember that the data were collected changing one factor at a time (OFAT). This means that you do not have any information regarding potential interactions whether analyzed using RSM or not.

Note this may be the reason why your regression analysis results were poor. When analyzing a regression model, you must manually code in all interactions of interest.

What types of graphical analyses have you performed such as matrix plots?

if only we could have the data to play with too....
when the intern ran his OFATs did he also record the values of the factors he wasn't deliberately adjusting?

Hello!
Well thank you for the interest. If I can attache the data I will!! How? The Excel file or the Minitab one?

when the intern ran his OFATs did he also record the values of the factors he wasn't deliberately adjusting?
Well that is a good point ... I have all the data of every factor he was playing with. The other factor were left at their "standard" values for this type of product.

Note this may be the reason why your regression analysis results were poor. When analyzing a regression model, you must manually code in all interactions of interest.

What types of graphical analyses have you performed such as matrix plots?

I guess I can't code in the interactions with the present data! More experiment then... Unless I assume no interactions? Or very little! (technical literature might help... more time...)
For the graphical analysis the Matrix Plot was Ok but the contour Surface plot of the Response Surface Analysis were very nice! but since I do not have the potential interactions of the different factors I might be looking at just a nice collection of pretty pictures.. but the data makes sense..

Thanks all...

In case I can't find a way to upload the data here is a paste (10 columns by 30 rows): (a bit of work to untangle it! Also the Density data are not all complete... don't know why!)

Lot temperature Inj Pres Inj speed Rot speed Back press dosage Izod Un Std dev Izod Density
211 120.0 25 8 30 3 19 413.94 8.90 1.4541
211 120.0 30 8 30 3 19 361.26 4.20 1.4521
211 120.0 35 8 30 3 19 388.20 6.10 1.4515
211 120.0 40 8 30 3 19 386.86 4.20 1.4510
211 108.9 40 8 30 3 19 300.19 7.70 *
211 114.4 40 8 30 3 19 322.48 2.40 *
211 120.0 40 8 30 3 19 386.86 4.20 *
211 125.6 40 8 30 3 19 178.98 9.10 *
211 120.0 40 8 30 3 19 386.86 4.20 *
211 120.0 40 8 35 3 19 301.14 7.20 *
211 120.0 40 8 40 3 19 244.63 3.50 *
211 120.0 40 8 30 3 17 245.06 4.10 1.4362
211 120.0 40 8 30 3 18 226.37 1.20 1.4368
211 120.0 40 8 30 3 19 386.86 4.20 1.4510
211 120.0 40 8 30 3 20 391.77 8.20 1.4494
211 120.0 40 8 30 3 21 410.35 4.72 1.4517
211 120.0 40 8 30 3 19 386.86 4.20 *
211 120.0 40 8 30 10 19 378.35 9.70 *
211 120.0 40 8 30 15 19 354.67 8.00 *
211 120.0 40 8 30 20 19 376.58 8.50 *
211 120.0 40 5 30 3 19 342.83 3.40 1.4317
211 120.0 40 6 30 3 19 376.82 6.60 1.4462
211 120.0 40 7 30 3 19 360.25 8.20 1.4465
211 120.0 40 8 30 3 19 386.86 4.20 1.4510
211 120.0 40 9 30 3 19 390.29 4.80 1.4476
211 120.0 40 10 30 3 19 401.84 3.90 1.4521
211 120.0 40 20 30 3 19 462.29 5.90 1.4543
211 120.0 40 30 30 3 19 384.08 4.50 1.4458
211 120.0 40 40 30 3 19 252.42 12.80 1.4424

I ran your data through Minitab. There did not appear to be any striking correlations between the factors and the responses.

I took a look at the RSM analysis. Do not trust any graphs that you may have developed. While the RSM seemed to identify significant factors, the R-squared (predicted) was 0%. This means that the model is worthless for making predictions, which is what I feared.

RSM is a specialized form of DOE. Neither tool is very good at analyzing historical data. Regression is the correct tool for that. However, the matrix plot does not show much worth pursuing.

Try designing and running an actual DOE. OFATs rarely work.

Hello,
I tried to salvage some data and I ran The Optimiser under the Response Surface Analysis section of Minitab.
I was looking for the best parameters (column 2 to 7) that could give me the maximum Izod value (8th column) and the analysis was interesting but I know I have no interaction between the parameters so it is probably untrustworthy. At least I can run a DOE with a bit of knowledge of what ''might'' be a significant parameter in the optimisation of the Izod value (a mechanical property of the plastic).

Again Thank you.

When I run the DOE and get real numbers (optimum parameters for a maximum Izod value) I will come back and compare.
Now for the curious here are the present ''optimum parameters'' that I will compare against:
Temperature : 116.3
Inj Pres : 25.0
Inj speed : 19.8
Rot speed : 30.0
Back press : 3.0
dosage : 20.7
Izod Un : should give me an amazing 551 .. wow! Well a bit suspicious
Std dev Izod : Hopefully as small as possible
Density : Maximum value possible .. we will see!

Have a great day.