Properly understanding SPC - Newbie SPC questions

J

Jat999

Hi all

First post and relatively newbie to properly understanding SPC. I hope I am using the correct forum.

I am trying to establish whether a new filling machine is capable to deliver consistent weights across a varying range of different weights and filling materials. The company I work for makes food ingredients and under uk law, has to conform to average weight legislation. Obviously as a business, we need to operate as close as we can to target weight and not be giving away too much to our customers.
The machine in question is a twin head filler. There are inbuilt weigh heads measuring eac deposit. I am able to capture eac weight from each weigh-head.
The material types range from very small sugar balls to small fudge cubes, the jar sizes range from 60grams to 500 grams.

In terms of data captured already, (I have 100’s of sequential data elements), for each weigh head, I am calculating the mean, the SD, the UCL, LCL and I have calculated the CP and CPk.

Where I am struggling is around understanding the outputs.
One head seems to be much more in control than the other. The control limits are closer and operating closer to the target, yet for both heads, the CP and CPk are the same, both values for both heads =1.

I believe my questions are
1. Am I approaching this correctly?
2. Is it possible to have CP and CPk for both to = 1 yet have very different control limit spread?
3. Are there any resources that may help me understand better the application I am trying?

Apologies for the long rambling questions, I do believe SPC is the correct methodology, I need to establish whether there really is capability and importantly, repeatability, in the machine over a wide range of materials before we let the machine supplier leave.

I am happy to share the spreadsheet I am using to show sample output.

John

Edit - Spreadsheet added
 
Last edited by a moderator:

Scanton

Wearer of many hats
I would suggest you use capability analysis to assess your processes ability to meet your requirements, and from the data you have supplied it doesn’t.

I have used your data to perform a capability analysis using SPC for Excel (a plug in software for Excel) using your max as the upper specification limit (USL) and your min and the lower specification limit (LSL) (see updated files attached) and can see that your data shows that your process is not capable with a Cpk of 0.58 for Filler 1 and a Cpk of 0.41 for Filler 2.

In real life conditions a process needs to have a Cpk above 1.0 in order to be able to meet specification requirements whilst allowing for natural process variation. In the automotive industry the minimum level of acceptability used to be 1.33, it then moved to 1.66 and are now asking for 2.0, I would suggest aiming for at least 1.33 to start with, and to achieve this adjustments/changes will have to be made to this process.

I have included an old excel based capability template for you to use in case you don’t have any capability software available.
I’m sure others will chip in with some helpful advice, and we would all be interested in how you get on.
 

Attachments

  • analysis_22-05_09-20 - 1.xlsx
    81.5 KB · Views: 311
  • Capability Study Template.xls
    243 KB · Views: 503

Miner

Forum Moderator
Leader
Admin
Before worrying about capability, it's always a good idea to determine whether your process is stable, and whether it is normally distributed.

The filler processes are showing some indications of instability. Also, Filler 2 does not appear to be normally distributed. This might be the result of the instability, or due to the process itself. Either way, the capability metrics Cp/Cpk are based on the assumption that the process is both stable and normally distributed.

Note: you did not define the upper/lower specification limits.
 

Attachments

  • Filler.jpg
    Filler.jpg
    36.1 KB · Views: 319
J

Jat999

Scanton & Miner
Many thanks your feedback. I will look through your files to reapply my data.

I also realised (as I explained, I am a newbie to this), I mistook the UCL and LCL interpretation for calculating CP and CPk for Upper and lower specification levels. I now have added these and adjusted the formulas accordingly.

Also thanks for advice on CP and CPk values.

Just one more question. Any good bedtime reading for beginners on the web which overs the area I am working on please?

John
 

bobdoering

Stop X-bar/R Madness!!
Trusted Information Resource
I believe my questions are
1. Am I approaching this correctly?

I recommend reviewing The CORRECT steps to implement an SPC chart


2. Is it possible to have CP and CPk for both to = 1 yet have very different control limit spread?

Absolutely, look at the attached Cpk=1 distributions. One would have very wide control limits, and the other would have very narrow control limits. In Shewhart charts (which are likely applicable in your application since fills are usually a random, independent condition, therefore meeting the statistical requirements of a Shewhart chart) capability has nothing to do with control. In fact, the specifications are not even a part of control limit calculations. All Shewhart controls charts are designed to do is tell you if a condition exists that meets the criteria of a special cause to act on. It never tells you if you are making good parts. It is process control, not part control.
 

Attachments

  • cpk.jpg
    cpk.jpg
    20.3 KB · Views: 327

bobdoering

Stop X-bar/R Madness!!
Trusted Information Resource
Before worrying about capability, it's always a good idea to determine whether your process is stable, and whether it is normally distributed.

The filler processes are showing some indications of instability. Also, Filler 2 does not appear to be normally distributed. This might be the result of the instability, or due to the process itself. Either way, the capability metrics Cp/Cpk are based on the assumption that the process is both stable and normally distributed.

Note: you did not define the upper/lower specification limits.

You need to determine if the fill process should be normal - even if it is stable. A low viscosity fluid should be close to normal. However, a thicker material may do some weird things beyond the minimum fill, such as string or hang up in the fill head. You can also have dramatic variation with some materials based on material temperature. Does it change over time? Very little actually behaves like a perfect bell curve. Outputs generally have many acting variances, not just one. (See my recommended reading above)
 
J

Jat999

Hi Bob

Thanks for coming back to me.:agree1:

In terms of "The correct steps" we have already found an external(?) variance - power fluctuations - causing the vibratory feeds to fluctuate. We are currently making a software feedback loop to compensate.

The materials we are dispensing are solids, ranging from small sugar beads through to small caramel chunks.

I am wanting to initially establish, by each material type if we have some level of control. Once we are OK here, then I need to establish repeatability of the process. Also as the machine has a number of parameters that can be changed, I need to understand which impact to what degree.

Biggest problem at the moment is stopping engineers who keep "tweaking" the parameters without understanding cause and effect :mad:

John
 

Miner

Forum Moderator
Leader
Admin
Also as the machine has a number of parameters that can be changed, I need to understand which impact to what degree.

Biggest problem at the moment is stopping engineers who keep "tweaking" the parameters without understanding cause and effect

There are three approaches to this.
  • The quick approach is to perform a designed experiment to identify the most important parameters.
  • A slower approach, which requires that you allow the "tweaks" is to collect data on all the parameters as well as the product. The reason you must allow the tweaks is to see enough variation to actually make a difference. You can then analyze the data using regression analysis.
  • Another approach recommended by Donald Wheeler may be found here.


If it were me, I would opt for the designed experiment approach. I have used this approach for more than 30 years with great success. It is more definitive, and will provide a better understanding of the process and it's interactions. However, you do have to know what you are doing. I suspect Wheeler's disagreement with this approach is due to people that did not know what they were doing.
 

outdoorsNW

Quite Involved in Discussions
From the OPs comments, the product appears to be along the lines of cake decorating material, or chocolate chips. Part of the difficulty here is there are two systems affecting the accuracy- the filling system and the product making system.

Some products appear to be discrete chunks, not true bulk materials such as liquids or powders. With liquids or powders, you can add or subtract a drop. With discrete pieces, you must add another whole piece. One time 50 pieces may make the correct weight, another time 50 pieces might be 0.1 g short, so the machine adds another piece, bring the total weight to the high end of the range.

So the variability in the piece weight matters a lot. It might be reducing piece to piece variability helps reduce packing machine variability.

Also, you said you are working with different types of products, and packaging machines likely work better with some product types than others-more variability.

You may want to try carefully selecting samples with reduced weight or size variation as a test to see if piece variation is a factor.

Low weight pieces are likely to give better results because the amount of weight added by the last piece is less, reducing absolute variability of the final result. What I mean is, in a perfectly performing system, with an average piece weight of 1 gram, and a max weight of 1.1g, I would expect the final package weight to be no more than 1.1g over the minimum (ignoring any safety factors). With an average piece weight of 10 grams and a max weight of 10.5g, I would expect the final package weight to be no more than 10.5g over the minimum.

Does the overweight condition matter for regulatory purposes?

Typically the concern is to make sure customers are not sold less than the package says. If you can simplify and mostly ignore the too heavy condition, you may have an easier time getting you packaging system to meet regulatory requirements. Later you can save money by reducing the max weight the system allows.
 
Last edited by a moderator:
Top Bottom