# How to Implement SPC

Discussion in 'SPC - Statistical Process Control' started by Cyril, Oct 29, 2015.

1. ### CyrilMember

Joined:
Oct 29, 2015
Messages:
6
0
Trophy Points:
1
Hello,
I was directed to start SPC implementation in our plant. Im kinda lost how to start it. But this is my initial plan:
-determine the appropriate control charts (based on the data to be monitored (attribute or variable)
- collect the data sets. Is 20-25 data sets okay?
- then after that I'll calculate the statistics and control limits then construct a control chart.
-Then evaluation for in control will follow.
- process capability determination
-corrective action
Is this the right path?
The pilot plant that was chosen packs products inside plastic cups. I'm thinking of just monitoring the weights of the output cups. I'll use x bar and R. How many samples should I take per data set to be representative of the population?

2. ### Bev DModeratorStaff Member

Joined:
Jul 30, 2015
Messages:
343
343
Trophy Points:
62
Location:
Maine
First begin reading the works of Donald Wheeler. Start at qualitydigest.com and also google "SPC Toolkit wheeler" for a quick find on some of his older columns. The first thing you want to understand is rational subgrouping.
Select a couple of critical characteristics to monitor. Don't try to do it all at once, you'll drown
think about the process - how should you subgroup the data? how often should you take a subgroup? what can go wrong and how will people react to out of control points
You might want to perform a MSA on the characteristic prior to collecting the data
20-25 data subgroups is OK to start - there is no real set number but this is a good rule of thumb.
plot the data and the control limits. now THINK about what it means...you can start with the standard charts but remember they are not always the correct choice based on the homogeneity of the process...

get some data and then come back and we'll help you through the process....

Cyril, Miner and Candi1024 like this.
3. ### Andy NicholsModeratorStaff Member

Joined:
Jul 30, 2015
Messages:
1,682
859
Trophy Points:
112
Location:
In the "Rust Belt"
When I tried to implement SPC one issue I came up with was trying to get people to agree to a feature which, above all else, needed controlling. I tried to introduce the concept of a "critical feature" which if wasn't correct, to spec, the rest wouldn't matter, kinda thing.

For cups, is the weight critical? If the diameter isn't correct, is that critical? How about wall thickness? Weight might not be all that important if it's got a hole in it, because of a fill problem, for example.

Cyril, Bev D and Miner like this.
4. ### MinerModeratorStaff Member

Joined:
Jul 30, 2015
Messages:
235
165
Trophy Points:
42
Location:
Greater Milwaukee USA
Cyril, Bev D and Andy Nichols like this.
5. ### CyrilMember

Joined:
Oct 29, 2015
Messages:
6
0
Trophy Points:
1
Thanks Bev. Attached is a sample of the monitoring from the department. This is the monitoring of the fill weights from each sampled cup. The process of the pilot plant (in-charge of packing products inside the cups):
1)inspection of the raw materials
2) filling the solid material
3) filling it with juice
3) sealing
4) cooking and cooling
4) putting it inside secondary packaging

For the fill weight data i will be using x bar chart.
Im not familiar with MSA will read on this today. When you say perform an MSA on the characteristic, characteristic of the product or the parameter or machine producing that product.
The following are being monitored:
burst pressure, percent oxygen content,brix, acid and vit c, ph, microload
sealing temp and time and pressure
cooking temp,cooking time,cooling water temp and cup centered temp after cooling
fill weight and net weight
metal detector
The critical control pts: are ph -3.5 target
cooking and cooling temp
burst test
metal detection
For SPC Im thinking of charting the fill weight, or im missing a lot of things? Thanks again

File size:
29 KB
Views:
2
6. ### Andy NicholsModeratorStaff Member

Joined:
Jul 30, 2015
Messages:
1,682
859
Trophy Points:
112
Location:
In the "Rust Belt"
My apologies, Cyril. I read what your product is wrongly in your OP. So fill weight IS the feature you wish to control. In that case, you could do worse than take 50 filled cups and measure them. The MSA is a complex study of those characteristics of the measurement process which causes variation. Study that so you know what variation is detectable. If your fill weight is +/- 5grms but the measurement taken can only discriminate to +/- 7 grms, then clearly you'll be in trouble. Since measurement is also a process, it's like doing the same kind of study and taking the filled product off the end of the line and measuring it...

Cyril likes this.
7. ### CyrilMember

Joined:
Oct 29, 2015
Messages:
6
0
Trophy Points:
1
Its okay Andy. I think they want SPC to control fill weight. BUt then Im also open if the approach is right or based on the process above. Are there other things that should also be considered or missed. Thanks

8. ### CyrilMember

Joined:
Oct 29, 2015
Messages:
6
0
Trophy Points:
1
Bev,
I read Dr.Wheeler's SPC toolkit and from that I graphed the data I had. I attached the file, please check if Im doing it right? thanks.

File size:
60.5 KB
Views:
5
9. ### Bob DoeringMember

Joined:
Jul 30, 2015
Messages:
36
24
Trophy Points:
7
The CORRECT steps to implement an SPC chart
You might have had some SPC training in the past, but it probably skipped over the correct steps to implement charting on the shop floor. Check it out, here they are:

1. Develop the total variance equation from the fishbone diagram of process variables. (Yes! This is the very first step!)

2. Determine which variance factors are adjustable, which are noise, and which can be set as a constant (CNX analysis.)
3. Choose the most important adjustable variable to chart (noise and constant variables can only generate "report card charts" which are of minimal value)
4. Ensure you have correct gaging and measurement techniques. Gage should have an ndc>10. For dimensionals, measure the same location on every part. Gage and measurement error tend to mask the true process variation with false normal distributions.

5. Minimize the variation of each of the participating variables - get the process in a steady state and capable. This includes eliminating your "special causes".
6. Prepare data in a time-ordered sequence as a capability study.

7. Accurately determine the correct distribution of the output. Distributions fitting software (e.g. Distribution Analyzer from variation.com are handy). NEVER ASSUME NORMALITY. NEVER ASSUME THE OUTPUT MUST BE NORMAL. Make sure they make sense! Determine if it is random or non-random (e.g. time function) data. Shewhart charts do not work with non-random variation.

8. Pick the correct chart to evaluate that variance factor (variable, attribute, Shewhart, non-Shewhart, etc.)
Please, please, please do NOT rubber stamp X-bar-R charts on the shop floor. Do your homework to get best results and not frustrate operators with nonsense.

Last edited by a moderator: Nov 4, 2015
Andy Nichols and Cyril like this.
10. ### CyrilMember

Joined:
Oct 29, 2015
Messages:
6
0
Trophy Points:
1
Thanks Bob. I'll look into CNX as well. I've never heard of it before though I know fishbone diagram.

11. ### MinerModeratorStaff Member

Joined:
Jul 30, 2015
Messages:
235
165
Trophy Points:
42
Location:
Greater Milwaukee USA

Cyril, CNX is just the classification of the each input variable (from fishbone diagram or parameter diagram) into those that are controllable to adjust the output (C), those that are not controllable over the long term and will vary as noise (N) and those that are controllable, but will not be used to adjust the process and are instead locked at a particular setting (X).

Cyril likes this.
12. ### Bev DModeratorStaff Member

Joined:
Jul 30, 2015
Messages:
343
343
Trophy Points:
62
Location:
Maine
An alternative to CNX which is characteristic specific you can perform a components of variation study (COV) study which groups factors into categories - or components - such as within piece, piece to piece, time to time, lot to lot vendor lot to lot, operator to operator, fixture to fixture or cavity to cavity. (Research "Multi-Vari" studies). I use this approach to asses non-homogeneity and develop my rational subgrouping, chart selection and sampling frequency strategy.

Cyril likes this.
13. ### Bev DModeratorStaff Member

Joined:
Jul 30, 2015
Messages:
343
343
Trophy Points:
62
Location:
Maine
AS an addition. I don't believe that COV studies replace CNX studies. I have just found that for rather complex processes such as assembly or processing of multiple components - especially when the components are of various types (electrical mechanical, biological chemical, etc.) and you are implementing a SPC monitor on an existing process, a CNX calculation can be overwhelming. In these cases I have found that a COV study gets the team started faster and the CNX can be done after monitoring begins. Certainly a CNX study will create a much more specific reaction plan to out of control conditions and so I think it's a judgment call for existing processes.

I do require them as part of new product or new process development along with a COV study...

Cyril likes this.
14. ### Bob DoeringMember

Joined:
Jul 30, 2015
Messages:
36
24
Trophy Points:
7

Process output is a result of the sum of all variances that affect the process. This example is a very upper-level example. It can be broken down into greater detail. The key is dealing with all sources of variation that are statistically significant special and common causes.

Cyril likes this.
15. ### CyrilMember

Joined:
Oct 29, 2015
Messages:
6
0
Trophy Points:
1
Thanks a lot for the help. I'll collect the data for MSA next week. I have a question do you perform MSA only for a certain type of device like the analog ones only?

Joined:
Jul 30, 2015
Messages:
36