How to start Process Capabilities Studies in Pig Iron Industry

A

Aashish

Please guide me on starting the process capabilities studies in Pig Iron Industry having 2 small blast furnaces.
 
N

NumberCruncher

Hi Aashish

There are many discussions about this on the Cove, just use the search button. Here is just one example. CPK.pdf - By Don Winton

In general, the starting point is to be absolutely clear. What are you measuring? How will you measure it? How accurate and precise are your measurements?

Since you haven't said what you want to measure, I will invent an example.

Suppose you are interested in the 'chemical purity' of the iron. Not good enough.

Exactly what do you want to measure? Carbon content?, Calcium silicate impurities?, Sulphur? How will you measure these? ICP-MS, ICP-OES, AA...? What is the precision and accuracy of the chosen method?

For this, you can buy a certified reference material (CRM) from N.I.S.T. in America. Each time you analyse a sample of your product, you measure a sample of the CRM. This will tell you how accurate and precise your analytical measurements are. This is important as you need to know how much of your variation is due to your process, and how much is due to your measurement method.

Repeat the sampling and measurement for at least 50 batches of metal, and plot the results. The details are given in the pdf file. You can also do a web search and find lots of web pages discussing process capability, and how to quantify your results.

The above is a bit general, but you need to state what you are trying to measure.

NC
 

Steve Prevette

Deming Disciple
Leader
Super Moderator
My father is in the foundry industry, so I have a "soft spot" for iron. Please feel free to make use of the SPC materials http :// www. efcog. org/wg/esh_es/Statistical_Process_Control/index.htm - DEAD 404 LINK UNLINKED

and more than willing to discuss what may be good variables for pig iron monitoring.
 
Last edited by a moderator:

bobdoering

Stop X-bar/R Madness!!
Trusted Information Resource
I always start with developing the total variance equation, or one can use the CNX approach, to determine what variables you have, which ones are significant, which are controllable, and which cannot be controlled (noise). From that, minimize as many as possible to statistical insignificance, then measure the variation of the others. The key point of the total variance equation is that the results are not one distribution - the distribution of the process - but rather the process and any compounded distributions of residual variation from measurement error, gage error, and noise. That makes the result more multi-modal than normal – which is important when rubber stamping capability indices. Just measuring and assuming the result is the process is the greatest mistake in process capability analysis.
 
Last edited by a moderator:
N

NumberCruncher

Hi Bob

I'm scratching my head. I have always viewed process capability as 'this is what we get'. The simplest example would be goods out the door. You measure a number of parts over a period of time, then calculate Cp, Cpk.

Yes, this does include all of the process variation such as batches, machines, gauges, men (or indeed women, or any sentient carbon based life form!). But isn't this the point? If your product is not good enough, then you improve the process by finding out what is causing the variation (by developing the total variance equation in fact!) then using all of the usual SPC techniques to fix the problems.

"...determine what variables you have, which ones are significant, which are controllable, and which cannot be controlled (noise). From that, minimize as many as possible to statistical insignificance, then measure the variation of the others."

I suspect that I am misunderstanding you, but you seem to be implying in your response that you should fix the process before deciding if it is capable, which is begging the question.


NC

:confused:
 
A

Al Dyer

Aashish, welcome!

What process do you want to control/prove capability in the Pig Iron business?

Input?
Output?
Content?
Materials?
Blows?
Leaks?

ASD... Again, Welcome
 

bobdoering

Stop X-bar/R Madness!!
Trusted Information Resource
I'm scratching my head. I have always viewed process capability as 'this is what we get'. The simplest example would be goods out the door. You measure a number of parts over a period of time, then calculate Cp, Cpk.

Yes, that is the simplistic approach and prone to error – with no clear evidence that the error exists. So, it will readily fool you if all you do is rubber stamp the calculations. You have to actually think about what you are calculating. Your measurement is the result of many things – all of those factors in the total variance equation. The perception is that the "total variance" is only the “process“ variation – a singular variation at that. In order for your measurement to represent only the process variation, you must have ensured all other variances are statistically insignificant. If you have, then good – proceed. If you haven’t, then you need to remove those variances that are not your process variance and are actually masking that variance. If you can’t, then you need to recognize you have multimodal distribution, so do not expect it to be ”normal”. If you think a process has to be normal to be stable, that is not necessarily correct. But, I you are going to properly analyze the capability, you have to apply the best statistical distribution model (curve fitting comparison) for it to be even close to useful. If it is non-normal, the marginally useful indices – Cpk, Ppk – become useless.

Yes, this does include all of the process variation such as batches, machines, gauges, men (or indeed women, or any sentient carbon based life form!). But isn't this the point? .
If you have gage or measurement error, that is not process error. It masks process error, and it makes your data invalid. So, that is not process error, and is not the point. It must be made statistically insignificant before any understanding of the variation is understood. If you have process variation (tool wear, for example), and add batch variation, multiple machines, etc., then it is mutltimodal, non-normal and Cpk and Ppk will be of no value. It would still be recommended to minimize the variation of those additional variances to be statistically insignificant, if it is economically feasible, to increase the process output stability across the streams..

If your product is not good enough, then you improve the process by finding out what is causing the variation (by developing the total variance equation in fact!) then using all of the usual SPC techniques to fix the problems.

If your point is that if your specification is so large that your total variation (not your process variation) is insignificant during the life of your process that you don’t have to care about any of the variations – I would still want to see that the measurement technique was correct and statistical gage resolution (e.g. ndc) was adequate before saying throw that data into any calculation, it is so capable nobody will care.


I suspect that I am misunderstanding you, but you seem to be implying in your response that you should fix the process before deciding if it is capable, which is begging the question .

What I am saying is understand the process - and what portion of the total variation the process may be - before thinking you can actually determine if it is capable. Don't just stuff meaningless data points into an equation thinking it knows more than you. It is not even as good as an 8-ball. Also, ponder if your study really has represented all of the variation that the process will see across its life. If not, then the calculation is truly whimsical.
 
N

NumberCruncher

Hi Bob

I think part of the misunderstanding is how we understand the word 'process'

If you have gage or measurement error, that is not process error.

I am viewing the factory as a black box process. Raw materials in one end, products out of the other. In this case, gauge error is part of the process error, along with batches, machines, production staff...

In this case, the final out-of-the-door measurements do represent the process variation, and the ultimate arbiter of quality is the customer specification.

I am making a couple of previously unstated assumptions.

1) The final out-of-the-door measurement is accurate. Clearly, if the measurement system is not up to scratch, that assumption is not valid.
2) Any non-normality, autocorrelation, time dependency of the data will be picked up by the person analysing the data (my on line pseudonym gives a clue here!).

If you view the factory as a white box process, then gauge error, batch variations etc do become noise factors to be removed, with the process error being 'what the equipment does' so to speak.

I hope that if we worked in the same factory, we would pass each other on the way, you from the inside working out, me from the outside working in. (I would say hello, by the way!).

Anyway, I have dragged this post way off the original topic (sorry Aashish). It's also late here in the UK, and I really shouldn't be trying to think at this time of night, so time for bed.

NC
 

Steve Prevette

Deming Disciple
Leader
Super Moderator
If we go to the subject of Iron - there are several critical factors to Iron production:

What was the temperature of the melt?
What is the carbon content?
What is the content of any alloying metals?
What is the Charpy test results?
What is the strength of the metal?

These are primarily output metrics, but important to the customer. I'd start at least trying to determine if the customer specifications are being met - or if there are any areas with relatively high variation against the specifications (aka process capability).

Then if there are indeed weaknesses, start looking at the processes as to what could be affecting the quality of the outgoing product.

Checking of the raw materials (iron, coke) can also prevent future problems from occurring. By the way, at a foundry my father worked at they stored their scrap metal outside. They dumped a load of scrap into the electric furnace, and unbeknownst to them one of the scrap castings had about a cup of water in it. Instant flash to steam, with considerable spewing of molten metal all over . . .
 
A

Aashish

Sir,

Very true indeed... There are critical parameters like Coke Rate, Hot Blat temperatures etc.... Should i start monitoring the process Capability in those process through Minitab.:thanx:
 
Top Bottom