# Statistics to define the capability of operator controlled manual machines

S

#### Steve Thomas

Where automatic machines can run without operator intervention for more than 1 hour, Statistic Cmk can be used to define their capability.

There are many manually set and run machines in the world which will not run for 1 hour without operator intervention and so Cmk statistic is not really valid. Cpk is also not valid because the machine/process is not in statistic control.

What statistic could be used to define the capability of operator controlled manual machines please?

Thank you....Steve

M

#### M Greenaway

Steve

The method of control has no bearing on the type of analysis undertaken.

S

#### Steve Thomas

Cm and Cmk capability studies normally require 1 hours worth of uninterupted sampling (operator not permitted to intervene in the 1 hour) in order that the Cm or Cmk statistics are valid.

If the machine won't run for 1 hour without operator intervention, how could Cm or Cmk be used? It might be possible to use less than 1 hour for the sampling, but if the time between operator interventions is very frequent such that sampling can only be done for say 15 minutes rather than 1 hour, doesn't this create a misleading Cm or Cmk statistic?

Thanks....Steve Thomas

M

#### M Greenaway

Sorry Steve, in my haste I thought you were talking Cpk. To my knowledge such constraint are not applicable to Cpk.

#### Mike S.

##### Happy to be Alive
Trusted Information Resource
I hope this isn't a double post. I'm having some trouble posting today and don't know why. Oh well...

Steve,

Welcome to the Cove. I agree with Martin - use Cpk. There are many situations where man and machine work closely and Cpk will work fine as a capability statistic here.

S

#### Steve Thomas

Thanks for your responses. Ok, looks like Cpk is the one to use in these circumstances. Need to work to pull the machines into statistical control so the statistics are valid.

One additional question: For Cpk, we would usually 4 hour capability study period. Would recommend less time for manually controlled machines, or the same 4 hours we use for automatic machines/processes.

Thanks....Steve

M

#### M Greenaway

Steve

When you set up such a study you should take samples initially frequently enough to discover if there is any process shift or instability. Then extend the time period as the data shows process stability.

R

#### Ravi Khare

Though Cpk is calculated on the basis of an estimated standard deviation (sigma obtained from R-bar/d2) over the observed range, isn't it based on the underlying assumption of Normality?

In case of operator intervention, which would be center shifting most of the time, would the normal distribution still hold good? It looks like you will get a multipeaked distribution due to shifting centres. In which case the assumption of Mean +/- 3sigma covering 99.72% of the data may not hold good.

S

#### Steve Thomas

With a manually operated machine the operator may have two intentions when re-adjusting a machine.

Let's assume the operator is running frequent checks of size as the components come off the machine (maybe also taking samples for SPC) and realises that size is approaching say top limit of tolerance. The operator may stop the machine to re-adjust or have some on-line method to re-adjust without stopping.

Either way, the operator may have to off-line the machine after re-setting to establish what the actual new size is. Often re-adjustment back to the intended size is more of an art than a science. The operator may not be able to hit the aim (assume centred) first time, and might be content to run the process provided the new size is approximately centred, or even maybe closer to the opposite extreme tolerance (bottom limit in this example) as making yet another adjustment to try and re-set to aim may not be successful and increases likelhood of scrap, and reduces process downtime.

How can any meaningful statistic such as Cm, Cmk, Cp, Cpk, PC, CR be applied to such circumstances?

M

#### M Greenaway

Steve

I dont think any of the stats quoted would be helpful in understanding your black art of machine setting, or your need to reduce downtime. I woulkd suggest that you try to understand the machine variables, and how they affect the product. Reducing the amount of variables, or the variation itself will help reduce your downtime.