Statistics - Where to start in ISO 9001?

[qualprod]

Hello everybody
I come looking for guide regarding statistical analysis.
Into an ISO 9001 system, I´ve been gathering a lot of data of production, in order to be analyzed to implement improvements.
I have a spreadsheet, with headings like: work orders, date, quantity produced, defective product, operator, machine, time spent in production, days spent in each work order, used raw material (description and quantity) and so on.
My background is not specialized in statistics, I know very little, but I want to start to analyze the data.
I can start to use pareto, histograms, average values, which I already know , but no more.
Could you guide me, where to start , what to do, what to read in order to make very good analysis of such information and
that finally can get sufficient info to be translated to improvement projects?
I know that into statistics , there are a lot of unknown terms for me, but probably are not applied for my purpose.
e.g. Binomial distribution, Box plot, Central limit theorem, Continuous distribution, Geometric distribution, Interquartile range, Pascal distribution
I don´t want to make sophisticated analysis, only those which could help me to improve the production processes.
The idea it is to use the less methodologies but that really can be applied to my business.
Thanks for your inputs.

 

[Bill Levinson]

ISO 9001 doesn't really have statistics requirements although IATF 16949 requires activities like measurement systems analysis (ISO 9001 does not). The thing to do in ISO 9001 is to use whatever methods are appropriate to fulfillment of your quality management system's mission. "Defective product" could, for example, be subject to control charts for defects or nonconformances.

On the other hand, some of the tools you mention such as Pareto charts are applicable to root cause analysis, and therefore corrective and preventive action (CAPA) which is required by the standard. The Pareto chart is applicable, though, only if you are trying to prioritize among several different problems. If you have different defect sources, the Pareto chart would be applicable.

 

[Big Jim]

All of that can be very insightful when needed but a huge waste of time when not needed.

I believe in the philosophy of management by exception. Spend you time on the abnormalities that pop up, not on the good stuff. Even with that philosophy, it is helpful to have a basic set and the basic set I use are 1) a measure of customer satisfaction, 2) a measure of product quality, 3) on-time delivery, and 4) a measure of supplier performance.

When an abnormality pops up, no matter how it is found, dig deeper with whatever method of investigation is appropriate, including any of the ones you mentioned. Otherwise it is paralysis of analysis.

 

[Tagin]

Could you guide me, where to start , what to do, what to read in order to make very good analysis of such information and that finally can get sufficient info to be translated to improvement projects?

Looking at it in this direction (data analysis -> improvement) is kind of a random data mining/sifting activity, where you are hoping the analyzed data will somehow show a pattern that you can somehow detect and then act upon.

It would be much simpler, and probably a lot more effective, to go in the other direction (improvement need -> data analysis). This will more clearly define the kind of analysis and statistical approaches needed for a specific goal.

The other problem with the first approach is that it can only show you patterns on data you have collected. There may be needs for improvement that require data you have not collected, or have measured differently than needed, and so those patterns would then be undetectable from your data.

 

[Bev D]

I echo what the others have said. Statistical methods are not to be used randomly. The best approach for you is to determine what is important to your organization FIRST. Meeting build plans? Capacity/productivity? Yields or defect rates? Then you really only need Paretos and run Charts. What are the biggest problems? (This may be quantified in terms of occurrences or cost or cycle time consumption). Trend the data fro each item in the Pareto - is it getting better, worse or is it stable? THEN pick an area to improve. You will need to go collect specific data on that problem - the answers won’t be in the general data you have in your spreadsheet.

 

[normzone]

What [Bev D] said.

A simple track-and-trend the percentage of success (meeting schedule for process dwell time or delivery, or passing quality inspection or test) yields valuable data. If those numbers are acceptable, then you can get fussier and slice your data finer. If they're not, look into why not.