Dear Respected RG Members,
I want to perform the statistical anslysis on the defects/faults data. If the data is related with the continuous process and show the mean of daily fault for 30 days.
Which statistical tests can be performed for the analysis of the specific defect?
Thanks and regards
Hi,
Please could you explain in some details what you need exactly to test? e.g. monitor the faults average.
Dear Dr. Ayman Amin,
Thanks for showing interest and to spare your precious time.
I am interested to to investigate effect of yarn fault/faults on the accepted and rejected yarn production.
Thanks & Regards
Are you also interested in relating the faults to the continuous process parameters? Perhaps statistical analyses to identify which process parameters are correlated with the fault rates; afterwards to establish the causation (if any) in the post-analyses phases. In this case you may want to look at cross- and auto-correlations, then further to represent conditional probabilities as Bayesian Belief Networks.
If there is only one data point per day it is a Bernoulli process and Ette is correct. If it is a continuous data colection, or very frequent relative to the number of events, and the faults are independent of each other, you can treat it as a Poisson process with some hazard rate lambda. (See Wikipedia) Test to see if the events are clustered or independent. If they are clustered, then it may be a Hawkes process meaning lambda jumps when an event occurs, then gradually mean-reverts. (Colloquially, this is referred to as "when it rains it pours") You can also test whether lambda is affected by any changes in personnel or machine settings, or is relatively stable day to day and month to month.
Dear Dr. Ette Etuk,
Thanks for expert inputs. Thanks also to spare your precious time for valuable answer.
Thanks & Regards
Dear Dr. Sanjiv Sharma,
Thanks for very detailed & effective expert inputs. Thanks also to spare your precious time for valuable answer.
Thanks & Regards
Dear Dr. Martin Goldberg,
Thanks for very detailed & effective expert inputs. Thanks also to spare your precious time for valuable answer.
Thanks & Regards
Mr. Sharma raises great points. The statistics by themselves may not solve what I assume is a yield problem. You haven't shared what the pass fail criteria are, but I assume there are more than one cause for failing. I think you need to look at the different reject events and classify them by type, along the lines of Mr. Goldberg's suggestion, and look at related causal factors in the process. if its sequential in nature then it is important to distinguish between the points at which defects are being found vs. where thay are being introduced, as earlier identification may allow correction rather than rejection of Product.
Statistical process is best viewed as a stochastic process.
I recommend Bernoulli process.
- Badges
- Science topic