Is it possible to determine a plausible data point average when using numerical and attribute data? Example: substrates of varying thickness are tested under a constraint for 10hrs and if the substrates withstand the constraint for the allotted time period, this is a pass. If it fails within the time period, this is a fail. Testing is run overnight so there is no evaluation at interval time periods to determine a difference in failure (eg..sub1 fails after 2hrs vs sub2 failing after 5hrs) a fail is an absolute fail. Please let me know if there's a way to analyze this type of data, statistically, in minitab and can the software suggest/predict a possible thickness that will still pass at a lower level (lower than 1 but greater than 0.1)?
Question: How to analyze numerical and attribute data
- Thread starter Dannikf
- Start date
If you were to have data in the middle region, the curve would be more accurate and you could do as you are asking. However, this particular curve is Minitab's best approximation since the data are missing. Therefore, there is much greater uncertainty.
Unless there is a significant cost to experimenting to fill in the missing data, I recommend doing that. Otherwise, you can start moving closer until results start falling.
Unless there is a significant cost to experimenting to fill in the missing data, I recommend doing that. Otherwise, you can start moving closer until results start falling.
A thought that may be a "Yeah, we already knew that"...
If the test period is 10hrs, and you do it overnight (18hrs, assuming single shift)...why not start the test when you arrive in the morning and check in on the parts every 30-60 min?
This could help get you data in the middle region.
Regardless, you also have a HUGE step in substrate thickness, with 100% of thins failing and 100% of thicks passing... do you ever have substrates in the middle thicknesses? If not, plotting a curve seems a bit overkill.
If the test period is 10hrs, and you do it overnight (18hrs, assuming single shift)...why not start the test when you arrive in the morning and check in on the parts every 30-60 min?
This could help get you data in the middle region.
Regardless, you also have a HUGE step in substrate thickness, with 100% of thins failing and 100% of thicks passing... do you ever have substrates in the middle thicknesses? If not, plotting a curve seems a bit overkill.
John Predmore
Trusted Information Resource
Could you hire an outside lab to run a one time test, and observe periodically (maybe every 20 minutes for 10 hours) for failure? In this way you would obtain variable data on failure points during the 10 hours run. My other idea is if you collect time-to-failure data for all samples, without stopping the test at 10 hours, you would have a better picture of the failure curve distribution, which enables better predictions. A Certified Reliability Engineer might offer other ideas, because they make a science out of failure rate analysis.
I remember one time, my company ran a product durability test that required manual application of a corrosive test fluid, every hour for 48 hours. All the team members voluntarily took a 1 hour shift throughout the night. I did my part, set my alarm for 2:30 am, did a squirt at 3:00 am and then another squirt at 4:00 am before I headed home and back to bed. I know this is not the question @Dannikf asked, but sometimes the barriers we think are insurmountable are really not that hard to overcome, depending on the culture of the workplace.
I remember one time, my company ran a product durability test that required manual application of a corrosive test fluid, every hour for 48 hours. All the team members voluntarily took a 1 hour shift throughout the night. I did my part, set my alarm for 2:30 am, did a squirt at 3:00 am and then another squirt at 4:00 am before I headed home and back to bed. I know this is not the question @Dannikf asked, but sometimes the barriers we think are insurmountable are really not that hard to overcome, depending on the culture of the workplace.
The data requirement has changed by lowering test time which allowed for a bit more data points. Based on the attached data, how can this be evaluated in minitab? I've been receiving error messages when I do statistical analysis and I'm not familiar with the software enough to know how to achieve what I'm looking for. Just by looking at this data, can you make a suggestion for what analysis can be used to logically evaluate this data?
A better question would be: is it possible to use some sort of predictive analysis to determine what the lowest film thickness can be added that will still achieve a passing after 360s? If this is not possible, what are some things that can be extrapolated from the data besides failing and passing at the extreme ranges of the test?
