Expertise‎ > ‎

Data Analysis

Data analysis seems easy. Put the data into an application such as Weibull ++ and see what comes
out. That's all. Or is it? Consider the plots shown in the figure to the right.There are three families of hard disk drives (HDDs). HDD #1 appears to fit a Weibull distribution well because the line is rather straight. But what if your data produces a plot like HDD #2 or HDD #3? What do you do then? The line is clearly not straight so the failures do not follow a Weibull distribution. What are the possible reasons that HDD #2 has such an upward swing and what analyses do you perform next? (For more details, see "Enhanced Reliability Modeling of RAID Storage Systems".)

Another case is shown in the figure below. Plotting the entire population produces a slightly curved plot. However, when the data are divided into "vintages", the results show distinct differences from one vintage to the next. In one case, I w
as able to determine the month in which the manufacturer made a manufacturing change that increased yield, but negatively affected the HDD reliability very significantly. When you know that certain vintages are more reliable than others, you can make better decisions regarding the disposition of field returns and future purchases.

Be careful when analyzing data and call upon someone who has analyzed large numbers of data sets if you want to get the best understanding of your field data so you can make good decisions. (For more details see "PhD Dissertation", p 52)