I'd like to produce a control chart that tells me if a given process will be within bounds in the future or not.

Currently, the process gives me simple timeseries data, per minute

Time  Latency in sec
15:25 1000
15:26 1200
15:27 950

I have about 1000 of these per day. We can take the assumption that the process works within bounds for now.

Based on this, I calculated the mean and standard deviation of the data points. Then I calculated ±1 standard deviation, ±2 standard deviations, ±3 standard deviations and charted the data.

If the data were normally distributed, one can assume that about 68% = 680 data points are within 1 stdev, 95% within 2 stdev, etc. but this was NOT the case. So the data is NOT normally distributed, which I already expected.

Now I read somewhere that the data doesn't need to be normally distributed. But we need to find a model such that the residuals need to be normally distributed.

Is that the way to go here? Find a model that gives me a normal distribution based on the data I have?

What I did in the meantime in my control chart is to create thresholds which I created in such a way that the 1000 datapoints indeed follow the 68-95-99 rule. But for example the first threshold is from ±4500 and the second threshold is from ±7000. Then 68% of the data is within the first control line and 95% of the data is within the second control line. Is this approach not valid?



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.