Detecting Trends

Does anyone have any recommendations on detecting trends.

We are going to be recording a series of measurements on a part, and I want to detect if there is a trend in the data over X number of parts. I have never done this before, and wanted to see if anyone had some pointers for statistical analysis.

We made a QC system for Coors that checks how a press stamped out the tops of cans ( ends ). We used a running average and running standard deviation to detect bad ends either from dirt on the aluminum or dirt in the die. We could see the bearings in the press start to fail because the standard deviations started to get bigger.
I think we had a queue of 256 parts which was only 25.6 seconds longs since the press stroke rate was 10 strokes per second. The queue may have been longer. I did this over 20 years ago.

Doing a simple limit check works for many applications. Looking at the rate of change can often detect errors before the limits are exceeded. Using standard deviations is a good way to dynamically set limits. Monitoring the changes in the standard deviation can tell you when your quality control is starting to drop.

Is this for SPC?

Trend rules would be five consecutive measurements on the same side of the Xbar value or seven consecutive measurements continually increasing or decreasing. Some people add 14 points in a row each consecutive measurement alternating above and below the Xbar value and 8 points in a row more than 1 standard deviation from the Xbar value in either or both directions.

Check with your customer because there are different trend rule sets in use for various SPC philosophies. Find out what trend rules they are using.