Predict and Prevent Process Failure Using an Autocorrelation Plot

Over the years we have seen numerous applications of Statistical Process Control which have resulted in a great deal of benefits to organizations looking to improve their processes. These benefits range from being able to detect potential failures in a current process to being able to predict and prevent such failures in similar processes in the future.

It’s always desirable that the data from a process should show random behavior. If we see that the data coming out of a process is not random, we should immediately investigate before it results in any major failure. Now, the challenge is how we come to know if the data is random or not random. It’s also important to detect the non-randomness at the right time.

Detecting Non-Randomness Using the Autocorrelation Plot

The autocorrelation plot is a widely used technique to detect such non-randomness in the data. It not only gives a signal for the non-randomness but it also shows if there is any typical pattern in the data, like trending upward or downward, a seasonal or cyclic trend, etc. It reveals if any correlation exists between the data points in different time lags and whether they are significant. The correlations can be checked with different significance levels. The autocorrelation is measured with the autocorrelation coefficient, which is always between +1 and -1. If the autocorrelation values are close to +/- 1, it indicates that there is a higher degree of correlation.

The autocorrelation can be plotted in Microsoft Excel using the formulas below. There are many Statistical Process Control software programs which have an inbuilt capability to do this analysis by themselves. Some software is also capable of showing an alert when the non-randomness is statistically significant. This triggers an investigation for the user so that he can take the necessary corrective action if required. Read more