Methodology for an Offline Outlier Detection in a Dynamic Measurement Data Set

The data acquisition process translates physical magnitudes into discrete, sampled data sets. Theoretically, this process executes undisturbed and without delay. But especially if accurate statistic distributions or extremal values are desired, outliers caused by physical errors in the measurement process falsify the result. Conventional approaches have difficulties with detecting outliers in dynamic and distorted data sets. This work shows two methods to classify and reliably improve these aspects. The first approach involves multiplying two factors to yield a quality factor determining outliers. In contrast, the second approach determines outlier status by identifying specific points through the addition of multiple factors, rather than by multiplying two factors. The combination of both methods leads to a more effective detection process, offering a time-efficient and robust technique for pinpointing outlying sampling points in transient measurement data sets.