Gas turbine engine operational data analysis for anomaly detection: Statistical vs. neural network approach

Multivariate data analysis by artificial neural network (ANN) approach was carried out in a previous work to identify anomalous regime in a gas turbine operational data. In this work, statistical hypothesis analysis using univariate time series exhaust temperature data from the same engine model has been carried out. The objective is to compare the results of these two methods. The applicability and consistency of parametric t-test and nonparametric rank sum tests are first assessed considering different sample sizes with respect to two typical realistic data sets. This builds up confidence in the approach. The anomaly measure tends to increase sharply with sample size. Anomalous regime in the data are identified by both hypothesis tests with high anomaly measures as found by ANN technique.