Dynamic time warping: Compensating for temperature variations

Abstract

Guided wave structural health monitoring has the potential to monitor large structural areas. Yet, temperature variations are known to misalign measurement data and cause false alarms. Current popular temperature compensation methods include the optimal signal stretch method and interpolation method. While these methods perform well for small temperature changes, they fail in large temperature variations. In this paper, dynamic time warping is used to realign signal responses that have been distorted by temperature. Baseline subtraction between the realigned signal and the baseline can then be used to detect and identify damage responses. We demonstrate dynamic time warping with guided wave measurements taken from an aluminum plate that is heated in cycles from roughly 75°F to 125°F. Dynamic time warping achieves a correlation between the chosen baseline and the pre-damage measurements greater than 90%. Applying no temperature compensation or more traditional methods, such as optimal signal stretch, demonstrate correlations of 0% and 30%, respectively. This shows that dynamic time warping removes temperature effects more accurately than other current approaches.