Fractal analysis is applied in a variety of research fields to charact
erize nonstationary data. Here, fractal analysis is used as a tool of
characterization in time series. The fractal dimension is calculated b
y Higuchi's method, and the effect of small data size on accuracy is s
tudied in detail. Three types of fractal-based anomaly indicators are
adopted: (a) the fractal dimension, (b) the error of the fractal dimen
sion, and (c) the chi-square value of the linear fitting of the fracta
l curve in the wave number domain. Fractal features of time series can
be characterized by introducing these three measures. The proposed me
thod is applied to various simulated fractal time series with ramp, ra
ndom, and periodic noise anomalies and also to neutron detector signal
s acquired in a nuclear reactor. Fractal characterization can successf
ully supplement conventional signal analysis methods especially if non
stationary and non-Gaussian features of the signal become important.