"Raw" data examination:
With experience, it can be possible to identify different corrosion mechanisms
and changes in corrosion behavior from "raw" noise data. Filtering out the
so-called dc drift (or very low frequency component of noise spectra) can be helpful.
Filtering can affect the "look" of noise data significantly - for example
pitting transients can have a significantly different appearance depending on the
application of filtering techniques. This approach can be criticized in terms of
subjectivity, its descriptive nature and lack of automation.Statistical
analysis:
Statistical techniques can be applied to quantify noise data, in terms of statistical
parameters. Opposing viewpoints still appear to exist as to whether certain noise signals
are stochastic, deterministic or a combination of the two. In the general field of signal
analysis, time dependent duality of stochastic and deterministic components have been
noted in certain time series.
Stochastic features imply a random, unpredictable
nature, whereas deterministic processes are predictable.
Statistical parameters employed in noise analysis include mean, standard deviation,
root mean square, skewness, kurtosis and localization index. It is obviously important to
consider an appropriate distribution in statistical analysis. Statistical analysis may be
tedious and require extensive computing, as the noise data is typically analyzed in
batches.
Frequency domain transforms:
These techniques transform noise data from a time series (time domain) into the
frequency domain. Fast Fourier Transforms (FFT, well known) and the Maximum Entropy Method
(MEM, lesser known) have been applied for these purposes. It has been proposed that the
roll-off slope of the resulting frequency spectra can give an indication of the corrosion
mechanism (localized vs. general corrosion and diffusion vs. activation control). Batch
processes are again involved in this approach.
Chaos theory:
Chaos has been described as essentially unpredictable (chaotic) behavior, arising in a
dynamic deterministic system. The unpredictability arises from the high sensitivity of the
system's behavior to initial conditions. As a deterministic process, the behavior is
theoretically perfectly predictable, provided complete knowledge was available
about the initial conditions. In practice, complex, unstable initial conditions that can
not be fully characterized and lead to subsequent "unpredictable" data are an
important feature of chaotic systems.
The so-called "butterfly effect" has been widely used as an example of
chaotic behavior, whereby a change in air pressure from a butterfly flapping its wings in
Chicago (a small initial effect) could ultimately lead to a tornado in Tokyo.
In summary, chaotic systems are highly sensitive to small fluctuations, leading to apparent
random, irregular data.
More comments on chaos ...
Wavelets:
Wavelet analysis has been applied to signals that contain non stationary statistical
features over time. Electrochemical noise signals are clearly
non stationary, with the signal frequency and amplitude changing over time.
Mathematically, wavelets essentially break up complex (spiky) data into different
frequency components, separating lower frequency fluctuations from higher frequency
events, as a function of time. The process has been likened to "seeing the trees and
the forest". Essentially the wavelet approach is one of simulating a complex time
series by "wave packets".
Neural Networks:
Neural networks are particularly suited to analyzing complex data, under the influence
of a large number of variables. The approach is one of breaking down a big complex problem
into smaller, simple networked computing tasks. The network "trains" itself to
optimize the solution to a complex problem. |
