21 High-Temperature Creep Damage Evolution of C/SiC …
237
21.2 Pattern Recognition Analysis Method
The creep failure process of C/SiC involves several different damage mechanisms,
for example, fiber breakage, matrix cracking and interface debonding. The complex
evolution of these damage modes determines the creep life of C/SiC. According to
their physical background, different damage mechanisms exhibit different acoustic
emission characteristics. Pattern recognition analysis is required to identify the
damage mode of each acoustic emission event, and thus clarify the damage evolution
during the whole creep experimental.
The K-means algorithm, an unsupervised pattern recognition method, was chosen
to determine the clustering of the AE signals. This mainly consists of three parts:
constructing a range of feature parameters, identifying the appropriate combination
of these parameters and applying K-means clustering analysis on this combination.
This protocol allows all relevant parameters to be included, and it avoids the impact
of human factors such as bias and other spurious effects. The optimal combination
of binary variables is determined through a voting method, thus the most reasonable
feature parameters suitable for CMC AE signal pattern recognition are obtained.
Different types of damage show different AE signal features, and different damage
types can be identified through K-means clustering. All C/SiC acoustic emission
pattern recognition analysis methods can be simplified as shown in Fig. 21.1.
21.2.1 AE Parameter Combination and k-means Cluster
In traditional AE pattern recognition, the selection of feature parameters is relatively
fixed, and those parameters are determined by those arising from metals, which are
not suitable for CMC. AE signals contain a lot of information, but much of this
is invalid for pattern recognition, so it is essential to determine feature parameters
suitable for CMC analysis.
Different parameter combinations have different effects on the eventual pattern
recognition, so many different parameter combinations were constructed in order
to create variables suitable for the following voting method. Before the parameter
combinations optimization, it is necessary to construct a set of parameter combinations that all parameters could be built up, and all the parameter combinations can
be obtained by method of the following.
The feature parameter combination is denoted a H and the number of clusters is
denoted as I. Then we can obtain a cluster CL, where CL = CL (H, I). The combined
sum of number of variable H is N and the number of variable I is L
N =
K
X =M
K
X
(K is max number of feature, X is the feature selected)
L = P − 1 P( is max number of clusters).
237
21.2 Pattern Recognition Analysis Method
The creep failure process of C/SiC involves several different damage mechanisms,
for example, fiber breakage, matrix cracking and interface debonding. The complex
evolution of these damage modes determines the creep life of C/SiC. According to
their physical background, different damage mechanisms exhibit different acoustic
emission characteristics. Pattern recognition analysis is required to identify the
damage mode of each acoustic emission event, and thus clarify the damage evolution
during the whole creep experimental.
The K-means algorithm, an unsupervised pattern recognition method, was chosen
to determine the clustering of the AE signals. This mainly consists of three parts:
constructing a range of feature parameters, identifying the appropriate combination
of these parameters and applying K-means clustering analysis on this combination.
This protocol allows all relevant parameters to be included, and it avoids the impact
of human factors such as bias and other spurious effects. The optimal combination
of binary variables is determined through a voting method, thus the most reasonable
feature parameters suitable for CMC AE signal pattern recognition are obtained.
Different types of damage show different AE signal features, and different damage
types can be identified through K-means clustering. All C/SiC acoustic emission
pattern recognition analysis methods can be simplified as shown in Fig. 21.1.
21.2.1 AE Parameter Combination and k-means Cluster
In traditional AE pattern recognition, the selection of feature parameters is relatively
fixed, and those parameters are determined by those arising from metals, which are
not suitable for CMC. AE signals contain a lot of information, but much of this
is invalid for pattern recognition, so it is essential to determine feature parameters
suitable for CMC analysis.
Different parameter combinations have different effects on the eventual pattern
recognition, so many different parameter combinations were constructed in order
to create variables suitable for the following voting method. Before the parameter
combinations optimization, it is necessary to construct a set of parameter combinations that all parameters could be built up, and all the parameter combinations can
be obtained by method of the following.
The feature parameter combination is denoted a H and the number of clusters is
denoted as I. Then we can obtain a cluster CL, where CL = CL (H, I). The combined
sum of number of variable H is N and the number of variable I is L
N =
K
X =M
K
X
(K is max number of feature, X is the feature selected)
L = P − 1 P( is max number of clusters).