17 Cluster Analysis of Acoustic Emission …
187
17.2 Tensile Test
C/SiC was manufactured by the Science and Technology on Thermo-structural
Composite Materials Laboratory of Northwestern Polytechnic University. The material mainly consists of C fibers, PyC and SiC matrix. The fiber volume content is
about 40%, the porosity is about 15%, and the density is 2.1–2.2 g/cm
3 .
The room temperature monotonic tensile test was carried out on Instron 8801
hydraulic servo fatigue testing machine. The loading rate was 0.5 mm/min. During
the test, online monitoring was performed using the AE machine (PAC PCI-II) of the
American Physical Acoustics Corporation (front amplification gain: 40 dB, signal
threshold: 45 dB, sample rate: 1MSPS). After the tensile test, the specimen cut-off
profile was observed using the Hitachi TM4000PLUS benchtop scanning mirror.
17.3 Clustering Analysis Based on Improved K-Means
Algorithm
The most commonly used clustering algorithm in AE signal analysis of C/SiC material damage is K-means, however, it may convert accurately and lose the global
optimal solution due to the influence of the initial clustering center. Therefore, this
paper introduces genetic algorithm, through the deep fusion of genetic algorithm
and K-means algorithm to solve the problem of K-means algorithm, so to achieve a
more reasonable clustering of AE data. Through the unsupervised clustering method
based on the improved k-means algorithm, the AE signals of 2D-C/SiC specimen
in the process of tensile damage is analyzed, and the damage mode identification
of the material under monotonic tensile load can be realized. To introduce genetic
algorithm to unsupervised clustering analysis, the following problems should be
solved: (1) the consistency of individual gene coding in genetic algorithm; (2) the
construction of fitness function and genetic operation; (3) the selection of clustering
effectiveness evaluation index. The flow of clustering analysis based on improved
k-means algorithm is shown in Fig. 17.1.
17.3.1 Gene Label Conversion in Individuals
Before using genetic algorithm to get the final result, it is necessary to solve the
problem of cluster label inconsistency. For example, the clustering {1, 1, 2, 2, 2,
3, 3, 3} and {2, 2, 3, 3, 3, 1, 1, 1}. Although their expressions are different, they
represent the same result. The corresponding relationship between basic clusters
must be established. For the clustering problem with m (m ≥ 2) base clusters, the
relative position relationship between the cluster center and the coordinate origin is
used to number all kinds of centers, then the corresponding element number of each
187
17.2 Tensile Test
C/SiC was manufactured by the Science and Technology on Thermo-structural
Composite Materials Laboratory of Northwestern Polytechnic University. The material mainly consists of C fibers, PyC and SiC matrix. The fiber volume content is
about 40%, the porosity is about 15%, and the density is 2.1–2.2 g/cm
3 .
The room temperature monotonic tensile test was carried out on Instron 8801
hydraulic servo fatigue testing machine. The loading rate was 0.5 mm/min. During
the test, online monitoring was performed using the AE machine (PAC PCI-II) of the
American Physical Acoustics Corporation (front amplification gain: 40 dB, signal
threshold: 45 dB, sample rate: 1MSPS). After the tensile test, the specimen cut-off
profile was observed using the Hitachi TM4000PLUS benchtop scanning mirror.
17.3 Clustering Analysis Based on Improved K-Means
Algorithm
The most commonly used clustering algorithm in AE signal analysis of C/SiC material damage is K-means, however, it may convert accurately and lose the global
optimal solution due to the influence of the initial clustering center. Therefore, this
paper introduces genetic algorithm, through the deep fusion of genetic algorithm
and K-means algorithm to solve the problem of K-means algorithm, so to achieve a
more reasonable clustering of AE data. Through the unsupervised clustering method
based on the improved k-means algorithm, the AE signals of 2D-C/SiC specimen
in the process of tensile damage is analyzed, and the damage mode identification
of the material under monotonic tensile load can be realized. To introduce genetic
algorithm to unsupervised clustering analysis, the following problems should be
solved: (1) the consistency of individual gene coding in genetic algorithm; (2) the
construction of fitness function and genetic operation; (3) the selection of clustering
effectiveness evaluation index. The flow of clustering analysis based on improved
k-means algorithm is shown in Fig. 17.1.
17.3.1 Gene Label Conversion in Individuals
Before using genetic algorithm to get the final result, it is necessary to solve the
problem of cluster label inconsistency. For example, the clustering {1, 1, 2, 2, 2,
3, 3, 3} and {2, 2, 3, 3, 3, 1, 1, 1}. Although their expressions are different, they
represent the same result. The corresponding relationship between basic clusters
must be established. For the clustering problem with m (m ≥ 2) base clusters, the
relative position relationship between the cluster center and the coordinate origin is
used to number all kinds of centers, then the corresponding element number of each