21 High-Temperature Creep Damage Evolution of C/SiC …
245
Fig. 21.5 Three-dimensional distribution of sample 6 cluster results
clustering for a single feature. The data in Fig. 21.4 are derived from sample No. 6,
it may be observed that the overlap between different clusters is small. Clustering
C, which represents low frequency signals, has a large separation distance from
classes A and B, while the separation distance of class A and class B is relatively
small. In subgraph (2, 3) (the subgraph of the third row in the second column), it is a
subgraph of frequency band 3 and center frequency, which shows an obvious overlap
between class A and class B. Other subgraphs can also clearly distinguish different
types of data. Meanwhile, there are two small clusters in subgraphs (1, 3), (1, 4),
(2, 4), and (3, 4), the distinguishing feature between them being peak frequency or
weighted frequency. In addition, the subgraph on the diagonals shows that the peak
frequency characteristics alone can distinguish different categories well. Therefore,
peak frequency will be used as the main feature to identify the damage evolution
process of materials in the subsequent analysis of material damages.
Figure 21.5 shows the clustering distribution map of different feature combinations. The data source is the same as Fig. 21.4, namely sample No. 6. It shows no
overlapping data, and subgraph (2, 2) should be noted. It is a subset of Fig. 21.4
(2, 3), but the overlap that was present originally is no longer present in the threedimensional subgraph. Therefore, it is more advantageous to use multiple variables
with three-dimensional graphical representation for distinguishing different data
samples.
245
Fig. 21.5 Three-dimensional distribution of sample 6 cluster results
clustering for a single feature. The data in Fig. 21.4 are derived from sample No. 6,
it may be observed that the overlap between different clusters is small. Clustering
C, which represents low frequency signals, has a large separation distance from
classes A and B, while the separation distance of class A and class B is relatively
small. In subgraph (2, 3) (the subgraph of the third row in the second column), it is a
subgraph of frequency band 3 and center frequency, which shows an obvious overlap
between class A and class B. Other subgraphs can also clearly distinguish different
types of data. Meanwhile, there are two small clusters in subgraphs (1, 3), (1, 4),
(2, 4), and (3, 4), the distinguishing feature between them being peak frequency or
weighted frequency. In addition, the subgraph on the diagonals shows that the peak
frequency characteristics alone can distinguish different categories well. Therefore,
peak frequency will be used as the main feature to identify the damage evolution
process of materials in the subsequent analysis of material damages.
Figure 21.5 shows the clustering distribution map of different feature combinations. The data source is the same as Fig. 21.4, namely sample No. 6. It shows no
overlapping data, and subgraph (2, 2) should be noted. It is a subset of Fig. 21.4
(2, 3), but the overlap that was present originally is no longer present in the threedimensional subgraph. Therefore, it is more advantageous to use multiple variables
with three-dimensional graphical representation for distinguishing different data
samples.