Recognition of subsurface defects in machined ceramics by application of neural networks to laser scatter patterns

摘要

Laser scatter has shown promise as a method to characterize damage microstructural variations as well as a method to characterize surfaces in optical translucent ceramics. Because large volumes of data need to be handled (and sorted) quickly, automated pattern recognition methods using neural networks have been implemented to recognize differences in patterns. A He-Ne laser ((lambda)=0.632(mu)) has been used to obtain scatter patterns from hot pressed Si(sub 3)N(sub 4) with various microstructural variations. By use of a backpropagation neural network running on an IBM PC clone 486/33 machine, a correlation was established between subsurface microstructure and position in Si(sub 3)N(sub 4) ball bearings. The data were confirmed by destructive analysis.