In this paper, we study the relationship between correlation dimension and signal structure based on nonlinear fractal theory. It shows that when the signal structure is more complex, the correlation dimension is higher. By analyzing background noise, the impact signals of loose parts, and the correlation dimensions of impact signals from background signals, we find that the change of correlation dimensions can reflect the situation of loose parts in reactor operation greatly, and the nonlinear feature of loose parts is consistent at some point. This method can be used to test the initial loosening of parts and to provide an effective way to improve the stability of loose-part monitoring systems.