In the presence of sustained random vibration loads， mechanical structures easily induce their own partial-order modes， leading to resonance and fatigue failure. Life expectancy analysis effectively prevents structural fatigue failure， exhibiting important engineering significance. Diverging from traditional approaches that establish empirical models based on probability density functions， this study focuses on the power spectral density segmentation. Combining theoretical derivation and numerical simulation， the research explores the dividing criteria for power spectral density resonance and non-resonance regions， investigates the impact of segmentation forms on damage estimation， and explores the coupling effects among multi-modal responses. The results indicate that considering only the power spectral density resonance region enhances damage estimation accuracy. The segmentation form has no effect on damage estimation. The proposed expression for multi-modal coupled damage estimation demonstrates superior applicability to random vibration fatigue compared to commonly used Dirlik and T-B methods.