Automated fiber placement （AFP） can effectively improve the manufacturing efficiency and quality of composite components. To meet the mechanical and placement requirements， the tow paths need to be optimized in terms of turn radius， fiber angle deviations and fiber coverage in the target domain. Most of the existing planning methods are generally divided into two steps： Initial path generation and path offset， which can hardly to compromise the contradictive requirements and to satisfy multiple optimization objectives as a whole. To solve this trade-off issue， this paper defines the global metric of paths for each requirement， converts the path planning problem into a generalized function problem in the parameter domain， uses physics-informed neural network （PINN） to solve the objective function， and finally extracts the objective function contour as the result of path planning. Compared with exiting path planning strategy， the proposed method finds global optimal tow paths with greatly improved alignment while maintaining small ply thickness variation and large turning radius， providing new paradigm for AFP path planning on complex part.