In order to solve the problems of low mechanical properties, multi-physical field coupling, complex mechanism, many process parameters and high solidification rate of molten pool in powder feeding metal direct forming process, etc, a closed-loop feedback control system for adaptive forming process based on neural network PSD algorithm is designed. According to the real-time image signal of laser molten pool monitored by optical system, the image feature set of molten pool is constructed by image fusion and visual detection technology, the dynamic relationship between process parameters and molten pool size is analyzed and the precision identification system of forming quality is constructed. The adaptive control of process parameters in LMD forming process is realized by optimizing the PSD algorithm of neural network. The simulation results show that the adaptive controller has high control accuracy for the width of molten pool, and can obviously improve the forming quality of manufacturing process and ensure the stability of process.