The scale of Industrial Internet networks continues to grow, and equipment intelligence continues to improve. Traditional Industrial Internet security solutions can no longer meet the requirements of privacy and efficiency. Trust management is an effective way to promote security, efficiency, and scalability in Industrial Internet networks. In this article, a dynamic trust management model for the edge devices in the Industrial Internet based on the feedback under the edge computing architecture is proposed. We divide trust into two parts, direct trust and indirect trust. The model calculates the direct trust between devices based on long-term and short-term trust. Long-term and short-term trust help honest devices restore their trust as soon as possible after experiencing short-term trust decline. They can also prevent dishonest devices from obtaining higher trust due to short-term honest behavior. Besides, the model calculates the indirect trust of the edge broker to devices based on the reward-punishment mechanism. The mechanism improves the accumulation speed of the indirect trust of honest devices and reduces the indirect trust of dishonest devices rapidly. The trust fusion algorithm is then used to aggregate the direct and indirect trust to calculate the comprehensive trust of the device. Experimental results show that the proposed model outperforms the existing methods in local and global performance in different attack scenarios.