In real-time Industrial Internet of Things (IIoT), e.g., monitoring and control scenarios, the freshness of data is crucial to maintain the system functionality and stability. In this paper, we propose an AoI-Aware Deep Learning (AA-DL) approach to minimize the Peak Age of Information (PAoI) in D2D-assisted IIoT networks. Particularly, we analyzed the success probability and the average PAoI via stochastic geometry, and formulate an optimization problem with the objective to find the optimal scheduling policy that minimizes PAoI. In order to solve the non-convex scheduling problem, we develop a Neural Network (NN) structure that exploits the Geographic Location Information (GLI) along with feedback stages to perform unsupervised learning over randomly deployed networks. Our motivation is based on the observation that in various transmission contexts, the wireless channel intensity is mainly influenced by distancedependant path loss, which could be calculated using the GLI of each link. The performance of the AA-DL method is evaluated via numerical results that demonstrate the effectiveness of our proposed method to improve the PAoI performance compared to a recent benchmark while maintains lower complexity against the conventional iterative optimization method.