In this article, research on the performance state assessment of compressor with damaged blades driven by realistic digital engine structural model is conducted, and a real-time aerodynamic performance simulation method for damaged compressor based on heuristic end-to-end in situ 3-D reconstruction is proposed. We present a heuristic coarse-to-fine 3-D reconstruction framework (HeurisRecon). Heuristic gated recurrent unit (GRU) is designed to introduce the prior knowledge of intact compressor drawn geometric model into the 3-D reconstruction of damaged blade. Based on the drawn geometric model, 2-D–3-D pose estimation method is adopted to solve the problem that structure from motion (SFM) cannot correctly estimate the camera extrinsics of borescope in the narrow internal space of aero-engine, while avoiding the cumulative error caused by the registration between damaged blade 3-D reconstruction model and compressor drawn geometric model. A high-performance image feature extraction network (TCNet) based on the feature interaction between transformer and convolutional neural network (CNN) is designed to provide multilevels of image features with rich semantic information for coarse-to-fine 3-D reconstruction. Transformer is used to achieve adaptive fusion in the back projection process from feature maps to feature volume. The realistic 3-D model of compressor with damaged blades obtained through our in situ reconstruction method is used for aerodynamic performance simulation evaluation. A simplified implicit difference scheme and $k{-}\varepsilon$ turbulence model are adopted to solve Navier–Stokes (N–S) equations in simulation. Experimental results verify that the proposed in situ 3-D reconstruction method can provide high-fidelity 3-D model for the aerodynamic performance simulation evaluation. This research provides a new idea for the performance state assessment of aero-engine in service and can be used to support the battle damage repair and assessment of aero...