<p>In high-precision manufacturing industries, such as aerospace and aviation, ensuring product conformity to quality specifications is critical for both functionality and safety. When quality characteristics exceed their specification limits, the product becomes nonconforming, and recovery through rework is required to adjust these characteristics and bring them within their defined specification limits before advancing to subsequent manufacturing stages. For products with complex geometric features, such as turbine blades that are produced through multistage manufacturing systems (MMS), interdependence among quality characteristics poses challenges to the rework process. Reworking characteristics that exceed their specification limits at a certain stage can cause other characteristics, initially within limits, to deviate from their specification limits, potentially resulting in further nonconformities. This article proposes a graph-based recovery optimization approach that integrates graph neural networks (GNN) with optimization algorithms to recover nonconforming products in MMS. The novelty resides in formulating nonconforming product recovery in MMS as a graph-based constrained optimization problem. The GNN models the relationships among quality characteristics during rework and predicts how rework of characteristics exceeding their specification limits impacts those already within their specification limits. Based on these modeled relationships, the optimization algorithms determine the optimal rework values for nonconforming characteristics to bring them within specification limits while preserving the conformity of other characteristics. The approach also incorporates a decision-support mechanism to identify products that can be recovered and those for which recovery would be unsuccessful. This step enables early rejection of infeasible rework attempts to save resources. The proposed approach was validated using real-world manufacturing data from an aviation turbine blade production process. The results demonstrate the ability of the approach to recover nonconforming products by achieving a 90.2% recovery success rate while minimizing the risk of additional nonconformities. The decision-support mechanism successfully identified products unsuitable for recovery, thus enabling early rejection and reducing resource waste. These findings highlight the potential of the proposed approach to advance zero-defect manufacturing, improve recovery efficiency, reduce non-value-added operations, and promote sustainability in high-precision MMS.</p>