Bilevel joint optimization for product design changes with a resilient supply chain based on deep reinforcement learning

High-tech supply chains are complex and capital-intensive, making them highly vulnerable to disruptions such as the COVID-19 pandemic and rising geopolitical tensions. Traditional risk management strategies — such as securing backup suppliers and stockpiling inventory — have revealed their limitations. In contrast, integrating product design change (PDC) strategies enables supply chains to adapt and restructure more flexibly, thereby enhancing resilience. This study emphasizes the critical interplay between PDC and resilient supply chains (RSC), proposing a design scheme that integrates both. Because PDC and RSC involve distinct decision-makers with hierarchical decision structures, we further construct a bilevel joint optimization (BJO) framework grounded in a Stackelberg game. To capture this dynamic, a novel bilevel 0–1 nonlinear programming model is formulated, and a bilevel deep reinforcement learning (DRL) algorithm is developed to solve it. The proposed DRL approach effectively handles multi-dimensional discrete decision variables and complex constraints. We conduct case studies on the smartphone industry. Comparative experiments reveal that the proposed BJO framework significantly outperforms traditional approaches. Moreover, our bilevel DRL algorithm achieves better results than conventional DRL and heuristic methods. Results from resilience-oriented experiments confirm that integrating both resilience and product design into supply chain activities significantly influences RSC decisions, thereby contributing to enterprise risk mitigation. We further conduct a sensitivity analysis on composite utility, design tolerance, and manufacturing adaptability tolerance parameters. The results show that changes in these parameters substantially affect the objectives of the decision-makers. Based on these findings, we offer practical managerial insights to guide real-world implementation.