针对航空发动机气路传感器在高温高压等运行环境下易发生漂移与失效，而传统硬件冗余方案存在体积、成本与共模故障约束的问题，论文提出了一种物理信息异构神经网络驱动的可测气路参数自适应虚拟感知与故障隔离重构方法。首先，基于气路定向因果依赖关系与部件物理特性，构建了物理信息异构体，通过部件级联子网络系统、自适应加权复合损失函数及分层训练策略，实现了数据精度与物理一致性的协调优化；其次，提出了基于弹性权重巩固的增量学习自适应更新机制，实现了历史知识保留与新增退化数据的协调学习；然后，建立了单传感特异性隔离准则与专用高精度重构模型，形成了故障检测、隔离与重构一体化方案；最后，基于某型小涵道比涡扇发动机的临近架次与长航时真实飞行数据，实现了对可测气路参数的精确跟踪，并利用真实传感漂移数据验证了其故障早期定位与信号重构能力，为机载传感容错与预测性维护提供工程化技术支撑。

Aero?engine gas?path sensors are vulnerable to drift and failure under harsh conditions such as high temperature and high pressure. Conventional hardware redundancy, on the other hand, suffers from inherent limitations in terms of volume, cost, and common?mode faults. To address these issues, this paper proposes a physics?informed heterogeneous neural-network?driven approach for adaptive virtual sensing and fault?isolation reconstruction of measurable gas?path parameters. First, a physics?informed heterogeneous body is constructed by exploiting the directional causal dependencies inherent in the gas?path and the physical characteristics of individual components. The network is trained with a component?cascaded sub?network architecture, an adaptively weighted composite loss, and a hierarchical strategy, which collectively reconcile data fidelity and physical coherence. Then, an incremental learning adaptation mechanism based on elastic weight consolidation is introduced, which enables harmonious retention of historical knowledge and effective assimilation of newly acquired degradation data. Moreover, single?sensor?specific isolation criteria and dedicated high?fidelity reconstruction models are established, thereby forming an integrated framework for fault detection, isolation, and reconstruction. Finally, the proposed method is validated using real flight data from successive sorties and long?endurance missions of a low?bypass?ratio turbofan engine, demonstrating accurate tracking of measurable gas?path parameters. Furthermore, real sensor drift data are employed to verify its early fault?localization and signal?reconstruction capabilities. This work provides engineering?oriented technological support for onboard sensor fault tolerance and predictive maintenance.