In intelligent manufacturing environments, the real-time awareness of material properties during heat treatment is essential for ensuring both product reliability and process stability. However, key internal states such as phase evolution and mechanical performance cannot be directly accessed using conventional physical sensors. In this study, a data-driven virtual sensor framework is developed for SAE 1038 medium-carbon steel to enable the nondestructive, real-time estimation of mechanical properties. Unlike conventional static regression models, the proposed framework incorporates the ideal critical diameter (DI) as a material descriptor to enable the adaptive normalization of batch-to-batch chemical variations. The virtual sensor integrates a comprehensive heat treatment database established through high-resolution dilatometry, Jominy end-quench testing, and mass effect analyses across multiple specimen diameters. Metallographic observations and SEM fractography are employed to validate the correspondence between predicted properties and underlying microstructural evolution. By coupling DI and tempering temperature into the proposed sensing logic, hardness and ultimate tensile strength are predicted with maximum errors of less than 5%. Furthermore, fracture morphology analysis reveals a systematic transition from quasi-cleavage to microvoid coalescence with increasing tempering temperature, providing a microstructural basis for defining the safe operating regime of self-sensing fasteners. The proposed virtual sensing framework demonstrates a practical and scalable approach for transforming conventional structural steels into data-enabled materials, offering a viable pathway toward real-time quality assurance and integration into intelligent heat treatment systems.