The rising demand for composite materials in aerostructures requires sophisticated Structural Health Monitoring (SHM) technology with the ability to continuously detect damage for structural integrity prediction. Piezoelectrets are an attractive prospect for piezoelectric sensor networks. These electro-active polymers with induced piezoelectricity showcase high flexibility, lightweight, and ambient compatibility, and can be seen as an alternative for traditional ceramic-based sensors. However, their development is hindered by computational complexity in multi-physics modeling and limited material property databases. The present work consists of a new reverse engineering framework combining computational homogenization and machine learning to improve characterization of piezoelectret sensors. The method applies the Mechanics of Structure Genome (MSG) theory via the SwiftComp software to perform the homogenization of the representative volume elements while dealing with complex electrostatic effects near voids by modeling them as pseudo ”elastic air” phases with effective properties. Material characterization is stated as a multi-objective optimization problem that aims simultaneously at reducing error in empirical data fitting and ensuring physical soundness of the determined parameters with explicit regularization. Machine learning algorithms were developed and validated as surrogate models and translated into substantial increases in computational velocity with no deterioration of accuracy. The developed framework showed gains in efficiency with the machine learning surrogates producing significantly faster computational speeds compared with SwiftComp. Systematic validation by using commercial sensors with complete reference data confirmed the methodology’s reliability, and rigorous ablation studies established the mathematical necessity of the multi-objective formulation of dealing with under-determined inverse problems. Application to the piezoelectret sensors demonstrated perfect convergence for idealized geometries (MSE ∼ 10−10) and accurately diagnosed analytical model shortcomings for realizable micro-structures. The proposed approach opens doors to otherwise inaccessible exploration of the design space and parametric investigations crucial to practical sensor designs and creates a solid ground work for embedding piezoelectret-based SHM systems into next-generation aerostructures.