This contribution explores the opportunities offered by physics-constrained automatic PDE learning, see for example [1], for identifying defects in mechanical structures. In this framework, defects are seen as limited zones in a domain where the constitutive law is different from that of the healthy material. One challenge in defect localization is that the impact of the defect on the constitutive law has to be known in order to perform its identification (for example by assuming the defect only affects Young modulus). The accuracy of the identification suffers when the impact of the defect in question is not fully known. The idea of the present contribution is thus to identify automatically the alteration of the constitutive law caused by the defect, as well as its localization. The resulting identification approach makes use of sparse regularization techniques [2] both for the constitutive law and the spatial localization of the defect. It will be applied to slender beams structures, and tested numerically for identification of fictive linear and nonlinear defects. Finally, the identification approach will be confronted to experimental data obtained through a simple dynamical test on a polymer beam. The input data for the algorithm will be obtained through post-processing of a series of digital snapshot of the sample during the test. REFERENCES [1] M. Flaschel, S. Kumar, and L. De Lorenzis. Automated discovery of generalized standard material models with EUCLID. Computer Methods in Applied Mechanics and Engineering, 405:115867, 2023. [2] Z. Zhang, Y. Xu, J. Yang, X. Li, and D. Zhang. A survey of sparse representation: algo- rithms and applications. IEEE access, 3:490–530, 2015.