Piezoelectric ceramics have recently become one of the most used materials in all kinds of electromechanical systems. However, the presence of defects in such materials prevents them from fulfilling their function. A number of numerical, analytical and experimental works are recently being developed to understand the behaviour of piezoelectrics with presence of damage, but very few aimed at locating defects. One of the current challenges in monitoring piezoelectrics is the correct interpretation of the readings from sensors, in order to reliably recover the defect characteristics minimizing uncertainties due to noise and model. An inverse problem strategy is proposed for this reconstruction, starting from the electromechanical response measurement as input data, and incorporating a numerical model that simulates that response. This model is solved using a Boundary Element Method (BEM), whose formulation is developed for the 2D static case. The damage identification inverse problem is solved using genetic algorithms for the minimization of the discrepancy or cost functional. The effect of noise on measurements and uncertainties in the model is studied in detail through a sensitivity analysis for some simple cases of defect.