Vibrations onset represents a paramount issue in all grinding processes. The related surface defects appear in the form of micrometric waviness that decreases the finishing quality and in some cases the functionality of the ground workpieces: sometimes, these defects can be also marked on the grinding wheel surface. This paper presents an online model-based approach to identify and quantify the level of waviness starting from multiple acceleration measurements, allowing a continuous monitoring of wheel and/or workpiece defects. The identification algorithm, that exploits a linear model of machine and process dynamics, is based on the application of Least Squares method in the frequency domain. Experiments confirm the good performance of the algorithm that, hence, can be exploited for developing advanced control schemes of the grinding process