This paper introduces the concept of secondary noise sources for passive defect detection and localization in structures. The proposed solution allows for the exploitation of the principle of Green's function reconstruction from noise correlation, even in the absence of an adequate ambient noise. The main principle is to convert a part of low-frequency modal vibrations into high-frequency noise by exploiting the frictional contact nonlinearities. The device consists of a mass-spring resonator coupled to a flexible beam by a rough frictional interface. The extremity of the beam, attached to the surface of a plate, excites efficiently flexural waves in the plate up to 30 kHz when the primary resonator vibrates around its natural frequency, i.e. a few dozens Hz. A set of such devices is placed at random positions on the plate surface, and low-frequency excitation is provided by a shaker. The generated high-frequency noise is recorded by an array of eight piezoelectric transducers attached to the plate. A differential correlation matrix is constructed by subtracting correlation functions computed from noise signals at each sensor pairs, before and after the introduction of a local heterogeneity mimicking a defect. A simple array processing then allows for the detection and estimation of the defect location from this differential correlation matrix. Beyond the successful proof of concept, influence of experimental parameters, such as the number of secondary sources or the variability of the position of the shaker application point, is also investigated.