In this paper, to increase the performance of the sparse reconstruction method in real complex engineering structures an adaptive dictionary learning framework is proposed which updates the dictionary matrix, to allow improved compatibility with the complex structure. This proposed framework was developed by combining analytical modeling with training data sets and learning methods. An experimental evaluation of the proposed dictionary learning framework was performed on an anisotropic composite plate with a stiffener. In this experimental evaluation, a moving magnet was used as the artificial damage to capture the training data set, and both artificial damage in several locations and real impact damage was used for detection and location of the target damage. The obtained results confirmed the concept of the proposed dictionary learning framework for the improved health monitoring of complex structures.@en