Abstract: The signal features of CFRP T-joint disbond under static extension testing were investigated by piezoelectric sensors and active Lamb wave monitoring technology, and disbond damages were identified with improved BP artificial neural networks. The experimental results show that interfacial disbond appears firstly in T-joint triangle filling area and then extend to flanges. Both of the signal energy and least square peak factor linearly decrease with time before failure, which can be used to describe disbond extension of T-joint. The network training data improved by adaptive particle swarm optimization algorithm are corresponded to experimental results with error range of 3.8%~4.7%.