Increased demands on the reliability of key equipment have created the need for its important components to be non-destructively inspected after main production processes, and detected on-line in the working state. Due to the difficulty of existing structural damage identification methods in meeting requirements for convenience or efficiency, a new model based on acoustic signals is constructed, with advantages such as non-contact measurement, high efficiency, and on-line detection. The emphases of the model are signal denoising, feature extraction, and the sensitive feature selection of damage based on distance evaluation. This model is applied to crack damage identification of the cantilever beam and switch machine′s traction pole. The results show a 100% correct recognition rate, with high speed of recognition. Thus, this model is effective and feasible, and provides a new method for the identification of structural damage.