Data and Decision Level Fusion-Based Crack Detection for Compressor Blade Using Acoustic and Vibration Signal

Crack detection is important for compressor blades. Due to the interference of strong noise, the existing methods based on single signal reach unsatisfactory performance. To improve the reliability and accuracy, a novel crack detection method is proposed by fusing the information from acoustic and vibration signals (AVS) in data and decision level. Firstly, the data-level fusion method is proposed by calculating Hoyer and improved cosine similarity, which is applied for vibration signals fusion based on its sparsity and similarity. Then, the raw AVS and data-level fusion samples are trained in the one-dimensional convolutional neural network (1D CNN), and the preliminary classification is obtained based on the probabilities of different cracks. Finally, the decision-level fusion method with changed reliability assignment is proposed through modifying and correcting the preliminary results of 1D CNN, which can reach reliable decisions and realize crack detection for compressor blade. The proposed method is tested by compressor experiments with three cracked blades under four single and one mixed working conditions. The results illustrate that the proposed method can make full use of AVS and detect cracks reliably under five working conditions. Furtherly, the superiority of the proposed method is validated by comparing with other crack detection approaches.