Measure and optimize sample confidence of acoustic signal for fault identification in ships

The shortage of sufficient “real” acoustic data under fault conditions on ships has been a challenge for accurate structure diagnosis. The common solution is generating semi-artificial data while the data quality is unknown. Thus, a kernel-based confidence measure (KBCM) is proposed for evaluating the expanded acoustic data. The deviation between the expanded and real data has been deduced to establish a KBCM model for measuring confidence. Furthermore, an optimization algorithm termed as the maximum class separability is formulated for the kernel optimization. Successful applications on experimental acoustics datasets under various fault conditions have demonstrated the proposed method's effectiveness.The shortage of sufficient “real” acoustic data under fault conditions on ships has been a challenge for accurate structure diagnosis. The common solution is generating semi-artificial data while the data quality is unknown. Thus, a kernel-based confidence measure (KBCM) is proposed for evaluating the expanded acoustic data. The deviation between the expanded and real data has been deduced to establish a KBCM model for measuring confidence. Furthermore, an optimization algorithm termed as the maximum class separability is formulated for the kernel optimization. Successful applications on experimental acoustics datasets under various fault conditions have demonstrated the proposed method's effectiveness.