Research on noise reduction algorithm of unmanned surface vessels fault signal based on frequency stacked noise reduction self-encoder

The relentless advancement of technology and the swift development of unmanned systems have positioned unmanned surface vessels (USV s) as versatile, multi-mission small platforms. These USV s exhibit substantial potential and value in various domains, including marine research, offshore engineering, and military operations. However, as the complexity of their operating environments increases, the reliability and safety issues of unmanned surface vessels have become increasingly prominent. Therefore, promptly and accurately diagnosing faults in USV s is crucial for ensuring their safe operation. However, the fault data of USV s often contains significant noise interference, which affects the accuracy of fault diagnosis. To address this problem, this paper proposes a fault signal denoising method for unmanned surface vessels based on frequency-stacked denoising autoencoders. First, the employment of a moving average filtering technique as an initial noise reduction strategy. Secondly, this study introduces a frequency stacking denoising autoencoder capable of identifying and extracting noise features, thereby facilitating profound denoising of the initially denoised vibrational signals from USV s propellers. To verify the effectiveness of the proposed method, this paper constructed a specialized experimental platform and conducted a comprehensive test using real-world operational data from USV s. The experimental results show that using this method to denoise vibration signals for USV s fault diagnosis achieves an average accuracy of 94.44 %. This represents a 20% improvement in both accuracy and reliability compared to using undenoised vibration signals, providing high-quality data support for precise fault feature identification and diagnosis.