Adaptive noise cancellation based on CEEMDAN-LMS for pipeline leak location under low SNR

Abstract

To reduce leak localization errors caused by background noise interference, an adaptive noise cancellation method based on the combination of complete ensemble empirical mode decomposition with adaptive noise and LMS filter (CEEMDAN-LMS) is proposed for pipeline leak location. First, the CEEMDAN is used to decompose the detection signal into several intrinsic mode functions (IMFs). To improve the algorithm’s adaptability and accuracy under low signal-to-noise ratio (SNR) conditions, an IMF selection strategy based on the cross-correlation (CC) function between each IMF and another detection signal is proposed. According to this strategy, the IMFs are classified into signal components and noise components. The signal components are then further denoised using an LMS filter. The quality of the reference noise in the LMS filter determines the filtering effectiveness. To obtain high-quality reference noise, this paper proposes a novel reference noise generation method, where selected noise components are summed and then decomposed again using CEEMDAN to create the reference noise. This reference noise is extracted from the original signal, highly correlated with the noise in the original signal, and has a narrow frequency band, which effectively improves the filtering performance. Finally, the filtered signal components are recombined to reconstruct the denoised signal. The time delay of the denoised signal is estimated using the CC function to pinpoint the leak location. In the simulation, two detection signals with initial SNRs of −18 dB were processed by CEEMDAN, resulting in improvements of 9.54 dB and 9.09 dB, respectively. After further processing with the LMS filter, the improvements were 1.9 dB and 1.97 dB, respectively. In real experiments, the localization error using direct CC of the unprocessed signal was 10.85 %, while the error after CEEMDAN-LMS processing was reduced to 2.99 %, showing a significant improvement in localization accuracy.