Gas turbine harmonic detection and modal identification based on underdetermined blind source separation

In gas turbine operational modal parameter identification, traditional methods like operational modal analysis or underdetermined blind source separation (UBSS) struggle with underdetermined time-delay mixture and harmonic interference. Based on the signal sparsity in the energy domain, this paper proposes a novel method based on improved UBSS with binary time-frequency masking (BTFM). First, signals are transformed to the time-frequency domain, and then to the energy domain by integrating over time. And peak frequency points are detected frequency energy sum curve from each channel. Second, to distinguish source signals, cosine distances between peak frequency points and other ones are calculated, and BTFM is constructed. Third, source signals are recovered and padding lines are added to reduce boundary effects. Finally, to detect harmonic components in the source signals, the probability density and kurtosis of each signal are calculated. Based on the separated modal response signals, the modal frequency of each order is identified, and the modal modes are determined using the mixing matrix. The effectiveness of the proposed method is validated through comprehensive analysis on a simulation system, a three-rotor test bench, and gas turbine datasets. Results show that it outperforms existing methods in achieving more accurate and reliable modal identification under harmonic interference. The proposed method facilitates operational modal identification and condition monitoring for large-scale equipment such as gas turbines, thereby providing guidance for structural optimization and vibration/noise reduction.