Abstract: Flue gas generator set is a kind of large high-speed rotating machinery in petrochemical industry, which function is transfer the energy in catalytic cracking regeneration flue gas into mechanical energy. The problem of extracting state feature weak information under large amount of background noise need to be resolved in order to ensure the flue gas generator set operation safe and stable. To this end, noise suppression algorithms on the basis of Birgé-Massart penalty function strategy is researched to obtain signal wavelet transform modulus maximum and threshold. The threshold is obtained through penalization strategy provided by Birge-Massart, then, different modulus maximum vertex neighborhood are constructed in different wavelet transform decomposition scales which influence the search process of modulus maximum point, so, the appropriate modulus maximum points sequence are obtained on various wavelet decomposition scales. Finally, signals are reconstruct using Mallat staggered projection. In order to validate the effectiveness of the algorithm, it was compared with four kinds of threshold noise suppression methods namely Rigrsure, Sqtwolog, Heursure, Minimaxi, and the results show that this algorithm has a better signal to noise ratio and mean-square error. The algorithm is applied in to extracting state feature weak information of vibration signals collected on the industrial scene, the results show that the signals after processed has good smoothness while retaining the mutation point and can be used for state feature information extraction of exhaust gas generator set.

Key words: Flue gas generator set, Modulus maximum, Punishment strategies, Weak information extraction