A novel modulation-sourced model and a modulation-carrier spectrum-based demodulation method for rotating machinery signal analysis

Abstract

Rotating machinery (RM) plays an essential role in modern industries. However, the complex internal motion behavior and vibration transmission paths lead to monitored signals characterized by various modulations, mainly amplitude modulation (AM), frequency modulation (FM), and their coupling effects. Although vibration signal models have been widely studied to interpret the vibration characteristics of RM, the generality and modulation properties of existing models remain to be improved. Furthermore, traditional frequency spectrum-based analysis and well-known demodulation methods face challenges in extracting modulation-carrier frequency pairs and the modulation types, thereby failing to thoroughly reveal the vibration characteristics of RM. Toward the above issues, firstly, a general modulation-sourced model is constructed by synthesizing as many AM-FM phenomena as possible to represent the monitored vibration signal of different RM, such as planetary gearboxes (PGs), rolling bearings (RBs), and so on. Secondly, drawing on the merits of the convolutional neural network (CNN), an elastic comb-shaped convolution kernel (ECCK) and a modified activation function called restricted-rectified linear unit (R-ReLU) are innovatively proposed to automatically obtain a modulation-carrier spectrum (MCS) from the frequency spectrum. Theoretical derivation of the MCS is provided to interpret the components in terms of the proposed model. More importantly, considering the inevitable pseudo demodulated components, a MCS-based demodulation strategy, the MCSD, is introduced to reveal the true modulation-carrier frequency pairs and their corresponding modulation types. Simulation studies with different modulation scenarios are analyzed to validate the effectiveness of the proposed method. Finally, experimental studies are conducted and four mainstream demodulation methods are picked for comparison, confirming the proposed method’s effectiveness and outperformance.