Abstract: To obtain the main vibration sources of the mechanical systems, the method to estimate the contributions of the vibration sources via multi-unit deflation constraint independent component analysis (cICA) is proposed. In cICA, the prior knowledge of the source signals is introduced to the contrast function of the traditional independent component analysis (ICA) as constraints, which can not only improve the accuracy of the independent components and the robustness of the algorithm, but also obtain the corresponding relationships between the independent components and the vibration sources. To remove the components in mixed signals, which corresponds to the independent component extracted at each iteration, the multi-unit deflation approach is introduced to the cICA. In multi-unit deflation, the energy reduction of the mixed signals corresponds to the contribution of the extracted independent component. The proposed algorithm is applied to the shell excitation test bench and the cylindrical structure test bench, and then the mechanical vibration source separation and contribution calculation of the test benches are successfully completed.

Key words: constraint independent component analysis;contrast function;prior knowledge;source contribution computation, particle swarm optimization based on logarithmic decrement, adjustment rules, boiler, mathematical modeling, steam power system, thermoelectric power plant, turbine