| Blades are key components in major equipment such as aero-engines and gas turbines,and online vibration monitoring is of great significance.Blade tip timing(BTT)is a cutting-edge technology for non-contact blade vibration on-line monitoring,but its inherent disadvantage is severe under-sampling.In practical applications,it must be reconstructed to obtain accurate blade vibration characteristics.Existing blade tip timing vibration signal reconstruction methods often assume a constant speed,and have disadvantages such as the need for prior knowledge and low reconstruction efficiency.It is difficult to meet the needs of rapid reconstruction of blind multi-frequency vibration of blades under variable conditions.Therefore,studying the reconstruction method of under-sampling blade tip timing vibration signal under variable conditions has important theoretical significance and application value.In view of the sparse characteristics of blade multi-frequency vibration response,this thesis cuts in from the perspective of compressed sensing and introduces deep learning theory,constructs a compressed sensing model of blade tip timing vibration signal under variable conditions,and proposes a convolutional neural network-based compressed sensing model.The sparse reconstruction method of deep learning is verified by simulation and experiment.The main work and conclusions of the thesis include:1.Aiming at the problem of non-constant blade speed under variable conditions,from time domain sampling to angular domain sampling,the blade tip timing measurement model of blade vibration signal angle domain under variable conditions is established,and on this basis,the variable conditions are constructed.The order-domain compressed sensing model of the blade tip timing vibration signal can effectively reduce the impact of variable conditions.2.Aiming at the compressed sensing model of the above-mentioned blade tip timing vibration signal,the equivalent cross-correlation coefficient is introduced to derive its reconfigurable sufficient conditions,and the optimal criteria for the number of blade tip timing sensors and their layout are established,and its effectiveness is verified by numerical simulation.3.In view of the above-mentioned unknown multi-band vibration signal reconstruction problem under variable conditions,the deep learning method is introduced,the deep compressed sensing sparse reconstruction model based on convolutional neural network is constructed,and an improved convolutional neural network structure is designed.The simulation results show that this method is significantly better than the traditional compressed sensing sparse reconstruction algorithm in terms of anti-noise influence and reconstruction fidelity.4.The blade tip timing vibration measurement experimental system was built,and the blade tip timing vibration measurement signal under variable speed was collected,and the deep compressed sensing method was experimentally verified.The results show that the proposed method can accurately reconstruct the synchronous vibration of the blade without prior knowledge of blade vibration. |
