Research On Recognition Of Bolt Connection Loosening State Based On Fractal Theory

Bolted joint structure is the most common way to connect components and fixed structures,which is widely used in aviation,bridge,ship,civil engineering and other fields.Its preload state plays an important role in the safe operation of mechanical equipment.However,due to the complex overall structure of bolted connection and the discontinuity of structural stiffness and damping,the collected signals often contain complex nonlinear components,and it is difficult to accurately extract features based on time-frequency domain method.The fractal theory has been widely concerned because it can realize fault diagnosis without dependence on the time-frequency domain.Therefore,this thesis studies the state recognition problem of bolted connection structure based on the fractal theory.Specific research contents are as follows:(1)In order to obtain the correlation dimension based on the single fractal theory,the traditional G-P method relies on subjective selection in the process of selecting the scale-free interval,which increases the randomness of the result.This thesis proposes an improved G-P algorithm based on density peak clustering algorithm to automatically select the scale-free interval.In this method,the second-order difference values of all the points set in the scale-free interval were classified by density peak clustering,and the best correlation part of the data near zero in the clustering results was determined as scale-free interval by combining the statistical theory,which improved the calculation accuracy of correlation dimension.The effectiveness of the proposed method was verified by Lorenz simulation system.Finally,the proposed method was applied to the identification of the loosening state of the bolt connection structure of the aero-engine rotor,and the recognition of different loosening states was realized by establishing a prior interval with the correlation dimension under different states.(2)Since the single fractal can only characterize and judge the nonlinear system from the whole,it cannot describe the inherent singularity of the structure from the local,and it also has the defects of establishing prior interval and requiring strict data requirements.A multifractal detrended wave analysis method combined with probabilistic neural network is proposed.The multi-fractal detrending wave analysis method was used to process the data in different states,and the multi-fractal spectrum was obtained to achieve the singularity characterization of different states.The multifractal spectrum parameters formed a multi-dimensional vector as the input of probabilistic neural network for training and testing.The results show that the proposed method can recognize different loosening states,and the effectiveness of the proposed method is verified by comparison.(3)The dimension of feature vector extracted from multifractal spectrum is too high,which leads to the confusion of recognition results,as well as the defect of artificial selection of smoothing factor in probabilistic neural network.Therefore,a dual optimization MPSPNN model based on principal component analysis and sparrow search algorithm is proposed to automatically determine the optimal smoothing factor.Firstly,the dimensionality reduction of the nine-dimensional feature set was realized by the principal component analysis method,so as to eliminate the redundant aliasing components and retain the best principal elements to form a new feature vector.Then,the new feature vector was trained and tested by combining the probabilistic neural network optimized by sparrow search algorithm.The results show that the proposed MPSPNN model is more accurate by comparison.