Structural Response Data Source Prediction And Abnormal State Early Warning Of Quayside Container Cranes

Quayside container crane(QCC)is as an important mechanical equipment for port transportation,its structural safety can not be ignored.Structural health monitoring of QCC is helpful to find out the potential damage hazards of the structure in time,and is of great significance to maintain the structural performance safety and reliability of port machinery.In the process of long-term structural monitoring,it is bound to produce an amount of monitoring data.The effective preprocessing of monitoring data is the premise to ensure the accuracy of structural response prediction and abnormal state early warning of QCC.On the basis of monitoring data preprocessing,this paper studies the structural response prediction and abnormal state early warning method of quayside container crane based on data-driven from the two aspects of monitoring data-driven and simulation data-driven: Taking high-quality health monitoring data as the data source,carry out the research of structural response prediction and structural abnormal state early warning;Taking the damage data simulated by the finite element model as the data source,carry out the research of structural abnormal state early warning.The main research contents include:(1)This paper analyzes and discusses three kinds of data quality problems existing in the monitoring data,such as data missing,data anomaly and data noise.Aiming at the problem of missing data,this paper compares with the recovery effects of four common data recovery methods on different amounts of data,and chooses mean interpolation to repair the missing data according to the comparison results;Aiming at the problem of abnormal data,according to the distribution characteristics of monitoring data,the abnormal data is detected by using laida criterion,and processed by elimination method.Aiming at the problem of noise for monitoring data,the concept of sample entropy(SE)is introduced into dual tree complex wavelet decomposition,and an adaptive denoising method of monitoring data based on dual tree complex wavelet transform(DTCWT)and sample entropy is proposed.The dual tree complex wavelet transform combines the advantages of sample entropy,and use the change rules of sample entropy to determine the optimal layer of dual tree complex wavelet decomposition and quantify the denoising threshold of each decomposition scale to realize adaptive denoising of signal.The effectiveness of the proposed method is verified by simulation examples and engineering cases.(2)In order to capture the changes of structural state in time and improve the accuracy of structural response prediction,a combined prediction model based on dual tree complex wavelet transform combined with autoregressive moving average model(ARMA)and support vector regression(SVR)called DTCWT-ARMA-SVR is proposed.Firstly,the method based on dual tree complex wavelet transform and sample entropy is used to preprocess the monitoring data to reduce the impact of noise.Secondly,the preprocessed signal is decomposed by DTCWT again to obtain the characteristics of different scales of the signal.According to the characteristics of different scales of signals,ARMA and SVR prediction models are built respectively.Finally,the prediction models of each scale are fused into the final combined prediction model.Through two experimental cases of structural monitoring data of QCC,the prediction effect of the combined prediction model is verified,and the prediction performance of the model is evaluated.(3)Considering the uncertain factors in the response prediction process based on time series,in order to carry out dynamic early warning of the structural abnormal state in time and optimize the early warning interval,an interval early warning method of lower upper bound estimation based on correlation vector machine(RVM)and particle swarm optimization(PSO)is proposed.In the frame of LUBE,the time-domain features are extracted from the structural monitoring data,and two RVM models are used to replace the neural network(NN)to construct the upper bounds and lower bounds of the early warning interval of structural health feature indexes.Particle swarm optimization algorithm is introduced to optimize the kernel parameters of RVM to obtain the best early warning interval,and the interval quality is evaluated.This method is applied to the structural health monitoring of QCC.The performance of early warning interval is verified by using the healthy structural monitoring data,and the effectiveness of this method is further verified by comparing with other methods of constructing early warning interval.(4)In order to solve the problems of insufficient damage conditions,incomplete data and incomplete damage information in the practical monitoring data set of QCC,the structural finite element model of QCC is established.Different damage conditions are simulated by reducing the stiffness of different elements,and the simulation data set is constructed.On the basis of having certain damage samples,the structural abnormal state early warning method based on fuzzy entropy ratio variation deviation(FERVD)is proposed.The characteristics of response signal are extracted based on the adaptive frequency band and fuzzy entropy of dual tree complex wavelet decomposition,and FERVD is constructed as a new structural damage early warning index.According to the dynamic changes of FERVD between healthy and damaged states,set the adaptive early warning threshold of the index,and analyze the anti-noise and damage sensitivity of the proposed early warning index.Finally,the early warning effect of the early warning strategy is verified by simulation data and monitoring data.(5)In order to solve the problem that the structural damage early warning results based on single time series are not comprehensive enough and the spatial correlation of different monitoring points of the structure is not considered,a structural abnormal state early warning method based on multivariate state estimation techniques(MSET)and kernel principal component analysis(KPCA)is proposed.For the structural multi monitoring point signals of QCC,KPCA is used to analyze the correlation of the multivariate time series,choose the main modeling variables,and establish MSET model of structural health condition.The difference between the observation vector and the estimation vector is analyze by using the similarity function,and the average similarity sequence is constructed by using the sliding window method.The early warning threshold of the average similarity sequence is determined according to the mean and minimum value of the average similarity sequence.The effectiveness and feasibility of the early warning method are verified by using structural monitoring data set and damage simulation data set of QCC.