Error Analysis And Precision Prediction Of The Whole Construction Process Of Section Structure

Flat section is a typical section of an offshore platform,and its construction accuracy is very important to the overall accuracy of the platform.The construction process of flat section is generally composed of cutting,bending and welding,and the overall construction precision is mainly determined by the precision of each procedure.Each process of cutting,bending and welding will produce out-of-plane deformation,and this deformation will accumulate as the process progresses.Therefore,it is necessary to predict the overall accuracy of the section structure according to the cutting error,bending error and welding inherent deformation,and verify whether the overall deformation will exceed its maximum allowable value,so as to reduce rework.Thermoelastoplastic finite element method and elastic finite element method are used to predict the overall deformation of multi-step structures.However,the former is not applicable to the prediction of the overall out-of-plane deformation of large section structures because it consumes a lot of computational resources and the calculation speed is slow.In this paper,the method of elastic finite element calculation combined with artificial neural network tool is used to study the out-of-plane deformation of a typical offshore platform cabin structure under the whole construction process of plane section.Specifically,by taking 36 groups of source errors into the elastic finite element calculation,the corresponding global out-of-plane deformation was obtained,and the artificial neural network training database was established.Based on the above database,the artificial neural network mapping model of source error and overall out-ofplane deformation is established.The prediction results after optimization of genetic algorithm show that the artificial neural network is feasible in predicting the overall out-ofplane deformation of offshore platform structure.The main conclusions of this paper are as follows:(1)The results of elastic finite element calculation are consistent with the deformation trend generated in the actual production,and the deformation of the bottom plate of the plane section stiffened plate on the offshore platform is smaller than that of the top plate without stiffened plate,which indicates that the stiffened plate has an inhibitory effect on the overall out-of-plane deformation of the section structure.The cutting error and bending error values were directly input into the WELD.DAT welding file for finite element calculation.It was found by point measurement that the closing effect of the calculation program on the opening with large error was very limited,and the error values entered in the welding file were almost entirely presented in the resulting model in the form of gaps.(2)The gap caused by cutting error and bending error can be effectively closed by surfacing,so the cutting error and bending error can be taken into account in the prediction of out of plane deformation.The elastic finite element calculation shows that the increase of out of plane deformation is mainly concentrated in the weld with gap.(3)After considering the influence of cutting error and bending error,the out-of-plane deformation of the offshore platform plane segment structure in all directions increases to a certain extent,among which the Z-direction out-of-plane deformation increases most obviously,with an increase of up to 33.4%,and the average displacement change range of some nodes is as high as 49.1% or more.It can be seen that the cutting error and bending error have a great influence on the overall out-of-plane deformation of the segmented structure,so that the cutting and bending error must be considered while predicting the overall deformation of the structure.(4)The results show that the BP artificial neural network is feasible in predicting the whole process of out-of-plane deformation of large welded structures,but its prediction accuracy and stability need to be improved.(5)After the initial weight and threshold of BP neural network were optimized by genetic algorithm,the maximum relative error between the predicted value and the true value was reduced from 16% to 2.5%,and the determination coefficient R^2 was increased from unstable(0.911,0.892,0.652)to about 0.99(0.984,0.991,0.999).It is enough to show that genetic algorithm plays a great role in optimizing the prediction accuracy and stability of BP artificial neural network model.The optimized BP artificial neural network can quickly and accurately predict the deformation values of subplane structures of offshore platforms.