Study Of Spin Wheel Drawing And Rolling Process Based On Numerical Simulation And Machine Learning

Spin wheel drawing and rolling process is an advanced manufacturing technology that integrates various process features such as extrusion,drawing,forging,rolling and rolling without cutting,which is an important method for processing high precision and high quality thin-walled tubes,and has been widely used in military,aviation,weapon,medical and other precision processing fields.However,during the processing of barrel-shaped parts by spin wheel drawing and rolling process,the difference of process parameters will affect the dimensional accuracy and forming quality of the finished tubes.Therefore,this paper explores and analyzes the spin wheel drawing and rolling forming process from finite element simulation analysis,regression prediction model,and modification of spin rolling equipment.Main research contents and conclusions:(1)The simulation model of Q235 steel pipe billet spin wheel drawing and rolling was established on the basis of metal elasto-plastic finite element by ABAQUS finite element numerical simulation software and finite element simulation was carried out.(2)The stress changes during the spin-wheel drawing and rolling process were analyzed,as well as the influence law of different process parameters on the surface residual stress and wall thickness of the formed parts after spin-drawing.The results show that the axial,radial and circumferential residual stresses of the formed parts after spin-drawing increase with the increase of the pressing amount,spindle speed and feed ratio.The wall thickness of the formed parts after spindrawing becomes smaller with the increase of the press-down amount;the wall thickness of the formed parts after spin-drawing becomes larger with the increase of the spindle speed;the wall thickness of the formed parts after spin-drawing increases slightly with the increase of the feed rate.(3)Several prediction models(BP neural network,RBF neural network,support vector machine regression)are explained in detail and algorithms are derived,and the particle swarm optimization algorithm is used to optimize the two main parameters of the RBF neural network,using the amount of press down,spindle speed and feed speed as input samples and the simulated value of wall thickness of the formed part after spin-drawing as output samples to train and prediction.The two prediction models are optimally weighted to obtain the final combined prediction model by combining the advantages of the PSO-RBF neural network prediction model with its strong nonlinear fitting ability and the support vector machine regression for small sample data.Four model evaluation criteria were used to evaluate the prediction effectiveness of these prediction models.The results show that the combined prediction model PSO-RBF-SVR has a strong nonlinear fitting ability and has the best prediction effect on the wall thickness of rolled pipes.(4)A design scheme of the average wall thickness measurement system of pipe billet based on laser displacement sensor is proposed,the design of the measurement installation device and the pressure down adjustment device is completed,and the sensor selection and control schematic are introduced.