Analysis Based On The Pressure Curve Tons Hydraulic Press Forging Process Quality Control System

With the development of aerospace, an advanced requirement is raised for the quality of large-scale high-strength aluminium alloy structural parts. It is always hoped to obtain the high accuracy dimension and better texture organization of structural parts by forging, so as to ensure the mechanicalness and application life of the shaped parts. But the randomized interference factors in the forging process of ten-thousand-ton multicored hydraulic die press, such as different manual operation, the undulate hydraulic pressure and so on, caused the large fluctuation in the internal structure of forgings and mass instability which hard to manufacture the higher quality products. The quality control system in the forging process based on the prediction in remaining pressurizing time was studied and applied aiming at those problems above.The relationship between the pressure curve and the key quality index has been established by analyzing the effect of die forging process on drop-forging, which provides theoretical basis for the analysis of hydraulic pressure curves. In order to resolve the problem of preferring hydraulic pressure curves, the stochastic approximation algorithm with randomly varying truncation has been studied. It makes the algorithm suitable for our problem to apply the concept of variable force applying work. And an operable method that applies stochastic approximation algorithm with randomly varying truncation to prefer the regular hydraulic press curve has been designed. The tendency item of hydraulic pressure curve is separated form the random item by exponential smoothing, so that the extraction of eigenvalues becomes easy. The neural network predictive theory was researched deeply, and the quality control system based on BP neural network prediction model has been realized according to the characteristic of die forging process and the structure of the hydraulic press computer control system. The Levenberg-Marquardt algorithm was employed to improve the BP neural network in order to resolve the problem that BP's low rapidity of convergence and trapping in local minimum area easily. The prediction model of remaining pressurizing time was established, which is used to direct the forging operation or to modify the preset forging time in order to uprate drop-forging quality by controlling the forging time.The emulation experiment indicated that the system designed in this thesis is effective and practicable, and possesses the merits of that high rapidity of convergence and high accuracy which can satisfies the requirement of die-forging. In practicable production, when the hydraulic pressure was up to the maximum, the predictive absolute error was less than 0.04s which transferred to drop-forging dimension error is 0.12mm.