基于反向步長遞減算法鋁厚板預(yù)拉伸夾持區(qū)的預(yù)測方法

基于反向步長遞減算法鋁厚板預(yù)拉伸夾持區(qū)的預(yù)測方法Holding area prediction of aluminum alloy thick plate for pre-stretching processes based on diminishing step algorithm with opposite direction

預(yù)拉伸是消除鋁合金板材內(nèi)淬火殘余應(yīng)力的主要方法，但拉伸機(jī)夾鉗對板材兩端的夾持不僅影響著板材端部殘余應(yīng)力的消除效果，而且還涉及到拉伸后板材的鋸切量、成材率等問題。因此，通過研究鋁合金板材預(yù)拉伸本構(gòu)模型、邊界條件、失效準(zhǔn)則等關(guān)鍵技術(shù)，建立了極限下壓量和滑移因子的有限元分析方法。其次，通過構(gòu)造初始夾持長度的計算函數(shù)，以一定步長正向從初始夾持長度繼續(xù)取值，根據(jù)當(dāng)前值與上一次取值之間滑移因子的差異，確定下一次取值的步長及其方向；若滑移因子相同則以相同步長繼續(xù)正向取值，否則以遞減的步長反向取值，直至步長的絕對值在閾值范圍之內(nèi)，構(gòu)建最小夾持長度反向步長遞減的確定算法，以此獲得板材厚度、伸長率為輸入的神經(jīng)網(wǎng)絡(luò)訓(xùn)練樣本。借助神經(jīng)網(wǎng)絡(luò)的非線性映射能力，通過有限組的訓(xùn)練樣本，構(gòu)建了最小夾持長度的神經(jīng)網(wǎng)絡(luò)預(yù)測模型。將預(yù)測值與相應(yīng)的有限元仿真值進(jìn)行比較，結(jié)果表明預(yù)測誤差在5%以內(nèi)，進(jìn)一步驗證建立的工件變形預(yù)測模型具有合理性。

Pre-stretching is the main method to eliminate the residual stress in aluminum alloy plate. But the holding of stretching machine clampers on the two ends of the plate can affect the cutting volume and yield of prestretched plate, as well as the elimination of residual stresses in end of plate. Therefore, by investigating the crucial technologies for the pre-stretching of aluminum alloy plate, including the constitutive model, boundary condition, failure criterion, as so on, the finite element analysis method is established for the limited pressure and slip factor. And then, while the function of initial holding length is constructed properly, the sum of the current holding length with a certain step can be taken as the next holding length. Thus, when the slip factors between the current holding length and the last one are calculated, the difference of the two can be judged. If the two is the same, the next holding length increases with the same step along the same direction. Otherwise, the next holding length decreases with the diminishing step along the opposite direction. The difference of the slip factor of current holding length with last holding length is iteratively validated until the absolute value of the step is not more than the given threshold value. Thus, the diminishing step algorithm with opposite direction of the minimum holding length can be suggested to obtain the neural network training samples with the thickness of the plate and the stretch rate as input. With the nonlinear mapping of neural network, a neural network prediction model of the minimum holding length is established by the finite groups of training samples. By comparing the predicted value with the corresponding finite element simulation, it shows that the prediction error is less than 5%. Obviously, the proposed diminishing step algorithm with opposite direction is reasonable to holding area prediction of aluminum alloy thick plate for pre-stretching processes.

全文下載：https://pan.baidu.com/s/1o8iMLEa