Abstract. In this paper, the forming process research of the front opening and closing mechanism hatch for multiple units was carried out, and the industrial grade 5083 aluminum alloy was selected as the original material. The uniform rapid superplastic forming process combined with hot stamping and direct-reverse superplastic forming was used to form the hatch body skin, the cold stamping and argon arc welding process were used to produce the connection support, and the hatch integral part was obtained by argon arc welding finally, which solved the problem of poor environmental protection and high ratio of traditional FRP hatch and high precision forming of the large complex thin-walled structural part that cannot be achieved by the traditional forming process. After process research and development, finite element analysis, and forming tests, the aluminum alloy hatch with good forming quality was successfully manufactured. The ultimate thinning rate of the part was 23.8%, the overall wall thickness was evenly distributed, the deviation of shape was controlled within 1 mm, and the mechanical properties met the relevant technical requirements.

超塑气胀成形工艺是复杂薄壁零件成形的主要技术.以5083铝合金为研究对象,利用Marc有限元软件对动车车窗的超塑气胀成形工艺进行研究,选择合适的正反胀时间和压力,设计了3套不同的反胀模具(底面左上端圆角大小不同以及左端是否有凸槽),依次进行热冲压、反胀及正胀有限元分析,最终得出正确的超塑气胀成形模具.结果表明:反胀过程中储料的多少对成形件有显著影响;其中两种成形件的右边储料过多,刚性太大,正胀时难以变形,出现褶皱缺陷,而另外一种成形件由于模具左端设有凸槽减少了储料,刚性也减小,故易于变形且无明显缺陷;成形件最小壁厚为2.496 mm,位于车窗内边缘棱边处,最大减薄率为37.6％,与模具贴合良好且质量稳定,整体壁厚分布均匀.

Due to countless orthogonal eigenstates, light beams with orbital angular momentum(OAM) have a large potential information capacity. Recently, deep learning has been extensively applied in recognition of OAM mode. However, previous deep learning methods require a constant distance between laser and receiver. The accuracy will drop quickly if the distance of testing set deviates from the training set. Previous deep learning methods also have difficulty distinguishing OAM modes with positive and negative topological charges. In order to further exploit the huge potential of the countless dimension of state space, we proposed multidimensional information assisted deep learning flexible recognition (MIADLFR) method to make use of both intensity and angular spectrum information for the first time to achieve recognition of OA Mmodes unlimited by the sign of TC and distance with high accuracy. Also, MIADLFR can reduce the computational complexity significantly and requires much smaller training set.