In order to investigate the optimal conditions for the formation of castor oil-based polyurethane (CO-PU) films and their comprehensive performance, a series of experiments were carried out utilising biomass castor oil (CO) as the main reactant, 4,4-diphenylmethane diisocyanate (MDI) as the other main reactant, benzoyl peroxide (BPO) as the radical initiator, and dibutyltin dilaurate (DBTDL) as the catalyst. The CO-PUs with different Rvalues (the number ratio of -NCO/-OH substances) were prepared by pre-polymerisation under single-variable control. The effects of different R-values, DBTDL contents and curing temperatures on the mechanical properties and film-forming state of the CO-PUs were then investigated. Finally, the liquid-resistant medium properties and thermo-gravimetric analyses of CO-PU films were investigated. The experimental results demonstrate that when the R value is 2.5, the DBTDL content is 0.1 parts per thousand, and the curing temperature of the film-forming process is 80 degrees C, the tensile strength and elongation at break of the CO-PU film are 36.4 MPa and 49.05 % respectively, and the hardness of the pencil at this time is 6H. The water absorption rate is 0.01 %, and the contact angle of water is up to 83.5 degrees. CO-PU displays a commendable resistance to sulfuric acid, sodium hydroxide, and ethyl acetate. Additionally, CO-PU exhibits a high thermal decomposition temperature. The comprehensive performance of COPU can be enhanced by modifying the film-forming conditions of CO-PU, which paves the way for further expansion of the applications of CO-PU in the future.
针对目前中文语义提取不充分,存在特征提取不完善、丢失关键信息的问题,提出了一种融合知识增强的语义表示(enhanced representation through knowledge integration,ERNIE)、卷积神经网络(convolutional neural networks,CNNs)、双向门控循环单元(bidirectional gated recurrent unit,BiGRU)的注意力机制(integration of ERNIE and CNNs-BiGRU's attention mechanism,ECBA)模型.首先,使用ERNIE模型预训练出动态词向量;然后,在特征提取层设置不同大小的卷积核提取局部特征,再利用BiGRU模型提取全局特征,并在BiGRU模型中引入注意力机制捕获关键信息,及融合局部特征和全局特征得到最终的特征向量;最后,通过softmax函数预测类别.将模型在公开数据集上进行验证.结果表明,相较于对照模型,ECBA模型在准确率、召回率、F1 值上分别达到了 95.07%、95.01%、95.03%的较好效果.ECBA模型用于新闻分类任务中,能促进特征提取技术在文本分类上的研究.
针对综合传动装置内不同类别零部件功能和工作风险差异化,提出了一种基于风险权重联合的综合传动装置任务可靠性预计新方法.首先,根据系统任务功能利用可靠性框图法评估综合传动装置系统可靠性.其次,在给定部件故障模式发生概率等级定义及评分准则基础上,引入模糊理论建立风险权重系数计算方法.最后,用某履带车辆综合传动装置的可靠性预计进行了实例验证.实验结果表明,该方法具备一定的合理性和有效性,可为管理部门提供可靠性设计需改进的重点关注部件,为传动装置后续优化设计提供更大的改进空间,具备广泛的应用推广前景.
为了解决目前物流仓储中心选址模型存在的脱离实际、需求不合理、成本模型不完备等问题.以长三角地区为例,采用二次指数平滑法预测出该地区 27 个中心城市未来 3 年的货物需求量,构建长三角地区物流仓储中心选址模型,利用K-means聚类算法和遗传算法进行仿真实验,并与多种算法进行对比分析.仿真实验表明,K-means聚类算法在求解备选仓储中心问题上有很好的效果,可以使仓储中心到需求点的总距离和最小,遗传算法相比于其他算法在求解仓储中心选址问题上求解速度更快、迭代次数更少、最终结果更精确.结果证明,使用K-means聚类算法和遗传算法求解长三角地区物流仓储中心可以大大提高该地区物流效率并降低物流成本.
利用多尺度特征策略进行特征提取的有效性不足是多模态医学图像融合领域存在的问题.为了增加融合结果的多尺结构信息,提出了一种基于残差多尺度网络(residual multi-scale network,Res2Net)、交错稠密网络和空间通道融合算法的多模态医学图像融合算法.Res2Net的编码器在提取多尺度特征时能保留更多语义信息;交错稠密网络减少了解码器和编码器之间的语义差异,丰富了融合图像的结构和细节信息;掩码鉴别器约束了脑瘤病灶区域,进一步提高了融合图像的质量;特征图通过空间通道融合算法融合减少了多模态图像之间的信息冗余.该算法在信息熵(entropy of information,EN)、互信息(mutual information,MI)、结构相似性(structure similarity index measure,SSIM)、多尺度结构相似性(multi scale structural similarity index measure,MI_SSIM)指标上拥有较高水平的性能表现,EN提高了6%,MI提高了3%.结果显示,所提出的算法在视觉感知和指标评估上达到了较高的融合质量.
