Predicting Engineering Properties in Injection Molded Parts by Neural Network Computation

该项目将开发一种方法来预测工程 注塑成型部件的性能。 这一方法将 结合联结主义学习系统的应用， 神经网络，对非牛顿粘弹性的处理 聚合物材料。 所需的工程性能， 注塑成型部件强烈依赖于 热机械历史，这反过来又极大地影响着 加工参数。 部分之间的关系 性能和热机械历史是复杂的， 线性，因此，拟制定的方法将基于 一种反向传播算法，它提供了训练的手段， 具有自适应非线性单元的网络。 网络将是 受过简单斑块的热机械历史训练，从 注射成型模拟程序的结果， 相应的工程性质。 该网络涉及 热机械历史对零件性能的影响 在训练中发展的加权连接。 因此当 来自模拟结果的热机械历史已知为 任何部件内的每个位置，相应的属性可以 被预测。 这种基于神经网络计算的方法 将为零件设计人员提供预测工程的能力， 注塑成型部件的性能，无需模塑和测试 实际组件。

The project will develop a methodology to predict engineering properties in injection molded parts. The methodology will incorporate the application of connectionist learning systems, neural networks, to the processing of non-newtonian viscoelastic polymeric materials. The desired engineering properties in injection molded parts are strongly dependent upon the thermomechanical history which, in turn, is greatly influenced by the processing parameters. The relationships between part properties and thermomechanical history are complex and highly non- linear, therefore the methodology to be developed will be based on a back propagation algorithm that provides the means for training networks with adaptive non-linear units. The networks will be trained with the thermomechanical history of a simple plaque, from the results of injection molding simulation programs, and the corresponding engineering properties. The network relates thermomechanical history to the part properties through the weighted connections developed during training. Thus, when thermomechanical histories from simulation results are known for every location within any part, the corresponding properties can be predicted. This methodology which is based on neural network computation will provide part designers with the ability to predict engineering properties in injection molded parts without molding and testing actual components.