机器学习增强的灰色盒软传感器用于在聚合物挤出工艺中的融度预测
Yasith S Perera1, Jie Li2, Chamil Abeykoon3
1Northwest Composites Centre and Aerospace Research Institute, Department of Materials, Faculty of Science and Engineering, The University of Manchester, Oxford Road, Manchester, M13 9PL, UK.
Scientific reports
|February 15, 2025
概括
这项研究引入了一种新的软传感器,用于在聚合物挤出中实时预测融粘度. 它将基于物理的模型与深度学习相结合,显著提高了比传统方法更准确的准确性.
科学领域:
- 聚合物科学与工程 聚合物科学与工程
- 过程控制和自动化过程控制和自动化
- 机器学习应用 机器学习应用
背景情况:
- 融粘度是聚合物挤出中的关键质量指标.
- 由于流量干扰和测量延迟,现有的风力计在实时监测方面存在局限性.
- 软传感器为监测难以测量的物理参数提供了可行的替代方案.
研究的目的:
- 开发一个实时软传感器,用于预测聚合物挤出中的融粘度.
- 将基于物理的知识与机器学习相结合,以提高预测准确度.
- 在动态工艺条件下克服传统风力计的局限性.
主要方法:
- 实施了一种灰色盒软传感方法,将基于物理的数学模型与深度神经网络集成在一起.
- 基于物理学的模型提供了初始融粘度预测.
- 一个深层神经网络可以弥补基于物理模型的预测错误.
主要成果:
- 拟议的软传感器实现了2.2 x 10−3 (0.22%) 的正常化根平均平方误差.
- 它的性能比完全数据驱动的模型 (MLP,LSTM) 和之前基于RBFNN的软传感器大约95%.
- 软传感器有效监测由于操作条件变化的粘度变化.
结论:
- 开发的灰盒软传感器为聚合物挤出提供了准确的实时融粘度预测.
- 这种方法增强了过程监控和控制能力.
- 该系统有效地检测来自操作转移的粘度变化,但不能检测材料性质变化.
相关概念视频
Classification and Mechanical Properties of Synthetic Polymers
22
Synthetic polymers are classified as elastomers, fibers, or plastics based on their crystallinity. Crystallinity, the degree of long-range order in the solid state, influences the mechanical properties (stretching or contracting) of elastomers. Elastomers are flexible polymers that can expand or contract easily upon the application of an external force. They have numerous crosslinks that pull them back into their original shape when stress is removed. Silicones, for instance, are highly elastic...
22
Determination of Molar Masses of Polymers I
11
Polymerization produces macromolecules with a range of chain lengths due to the random nature of molecular growth processes. As chains form and terminate at different stages, a single polymer sample contains molecules of varying sizes rather than a uniform structure. This variability is described using average molar masses and distribution-related parameters, which together provide a comprehensive understanding of polymer characteristics.The distribution of molar masses plays a critical role in...
11
Determination of Molar Masses of Polymers II
4
Polymer samples typically consist of macromolecular chains with a distribution of lengths, resulting in a range of molar masses rather than a single discrete value. Conventional descriptors such as the number-average molar mass and weight-average molar mass quantify this distribution but do not fully capture polymer behavior in solution..The viscosity-average molar mass provides a more realistic description of polymer behavior in solution because it accounts for the enhanced contribution of...
4


