Machine learning for additive manufacturing of electronics
Machine learning for additive manufacturing of electronics
复制标题
用于电子增材制造的机器学习
DOI：
--
复制
发表时间：
2017
期刊：
影响因子：
--
通讯作者：
C. Bailey
中科院分区：
文献类型：
--
作者：
S. Stoyanov;C. Bailey
采用增材制造（AM）技术（如3D喷墨打印）制造的电子产品的质量可以通过采用主动预测模型进行工艺条件监测而不是使用传统的制造后评估技术来保证。本文详细介绍了一种基于模型的方法和相关的机器学习算法，可用于在生产运行期间实现和保持最佳产品质量，并实现模型预测过程控制（MPC）。基于电子制造3D喷墨打印动态行为的状态空间建模的数据驱动的统计学是新的，并使其成为一个原始的贡献。电子电路导线的3D打印是一个主要的目标应用，用于演示和验证从测量的过程数据开发的机器学习模型的性能。结果表明，对于3D打印过程的适度非线性动态，即使在较大的预测范围内，状态空间模型也可以提供预期的过程趋势（状态）和相关的产品质量特征。这些模型还可以支持实现模型预测过程控制，以实现最佳目标性能。
Quality of electronic products fabricated with additive manufacturing (AM) techniques such as 3D inkjet printing can be assured by adopting pro-active predictive models for process condition monitoring instead of using conventional post-manufacture assessment techniques. This paper details a model-based approach, and associated machine learning algorithms, which can be used to achieve and maintain optimal product quality during production runs and to realise model predictive process control (MPC). The investigated data-driven prognostics based on state-space modelling of the dynamic behaviour of 3D inkjet printing for electronics manufacturing is new and makes it an original contribution. 3D printing of conductive lines for electronic circuits is a main targeted application, and is used to demonstrate and validate the prognostics capability of machine learning models developed from measured process data. The results show that, for moderately non-linear dynamics of the 3D-Printing process, state-space models can inform on the expected process trends (states) and related product quality characteristics even over large prediction horizons. The models can also support the realisation of model predictive process control for optimal target performance.