Investigating the relationships between process variables and resource consumption and product quality in a production line is critical to optimizing the processes and improving manufacturing, energy, and material efficiency as well as product quality. As a production line is composed of multiple processes that are interrelated, modeling of their interdependency towards efficiency and quality improvement can be formulated as a high-dimensional, non-linear system modeling problem. In addition, some manufacturing processes are well understood and processes can be optimized with existing physical laws, whereas others may not be completely the case, due to the complexity involved. This research will investigate recent advancement in data science (e.g., Bayesian inference and deep learning) and develop manufacturing-specific machine learning algorithms for better characterization of process dynamics, towards efficiency and quality improvement.
The modeling of an entire production line is partitioned into the unit process level and system level. Upon the modeling at the unit process level, multiple processes in a production line can be modeled as a Markov process, and the resource consumption of a latter process is dependent on its precedent process.
Modeling of the unit process is based on hybrid machine learning, integration of deep learning with stochastic modeling. We are developing a novel machine learning architecture, manufacturing domain knowledge-guided machine learning, to improve the trustworthiness and robustness of machine learning in manufacturing applications.