This paper proposes a Hierarchical Bayesian Network (HBN) approach to estimate the uncertainty in performance prediction of manufacturing processes by aggregating the uncertainty arising from multiple models at multiple levels. A HBN is an extension of a Bayesian network (BN) for modeling hierarchical or multi-level systems where each node may represent a lower-level BN. The BNs at different levels can be constructed either using physics-based models or available data or by a hybrid approach through a combination of physics-based models and data. An improved BN learning algorithm is presented where the topology is learnt using an existing algorithm but different parametric and non-parametric models are fit to represent the conditional probabilities. Data for model calibration may be available at multiple levels such as at the unit process level or line level or sometimes at the factory level. Using all the data for calibration can be computationally expensive; therefore, a multi-level segmented approach for model calibration is developed. The injection molding process is used to demonstrate the proposed methodologies for uncertainty prediction in its energy consumption.

This content is only available via PDF.
You do not currently have access to this content.