Bayesian Optimization and Manufacturing Processes

Bayesian optimization (BO) is a powerful method for optimizing complex manufacturing processes, often considered "black boxes." However, its effectiveness decreases in high-dimensional multi-stage systems where intermediate outputs are available. Standard BO models ignore these intermediate observations and the underlying process structure.

POGPN-JPSS: A New Approach

The POGPN-JPSS framework combines Partially Observable Gaussian Process Networks (POGPN) with Joint Parameter and State-Space (JPSS) modeling to leverage information extracted from intermediate outputs. Expert knowledge is used to extract low-dimensional latent features from high-dimensional data.

Results and Benefits

The results obtained on a simulation of a bioethanol production process demonstrate that POGPN-JPSS significantly outperforms existing methods, achieving the desired performance in half the time and with greater reliability. This faster optimization translates into significant savings in time and resources, highlighting the importance of combining expert knowledge with structured probabilistic models for rapid process maturation.