Automatic Tuning of Level Controllers in a Flotation Bank using Bayesian Optimisation

Abstract

A flotation bank consisting of 6 cells in series under Single-Input-Single-Output (SISO) Proportional Integral (PI) level control is automatically tuned using Bayesian Optimisation (BO). Open loop step tests from the valve position to the level are used to identify first-order plus time-delay (FOPTD) models for each flotation cell. The PI controller settings are tuned according to the Skogestad Internal Model Control (SIMC) tuning rules. Stability bounds derived from µ-analysis are defined using these SIMC settings. As the optimum achieved by the Bayesian optimiser is largely dependent on the parameter space provided to the tuning algorithm, this space is selected first to ensure stability and secondly for performance. The BO framework is able to tune each of the six SISO PI controllers to provide significantly improved level control over the original SIMC controllers with regards to different forms of the integrated error when the plant is subjected to step changes in the level setpoints and disturbances to the feed flow. This improvement comes at the cost of an increased number of tests to conduct.