Adaptive Model Predictive Control for Heat Pipe–Cooled Microreactors Under Normal and Heat Pipe Failure Conditions

Microreactors, a specific class of nuclear reactor, feature a thermal power output of <20 MW, with intended use cases ranging from power production for remote localities and industrial facilities, to military applications, to disaster relief. Because the remote locations of these reactors make repairs difficult, and with continuous power production being essential for the intended use cases, the control system for microreactors should be able to operate or safely shut down the reactor under abnormal conditions (e.g. cases of component failure). The nuclear industry is currently pursuing various microreactor designs, one of which is the heat pipe (HP)–cooled microreactor. A potential failure mechanism in this type of microreactor is individual HP failure. The present work explores the notion that even if a single HP fails, an HP-cooled microreactor may still be controllable in its degraded state. A framework is presented for the stable control of an HP-cooled microreactor system’s thermal output power and temperature regulation under both normal and HP failure conditions, using adaptive model predictive control (A-MPC). A-MPC was implemented for its ability to maintain optimal controller performance under changing plant state and system constraints. The complex, nonlinear physical phenomena present in an HP-cooled microreactor make using a physics-based model as the A-MPC controller’s internal predictor impractical. Thus, a data-based surrogate predictor model was developed for use under both normal and HP failure conditions. The subject under study is a 37-HP system intended to simulate the HP and core thermal behavior of an HP-cooled microreactor. This system was modeled and simulated in DireWolf, a Multiphysics Object-Oriented Simulation Environment (MOOSE)–based application designed to simulate HP-cooled microreactors. The resulting model was used to generate training data for the data-based predictor model and served as the plant simulator when coupled with the A-MPC controller. This paper presents the data-based predictor model of the 37-HP system, the A-MPC controller architecture that proved suitable under both normal and HP failure microreactor conditions, and the performance of the controller when coupled with the DireWolf simulation of the 37-HP system under both normal and HP failure conditions.