Experimental validation of a multiple model predictive control for waste heat recovery organic Rankine cycle systems

Waste heat recovery systems are today considered as a valuable solution to increase energy efficiency of industrial applications and heavy-duty vehicles, as it uses a thermodynamic organic Rankine cycle system to recover the heat losses to produce electrical or mechanical power. Optimal performance of such machines is often achieved at conditions where complex time-varying nonlinear dynamics are encountered, making the automatic control strategy a fundamental element to maximise the energy efficiency. In this paper the development of a multiple model predictive controller suitable for industrial implementation is presented, and its effectiveness is experimentally validated for the task of maximising output power of a $11{\mathrm{kW}}_{\mathrm{el}}$ small-scale ORC power unit used in a waste heat recovery application. The main advantage of the proposed controller is the possibility to use different model structures to describe local dynamics without increasing complexity of the optimisation problem. Additionally, experimental results illustrate that the entire operating range of the system might be classified in two regions, a quasi-linear and a highly nonlinear region for 'high' and 'low' superheating degrees respectively. Closed-loops tests lead to the conclusion that a single linear model predictive controller might only be used under suboptimal operation of low power production (on the quasi-linear region for 'high' superheating), otherwise leading to poor performance or even instability. Alternatively, the proposed strategy keeps the cycle stable over the entire range of conditions and allows to increase the net electrical energy produced by at least $6\%$, even under drastic waste heat source variations, when operating closer to the minimum allowed superheating degree.