Waste heat recovery system for marine engines optimized through a preference learning rank function embedded into a Bayesian optimizer

Abstract

Waste heat recovery is a proven process to improve efficiency on engines and meets current necessities of the maritime industry. Since January 1, 2023, already built vessels must meet the energy efficiency indicators known as EEXI and CII. Aiming to reduce fuel consumption and mitigate pollution emissions, a novel waste heat recovery system composed of steam Rankine cycle, organic Rankine cycle, thermoelectric harvesters and desalination is presented. High, medium and low-grade waste heat from exhaust gas, jacket water, lubricating oil and engine block radiation are targeted for recovery. Performance assessment of each subsystem when implemented on a real case study 6-cylinder medium speed marine engine is analyzed. The equivalent electricity production concept was used for the assessment of the desalination subsystem. The proposed system effectively recovers waste energy, offering economic benefits, reducing pollution and satisfying the daily demand of fresh water. Also, optimal states of the waste heat recovery are provided via Bayesian optimization, which requires an evaluation function for the system to be optimized. However, this function is not available and cannot be straightforwardly established, since the quality of waste heat recovery depends on some indicators with a trade-off among them. Hence, a preference learning procedure that exploits expert knowledge is proposed to induce a function of this kind from those indicators in order to be embedded into the Bayesian optimization procedure afterward. Applied to the case study engine, a fuel consumption reduction of 15.04% is achieved. Fuel savings lead to an improvement in energy efficiency indicators, achieving a reduction of 6.98% on the EEXI and a 13.85% on the CII.