Integrated machine learning-quantitative structure property relationship (ML-QSPR) and chemical kinetics for high throughput fuel screening toward internal combustion engine

Abstract

This work proposes a high throughput fuel screening approach to identify molecules with desired properties for internal combustion engine at the early stage of property-oriented fuel design. The virtual screening is a funnel-like approach containing Tier 1 fuel physicochemical property screening and Tier 2 chemical screening. Tier 1 screening is based on the machine learning quantitative structure property relationship (ML-QSPR) models for 15 properties of melting point, boiling point, vapor pressure, enthalpy of vaporization, cetane number, research octane number, motor octane number, ignition temperature, flash point, yield sooting index, liquid density, lower heating value, surface tension, lower/upper flammability limit. The key is to identify the target values for the selected properties for a given engine architecture and combustion strategy. Tier 2 screening inspects the ignition delay time, ϕ-sensitivity and laminar flame speed to evaluate the fuel reactivity, the potential of ϕ stratification combustion, combustion rate and dilution tolerance. Merit function provides a simple tool to assess the potential benefit of fuel-engine interaction based on the properties computed in Tier 1 and Tier 2 screenings. A case study for boosted spark ignition engine is performed to showcase the fuel screening workflow. The virtual screening can accelerate the property-oriented fuel design in a time, resource, labor, cost-effective way and identify the promising candidates for the experimental test as ultimate validation. This paradigm aims at inspiring the new ideas of data-driven fuel screening and promoting the application in the energy sector.