8 July 2024 in Nature Energy
Clean transport requires tailored energy carriers. For heavy-duty transportation, synthetic fuels are promising but must fulfil the key challenges of achieving carbon neutrality while reducing air pollution and ensuring scalability through compatibility with existing infrastructure. Here we show that hydroformylated Fischer–Tropsch (HyFiT) fuels composed of optimized alkane–alcohol blends simultaneously address these challenges. First, the design of the HyFiT fuel process flexibly closes the carbon cycle by employing biomass or carbon dioxide as feedstock, while being scalable through mature technologies. Second, fuel testing shows that HyFiT fuels comply with global fuel standards. Material compatibility is demonstrated for two standard sealing materials, enabling the retrofit of today’s vehicle fleets. Third, vehicle testing shows that HyFiT fuels substantially reduce combustion-induced particulate matter and nitrogen oxides. Fourth, a well-to-wheel life cycle assessment finds that HyFiT fuels enable the transition to net-zero greenhouse gas emissions, showing simultaneously a favourable profile in other environmental parameters. HyFiT fuels can thus complement electrification for heavy-duty transportation.
14 May 2026 in Digital Chemical Engineering
20 July 2025 in International Journal of Hydrogen Energy
30 October 2026 in Computers & Chemical Engineering
21 August 2026 in SSRN Electronic Journal
Recently, technological advancements have enabled various automatons in chemical engineering, from process simulation and optimization to production scheduling. Despite these advances, however, fault detection remains a largely manual task due to process safety concerns.In this paper, we conduct a feasibility study on the use of symbolic machine learning for fault detection and diagnosis. In particular, we use a state-of-the-art framework, leading to interpretable logic rules that can be used to detect faults. Furthermore, we define two methods for identifying a minimal, cost-effective set of process variables from which fault detection rules can be built. Our first method involves running the framework for different sets of variables, comparing the accuracy of the learned rules with their associated cost; the second method directly embeds the cost consideration as a parametrized optimization constraint.The study is conducted on a simulated ethylene oxidation process, focusing on ten faults. Experimental results show that our fault detection approach is comparable to statistical machine learning methods such as random forest in terms of accuracy, while providing much better interpretability. Furthermore, we found that our cost optimization methods enable identifying which process variables can be ignored without incurring a loss in fault detection accuracy, while identifying low-cost solutions with a moderate impact on fault detection accuracy. We finish by outlining a plan to overcome the main limitations of the approach, namely, scalability and the assumptions made to keep the problem tractable, and discuss a methodology for applying a future iteration of the framework to a real process.