AI-assisted Advanced Propellant Development for Electric Propulsion

TL;DR

The paper addresses the challenge of finding viable xenon substitutes for electric propulsion by predicting ionisation properties and fragmentation of candidate molecules. It introduces a pipeline that uses Morgan fingerprints (ECFP, 4096-bit, radius 2) as inputs to neural networks (MLP for IE, AE, and ion mass; LSTM/Bi-LSTM for mass spectra) trained on open NIST WebBook data, with a 90/5/5 data split and evaluation via mean relative error and spectral similarity metrics. Key results include IE with $6.87\%$ MRE, AE with $7.99\%$ MRE, ion mass from AE fragmentation with $23.89\%$ MRE, and mass spectra predictions achieving a cosine similarity of $0.6395$ and recall@10 of $60.63\%$ (improving to $78\%$ with a ±30 Da mass filter). The study demonstrates a feasible ML-driven framework to screen propellant candidates and observes limitations from open-access data (mass bias toward $50$–$200$ Da) and fixed EI energy (70 eV), suggesting future gains from graph-based molecular representations and multi-energy fragmentation databases to enhance screening for Xe substitutes.

Abstract

Artificial Intelligence algorithms are introduced in this work as a tool to predict the performance of new chemical compounds as alternative propellants for electric propulsion, focusing on predicting their ionisation characteristics and fragmentation patterns. The chemical properties and structure of the compounds are encoded using a chemical fingerprint, and the training datasets are extracted from the NIST WebBook. The AI-predicted ionisation energy and minimum appearance energy have a mean relative error of 6.87% and 7.99%, respectively, and a predicted ion mass with a 23.89% relative error. In the cases of full mass spectra due to electron ionisation, the predictions have a cosine similarity of 0.6395 and align with the top 10 most similar mass spectra in 78% of instances within a 30 Da range.

Figures (10)

• Figure 1: Distribution of molecular weights of chemical compounds in the three databases. The histograms indicate the percentage of compounds in specified ranges of molecular weight.
• Figure 2: Distribution of the prediction parameters in the datasets.
• Figure 3: Overview of the library matching. Modified from Wei2019.
• Figure 4: Performance from the IE prediction model with the configurations from Table \ref{['tab:hyper']} using the training and validation sets.
• Figure 5: Performance from the minimum AE prediction model with the configurations from Table \ref{['tab:hyper']}.