UnblindMix: An unsupervised reference-free framework for multi-species detection in high-temperature combustion diagnostics

High-temperature combustion processes are characterized by the rapid formation and consumption of numerous reactive species, presenting substantial challenges for accurate diagnostic measurements. Traditional approaches often require labor-intensive collection of high-temperature absorption cross-section spectra, which limits scalability and efficiency. To address these challenges, we present UnblindMix, an unsupervised, reference-free diagnostic framework that leverages blind source separation (BSS) to simultaneously infer species concentrations and reconstruct high-temperature reference spectra directly from composite mixture spectra. This study demonstrates the capability of UnblindMix in pyrolysis experiments of n-butane and iso-butane at 923 K and 1 atm, as well as in non-reactive mixtures of $C_1-C_3$ hydrocarbons in a high-temperature static cell. Using an interband cascade laser (ICL) operating over 2984.5–2989.5 cm⁻¹ range, the model accurately predicts time-resolved species profiles and reconstructs high-temperature reference spectra with minimal reliance on pre-existing spectral databases. Validation against AramcoMech 3.0 simulations and experimental data revealed good agreement. UnblindMix represents a significant advancement absorption diagnostics, reducing the dependency on extensive spectral databases and offering a scalable, efficient solution for multi-species detection in laboratory and industrial applications.

Novelty and significance

The novelty of this research lies in the introduction of UnblindMix, an autoencoder-based blind source separation model capable of inferring species concentrations and reconstructing high-temperature reference spectra solely from mixture spectra. It is significant because it eliminates the reliance on labor-intensive high-temperature absorption cross-section datasets, providing a scalable and robust framework for multi-species detection in combustion diagnostics. The model’s application to real-time pyrolysis of n-butane and iso-butane demonstrates its potential to revolutionize high-temperature spectroscopic analysis for both research and industrial applications.