Predicting odorous emissions from bitumen collected at 13 refineries: A combined GC–MS and supervised learning approach

Abstract

During asphalt paving operations, bitumen emissions occasionally give rise to unpleasant odours attributed to volatile organic compounds. While infrequent, these odours can significantly disrupt community well-being, local air quality, and workers' productivity. Predicting odours from a bitumen source before its use in the field is an ideal strategy to address these challenges proactively. This study introduces a novel Linear Discriminant Analysis (LDA) method, utilising data from headspace gas chromatography-mass spectrometry (HS-GC–MS) of bitumen samples to forecast the likelihood of odours in bituminous road binders. The LDA model, developed using HS-GC–MS results from sixteen straight-run binders of known odour status collected globally, demonstrates high accuracy in odour prediction through two cross-validation techniques. This accuracy enables the rapid identification of odorous bitumen samples using GC–MS data. Furthermore, our method suggests a substantial contribution to odour from alkanes and arenes. The proposed approach provides a simple and practical tool, offering the potential for selective use or pre-treatment of bitumen, thereby reducing the introduction of highly odorous binders into paving projects. This methodology presents an innovative step towards proactive odour management in asphalt paving, contributing to community well-being, environmental quality, and the efficiency of paving operations.