DOI: 10.12688/f1000research.191871.1 ISSN: 2046-1402

Multi-Chamber Electronic Nose for Biopharmaceutical Powder Discrimination: A Comparative PCA–ANN Pattern Recognition Approach

Andi Setiawan, Radi -, Nur Afni Rezkika, Siti Jualiah, Nur Imam Satria, Ilmania Syavitri, Muhammad Salam Ibnu Chaer, Muhammad Abimansyah Saragih, Aziz Ahmad Bukhori, Yosefina Mangera, Monaeka Sitinjak, Qurratul Uyun
Show PDF Cite
Background Reliable identification of herbal powders is important for quality control, but grinding reduces distinctive morphological characteristics and makes visual identification difficult. Electronic noses offer an alternative by detecting sensor response patterns associated with sample aroma. This study aimed to evaluate a multi-chamber electronic nose (e-nose) combined with principal component analysis and an artificial neural network to discriminate six herbal powders. Methods A six-chamber electronic nose system equipped with an array of 13 gas sensors was used to analyse garlic, moringa leaf, green chiretta, black pepper, ginger, and lemongrass powders. Each powder type underwent 50 replicate measurements comprising flushing, collecting, and purging stages, yielding 300 observations. Mean sensor responses were analysed using principal component analysis (PCA) for pattern exploration, while normalised responses were used for artificial neural network (ANN) classification. The ANN was trained on 240 observations and evaluated on 60 test observations. Results The first two principal components explained 88.01% of the total variance. Black pepper formed a clearly separated cluster, whereas the remaining herbal powder classes exhibited partial overlap in the PCA score plot. The ANN correctly classified 56 of 60 test observations, achieving an overall accuracy of 93.33%. Class-specific accuracies were 100% for black pepper, garlic, ginger, and lemongrass, 70% for moringa leaves, and 90% for green chiretta. Conclusions The sensor response patterns generated by the multi-chamber e-nose contained discriminative information that supported ANN-based classification of the six herbal powders under the experimental conditions. PCA provided exploratory information on class distributions, while ANN enabled classification despite partial overlap in the PCA score plot. The integrated system shows potential for herbal powder identification and quality control, with further validation needed across independent sample batches.

More from our Archive