Anal Methods. 2025 Oct 13. doi: 10.1039/d5ay01251a. Online ahead of print.
ABSTRACT
This study presents a low-cost, 3D-printed portable device that integrates computer vision and an artificial neural network (ANN) to quantify biodiesel content (1-30% v/v) in fossil diesel blends using a solvatochromic assay with Reichardt’s dye. A total of 105 samples (35 biodiesel blend levels in triplicate) were analyzed with both the proposed method and the official Brazilian standard ABNT NBR 15568/2008 (FT-IR). While the standard method requires laboratory infrastructure and specialized equipment, the proposed system provides comparable accuracy directly at the point of fuel distribution. It achieved a mean absolute error (MAE) of 1.5% (R2 = 0.969) for training data and 0.5% (R2 = 0.995) for independent test data. Robust cross-validation confirmed model stability and absence of overfitting, and a paired Student’s t-test showed no statistically significant difference between the two methods (p > 0.05), confirming statistical equivalence. Beyond analytical performance, the device offers practical advantages: controlled lighting and webcam-based image acquisition coupled with ANN processing enable rapid, on-site biodiesel determination without specialized training. This contrasts with the official method, which is restricted to laboratory settings. By combining portability, low operational complexity, and real-time analysis capability, this system represents a significant advancement for fuel quality monitoring, allowing reliable control of BX diesel blends directly at fueling stations and other non-laboratory environments. BX diesel, dye solution, and ethanol.
PMID:41078134 | DOI:10.1039/d5ay01251a
