Machine Learning–Based Identification of Solvents from Post-Desiccation Patterns.
J. I. Morán-Cortés, F. Pacheco-Vázquez · 2026
Read Machine Learning–Based Identification of Solvents from Post-Desiccation Patterns. by J. I. Morán-Cortés, F. Pacheco-Vázquez (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-15660.txt.
Introduction
Morán-Cortés and Pacheco-Vázquez teach a neural network to name a solvent from dried crack patterns. Starch-liquid slurries crack as they dry, and the solvent is gone by the end. The team photographed patterns from water, ethanol, acetone, and mixed solvents. Nine shape traits were counted into frequency histograms. Networks trained on the histograms reached about 96 percent accuracy. Sets with the crack area spread worked best. The protocol may guide pattern tasks in other fields.
Catalog record
| CREATOR | J. I. Morán-Cortés, F. Pacheco-Vázquez |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 33234 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.15660
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Machine Learning–Based Identification of Solvents from Post-Desiccation Patterns. J. I. Morán-Cortés, F. Pacheco-Vázquez Abstract We introduce an optimized protocol of fracture pattern classification using an artificial neural network to identify the solvent involved in the desiccation cracking process of starch-liquid slurries, even after it has been completely evaporated. For this purpose, image analysis techniques were used to characterize patterns obtained from drying suspensions using single solvents (water, ethanol, acetone) and two-component solvents (water-ethanol mixtures at different concentrations). Frequency histograms were generated based on nine morphological features, taking into account their size, shape, geometry and orientational ordering. Subsequently, we used these histograms as input data into artificial neural network variants to determine the set of features that lead to the higher accuracy in solvent identification. We obtained an average accuracy of 96 ( ± 1 ) % 96(\pm 1)\% considering all solvents in the analysis. The highest accuracy was obtained with sets of features that include the crack area distribution. The proposed protocol can help to determine the combination of features that optimize pattern recognition in other fields of science and engineering. 2021 keywords Solvent identification, pattern recognition protocol, artificial neural networks, cracking process. 1 Introduction In different scientific fields, neural networks stand out for their effectiveness and efficiency in automatically classifying patterns based on their morphological features 29 ; 27 ; 43 ; 34 ; 35 ; 30 ; 40 ; 44 ; 15 ; 10 . Ponomarev et al. 34 ; 35 used an automatic classification method of immunofluorescent images for autoimmune disease diagnostics based on the number, size, shape, and localization of individual cells in medical tests, achieving an average accuracy of 84(\pm 6)\% and showing performance comparable to that of the texture-based method 30 . Umer et al. 40 proposed a person identification system based on multiscale morphology to extract structural and textural features from human iris images, with average accuracy of 99(\pm 1)\% . Wang et al. 44 developed a model for material property classification based on X-ray and neutron diffraction patterns of crystal structures, reaching accuracies of 66(\pm 14)\% for the perovskite family \textup{ABO}_{3} . Fan et al. 15 propose a model for classifying multi-defect wafer map patterns by combining a densities clustering method with geometric features, including form factor (area/perimeter) and eccentricity, obtaining average accuracy of up to 94\% , whereas more recently, Choi and Suh 10 achieved over 98\% accuracy for the same purpose with a model focused on the active contour. Although neural networks with statistical approaches are feasible tools for pattern recognition, a correct selection of parameters is essential to avoid overfitting and increase the accuracy and efficiency of the method. Desiccation patterns are probably the most common examples of how nature spontaneously generates configurations through complex processes that occur over multiple time and length scales. For example, at microscopic scale, slow evaporation of gold nanoparticle suspensions results in microwire patterns induced by drying liquid bridges 41 . Evaporation of blood droplets produces dendritic patterns with radial fissures in the periphery related to a high concentration of lipids 19 ; 3 . On a much larger scale, drying processes in basins and lakes generate polygonal salt patterns 21 , while the shrinkage of frozen ground (permafrost) creates ice-filled crevices and soil 4 . Furthermore, these polygonal patterns are not limited to Earth alone, but are also found on the surface of Mars 4 ; 26 ; 25 ; 28 ; 36 ; 32 and the Moon 28 . Such patterns exhibit great morphological diversity due to the influence of multiple factors during their formation; however, they can be effectively analyzed and classified using image analysis and machine learning techniques (for example, to predict the number of cracks in the soil based on the initial moisture content, layer thickness, and sample size 11 ), provided that adequate selection of parameters is made. Thin layers of colloidal suspensions have been extensively used at laboratory scale to explore desiccation processes. It has been shown that the morphology of the resulting cracking patterns depends significantly on the thickness h of the layer. Theoretically, thin films crack upon reaching a critical stress, which relaxes over a distance proportional to h 18 , while the crack spacing L follows a power scaling with exponent n=2/3 28 ; 38 , in reasonable agreement with experiments ( n=0.65\pm 0.01 for silica particles 2 ; 22 , n=0.71\pm 0.01 for cornstarch and n=0.53\pm 0.02 with clay 28 ), for which the characteristic lengths L_{1}=\sqrt{A_{\textup{c}}/N_{\textup{c}}} and L_{2}=A_{\textup{c}}/L_{\textup{T}} have been proposed, where A_{\textup{c}} is the size of the container, N_{\textup{c}} the number of cells, and L_{\textup{T}} the total length of the cracks. In addition, the average displacement of the pattern vertices increases linearly with h during repeated wetting and drying cycles 18 . Likewise, the characteristic area A of the patterns is directly proportional to h^{4/3} when using cornstarch with water, as well as calcium carbonate (\textup{CaCO}_{3}) with different solvents such as water, isopropanol and silicone oil 25 ; 28 . Differences in the fitting parameters are evidence of the influence of the solute and solvent on the size of the patterns. In order to classify fracture patterns, frequency histograms are a useful tool to represent their morphological distribution. A variety of parameters exists to characterize the geometry and topology of these patterns based on radial density profile 19 , average areas 25 ; 31 ; 39 ; 1 ; 37 , lengths 28 ; 32 ; 2 ; 22 ; 39 ; 7 , number of neighbors 32 ; 37 ; 6 , fractal dimension 13 ; 5 and lacunarity 5 , to name a few, but only some studies include a large dataset of measurements. However, histograms based on the normalized isoperimetric ratio, defined as \lambda=4\pi A/P^{2} , where A is the area and P is the perimeter of the patterns, have been proposed by Roy et al. 37 as the most efficient parameter to distinguish between fracture patterns, which they qualitatively evidenced in cracked materials of natural mud, clays (such as laponite and bentonite), corn and potato starch extracts, and polymers (such as PDMS), among others. From a statistical point of view, this should hold true even if the successive cleavage of the fracture patterns does not follow a regular or uniform process 6 . For this reason, histograms of the area - perimeter set and other morphological features may contribute to the automated identification of such patterns, but the effectiveness and efficiency of these features for classification is still unknown. In this work, we propose an optimized protocol to use neural networks for the recognition and classification of fracture patterns produced during the desiccation of cornstarch-liquid slurries involving different solvents. For this purpose, an experimental array (described in Sec. 2 ) was used to follow the evaporation process of thin layers of the slurries until the fracture patterns appear. In Sec. 3 , we describe the methodology used for the image analysis. Then, we propose a set of features represented in frequency histograms to characterize fracture patterns from their size, shape, geometry and orientational ordering. On the one hand, size is determined through the area and perimeter of the cells and the area of the cracks, all