The ability to tailor the mechanical properties of synthetic fibers for specific applications has driven significant innovation in materials science. While the market pricing of established fibers is generally understood, estimating the economic worth of new or artificial fibers is a difficult task. Although external pressures such as supply limitations, infrastructural development, and overall market needs can alter the eventual selling price, the fundamental worth of a fiber, especially a novel one, originates from its operational qualities. These qualities are directly observable in its key structural attributes: tensile strength, elongation, and Young's modulus. These properties dictate how a fiber will perform under various conditions, directly influencing its suitability for different applications. For instance, a fiber with high tensile strength and modulus, for example, is inherently more valuable because it can withstand greater stress and maintain its shape under load, making it suitable for demanding applications like aerospace, high-performance composites, or structural materials [1]. In contrast, a fiber with diminished strength and modulus will be restricted to less demanding uses, thereby limiting its potential market value [2].
This study addresses the challenge of predicting the market price of synthetic fibers based on their fundamental mechanical properties: tensile strength, elongation, and Young's modulus. Given a novel fiber 'f' with these known properties, our objective is to estimate its market value. To achieve this, we developed a methodology that leverages a dataset of 121 fibers, each represented by a 4D elliptical range encompassing the mechanical properties and price. The mechanical properties and predicted price are in ranges – tensile strength 0.19 – 810 (ksi), Young’s modulus 2.76528e-05 – 65 (106psi), elongation 0.3 – 1800 (% strain) and price $0.60 - $137.
We then deduce the fiber that most closely resembles f, using a distance measure. We extract data points from this closest ellipse using multinormal and beta distributions, creating diverse training scenarios. We train Machine learning (ML) models using Neural Network, Decision Tree and Multinormal distribution, to estimate the price of f. Furthermore, we automate this process by developing a lookup table of ML models for all 121 fibers, in all three approaches. Given a fiber, and a distance measure, the lookup table is used to estimate the price of f.
A critical aspect of our approach is the selection of the distance measure. We explored two distance measures: (1) minimizing the distance to the ellipse's center and (2) minimizing the perpendicular distance to the ellipse's surface. Approach 1 is numerically cheap to compute but may misrepresent the closest resembling ellipse if a bigger ellipse is almost near f but is away from its center. Approach 2 is numerical expensive as it involves solving for an optimization routine to find the closest ellipse but gives a more realistic estimate of the notion of the closest resembling fiber.
We used the above models on mechanical properties and protein concentrations data of diverse fibers from an external data source of Koeppel and Holland [3]. The data comprises of spider silk fibers and our model identified the most relevant existing fiber matches for given mechanical properties and respective price prediction based on above trained model. Our method provides a novel approach to estimate both the price, and the most relevant closest existing fiber to tally with, using a combination of machine learning and geometrical arguments that account for the inherent variability in property-price relationships. Given the dataset, while diverse, is limited in size (121 fibers), which could impact the generalizability of the models. For future work we will expand our dataset by adding experimental results and incorporate economic factors and market dynamics into the model to improve price prediction accuracy for novel synthetic fibers.
References
[1] J. Denk, X. Liao, W. Knolle, et al., “Novel multifibrillar carbon and oxidation-stable carbon/ceramic hybrid fibers consisting of thousands of individual nanofibers with high tensile strength,” Sci Rep 14, 18143 (2024). https://doi.org/10.1038/s41598-024-68794-w
[2] M. Kurpińska, M. Pawelska-Mazur, Y. Gu, et al., “The impact of natural fibers’ characteristics on mechanical properties of the cement composites,” Sci Rep 12, 20565 (2022). https://doi.org/10.1038/s41598-022-25085-6
[3] A. Koeppel, C. Holland, Progress and Trends in Artificial Silk Spinning: A Systematic Review. ACS Biomater. Sci. Eng. 3, 3, 349–368 (2017).