R. Geppert, S. Bajaj, T. Gupta, G. Menzl
Refiberd,
United States
Keywords: textile testing, AI, sorting
Summary:
The textile industry currently relies upon calculations that can not be validated. Material costing information or appropriate duties are assigned based on supplier-provided composition data, yet as a result of opacities in the textiles value chain, intentional and unintentional production material substitutions, and increasingly complex blends, nearly 40%+ of material content labels are inaccurate. While many brands and manufacturers choose or are required to complete batch testing of their products, this analysis is infrequent and results unreliable, as testing often happens before final production runs are completed. Chemical testing is often used for these singular tests, and due to the destructive, costly, and time consuming nature of the analysis, only small portions of materials can be validated, leaving the composition of most produced material untested and its composition data dependent on manual entry and reporting. This material composition data crisis leads to a loss of billions of dollars across the industry; impacting private stakeholders like mills and brands through incorrect costs, public entities through missed tax revenue, and downstream organizations like recyclers and textile sorters through inaccurate feedstock, all due to mislabeled and inaccurately tested textile goods. Refiberd, a seed stage start-up based in San Francisco, CA, has developed a novel technology using hyperspectral imaging and artificial intelligence that has the potential to detect the fiber composition of textiles within 2% of actual composition. Since the company’s inception in 2020 and subsequent SBIR funding in 2024, we have completed 10+ pilots with global brands, textile recyclers, and material testing facilities, demonstrating success in the accurate material detection for textile samples with 4+ fiber complex blends, including contaminants like elastane and dark dyes. The primary challenge in developing an accurate software for fiber composition, lies in the lack of “ground truth” data available for most textiles, making it difficult to build and train a generalizable and precise algorithm. Refiberd has sourced over 15,000 material samples from the industry with composition validated by trusted suppliers or chemical analysis in order to ensure the accuracy of our technology. Refiberd’s accurate, efficient, and non-destructive technology offers significant positive return for both the private and public sectors. The technology provides the opportunity to accurately detect the composition of a material within milliseconds, allowing for high volume throughput and decreased costs compared to chemical analysis. The impact of this technology has the potential to provide 2X ROI to material testing facilities after the first year of implementation by eliminating the need for chemical testing for materials with high-confidence predictions, and save brands up to $39 billion annually by preventing substitutions in production materials. At second-life and end-of-life, Refiberd’s technology can recoup $20 billion of counterfeit goods which currently go undetected, and save textile recyclers up to $7.5 billion in production yield losses. It is imperative that accurate and scalable technology is adopted to close the textile data gap in order to prevent future financial loss and increase transparency for all stakeholders.