BEST-BEFORE: Optimizing clothing service life through predictive analytics for sustainable longevity
| Reference number | |
| Coordinator | Högskolan i Borås - Textilhögskolan |
| Funding from Vinnova | SEK 497 189 |
| Project duration | December 2019 - January 2021 |
| Status | Completed |
Important results from the project
Currently in the clothing industry, design for longevity clearly has a crucial role to influence sustainability. However till date, clothing longevity is estimated during the service lifetime based on consumers’ perception, or by conducting durability tests in product development, that gives only a static evaluation of garment lifetime. BEST-BEFORE proposes a predictive model based on machine learning that can approximate how long a garment can last based on different properties, by producing “digital” image twins and proposing extrapolated end-of-lifetimes.
Expected long term effects
Best-Before’s predictive model based on computational GAN provides a good introductory tool and a groundbreaking concept, for approximating how long a garment can last by capturing the visual degradation through washes, and producing the “digital” image twins having features visibly similar to real samples. Qualitative evaluation of the synthetic images produced can be extrapolated to predict the end-of-life (in terms of individual properties such as visual degradation and wear-and-tear) and also in an aggregated way.
Approach and implementation
Best Before is divided into four activities. An extensive mapping of wash-related durability requirements and relevant standards first led to design of the experiment. Next, wash experiments were conducted on Fjällräven’s Abisko Lightweight trousers and G1000 fabrics over 70 cycles, to create a library of visual images and abrasion resistance test data. Acquisition, labelling, and training using conditional GAN model resulted in predicting end-of-life by creating digital image twins. Finally, feasibility was assessed through expert evaluation and risk assessment.