This talk will initially present a bibliometric analysis of artificial intelligence (AI) in textiles, covering research from 1987 to 2024 involving over 3,000 published papers. The analysis reveals that AI methods int textiles have largely remained consistent over the years. However, specific areas such as fabric, quality, defects, fashion, yarn, color, apparel, waste, and spinning have attracted significant research interest, indicating a strong focus on consumer-facing aspects of the textile industry. In contrast, textile manufacturing processes like carding, apparel sewing, knitting, dyeing, and finishing show relatively low levels of AI implementation, suggesting considerable opportunities for innovation in these areas.

The second part of the talk will address a specific critical challenge of using AI as a tool to predict the color of a dry fabric from reflectance measurements of a wet fabric.  The study involved experimentation in a controlled environment of 700+ fabric samples to model the mapping of colors in the wet state to their dry state.

An optimal deep learning neural network architecture was developed, featuring four input neurons, ten hidden layers with fifty neurons, and an output layer with three neurons. Additionally, several traditional machine learning models were implemented for comparative analysis. The neural network's prediction error was evaluated using mean squared error (MSE) and a custom loss function (ΔE2000). Results on test data showed that the neural network with the custom loss function achieved a mean color difference prediction of under 1 over the color gamut, significantly lower than the mean baseline color difference of 13.1. Neural network-based AI methods can effectively predict fabric color, offering the potential to automate critical textile processes, reduce dye wastage, minimize water pollution, and improve overall efficiency and cost-effectiveness in dyeing operations.