Neural Network Applications in Yarn Quality Prediction

Neural networks have emerged as powerful tools for predicting key quality attributes of yarn, including tensile strength, evenness, density irregularities and functional performance. By capturing non-linear relationships between fibre properties, spinning parameters and mechanical behaviour, various architectures—ranging from feedforward multilayer perceptrons to recurrent networks with attention mechanisms—enable accurate forecasts of yarn performance under diverse production conditions. These data-driven models support real-time process control, reduce trial-and-error sampling and drive efficiencies across spinning, blending and finishing stages. Recent innovations incorporate metaheuristic training algorithms and hybrid analytical-neural frameworks to extend predictions from elastic to viscoelastic–plastic regimes and to model time-dependent or sequential process data. Such approaches underpin global initiatives to enhance textile sustainability and product consistency.

Research from Nature Portfolio

Recent studies have introduced a hybrid analytical-geometrical neural network model to predict tensile behaviour of close-packed multifilament yarns containing two to five monofilaments in the core. By integrating a geometrical representation with a feedforward artificial neural network, this approach extends predictive capabilities from elastic through viscoelastic–plastic responses. Experimental validation and numerical simulations demonstrate excellent agreement with measured tensile curves, highlighting the model’s potential to improve understanding and control of yarn mechanical properties in advanced textile structures.

Neural Network Applications in Yarn Quality Prediction publication trend

The graph below shows the total number of articles in neural network applications in yarn quality prediction across all publications each year (not limited to Nature Index journals).

Technical terms

Artificial neural network: A computational framework of interconnected processing units that learns complex input-output mappings by adjusting connection weights.

Multilayer perceptron (MLP): A feedforward neural network composed of multiple layers of neurons with nonlinear activation functions, trained by backpropagation.

Gated recurrent unit (GRU): A type of recurrent neural network cell that captures temporal dependencies through update and reset gating mechanisms.

Attention mechanism: A technique that enables a model to weight input features differentially, focusing on the most relevant information for prediction.

Broad multilayer neural network (BMNN): An expanded neural architecture integrating broad learning systems with multiple hidden layers to enhance training efficiency and predictive performance.

Extreme learning machine (ELM): A rapid training single-hidden-layer feedforward network with randomly initialised hidden nodes and analytically determined output weights.

Viscoelastic–plastic behaviour: Material response exhibiting both time-dependent elastic deformation and irreversible plastic flow under load.

References

1. A viscoelastic-plastic model for the core of various close-packings of multifilament polyamide-6 yarns. Scientific Reports (2024).
2. Prediction of Cotton Yarn Quality Based on Attention-GRU. Applied Sciences (2023).
3. Prediction of yarn unevenness based on BMNN. Journal of Engineered Fibers and Fabrics (2021).
4. The use of extreme learning machines (ELM) algorithms to prediction strength for cotton ring spun yarn. Fashion and Textiles (2016).