The rapid advancement of artificial intelligence (AI) has led to its widespread adoption across various engineering fields, including composite materials research. Composite materials, known for their superior mechanical properties and lightweight characteristics, play a crucial role in industries such as aerospace, automotive, and robotics. However, their inherent complexity-such as anisotropic behavior, nonlinear characteristics, and intricate microstructures-poses significant challenges for traditional design and analysis methods. To address these challenges, AI-driven approaches have emerged as powerful tools, offering solutions in prediction, generation, and automation. This review systematically explores applications of machine learning and deep learning in composite materials research, categorized into three major approaches: predictive, generative, and automation models. Predictive models enhance the accuracy of property prediction and microstructure analysis. Generative models facilitate novel material discovery and microstructure design. Automatic models improve quality control and can be used to optimize manufacturing processes through real-time data analysis. By leveraging diverse large-scale datasets, AI provides innovative solutions to the key challenges associated with composite materials and enhances research and design efficiency. This review highlights the transformative potential of AI in composite materials research, providing insights into future research directions and challenges.

Additive friction stir deposition (AFSD) is a solid-state manufacturing technique capable of producing high-strength, defect-free metal components. The complexity of its process parameters has driven growing interest in machine learning (ML) for improved predictive accuracy and process control. This study presents a novel biomimetic ML approach to predict the mechanical properties of AFSD-fabricated aluminum alloy-walled structures. The methodology integrates numerical modeling of the AFSD process with genetic algorithm (GA)-optimized ML models to predict von Mises stress and logarithmic strain. Finite element analysis was employed to simulate the AFSD process for five aluminum alloys: AA2024, AA5083, AA5086, AA7075, and AA6061, capturing the complex thermal and mechanical interactions involved. A dataset of 200 samples was generated from these simulations. Decision tree and random forest (RF) regression models, optimized using GAs, were developed to predict key mechanical properties. The RF model demonstrated superior performance, achieving R² values of 0.9676 for von Mises stress and 0.7201 for logarithmic strain. This innovative approach provides a robust tool for understanding and optimizing the AFSD process across a range of aluminum alloys, offering valuable insights into material behavior under various process parameters.

The integration of artificial intelligence (AI) into textile design enhances functionality, automation, and user interaction. While gesture recognition has been explored in smart textiles, contactless interactive systems for healthcare remain underdeveloped. This study presents a human-centered co-design approach to the development of an AI-integrated gesture recognition system embedded in illuminative textile wall panels, aimed at enhancing spatial engagement in healthcare environments. The research was conducted in three key stages. First, a co-design workshop was conducted to explore user preferences in textile materials, graphic design, and gesture interaction. Second, intelligent illuminative textiles were developed by knitting polymeric optical fiber into base wool yarns to enable illumination. A camera was embedded and integrated with a computer vision-based deep learning model for detecting landmarks on the hands, shoulders, and head. The recognized gestures and body movements triggered specific pre-programmed color changes on the textile surface through edge-integrated light-emitting diodes. Finally, a prototype was fabricated and installed in a government-established District Health Centre in Hong Kong to support physical activity and rehabilitation for elderly users. Semi-structured interviews with stakeholders - including co-designers, users, and occupational therapists - were conducted to evaluate usability and inform design refinements. Stakeholders reported high levels of satisfaction, emphasizing the system’s ability to enhance community connection, therapeutic engagement, intuitive usability, and compelling visual feedback. These findings suggest that AI-driven interactive textiles present promising opportunities for rehabilitation, therapeutic environments, and the promotion of elderly well-being.

Controlling shrinkage behavior in electrospun membranes is critical for applications that require precise dimensional or mechanical performance. However, experimental variability and limited datasets often hinder the development of robust process models. This study introduces a data-driven framework that combines machine learning with Monte Carlo simulation to enable both accurate and stable shrinkage control in electrospinning using a small experimental dataset. Multiple regression models were trained to predict biaxial shrinkage ratios and their variability, with support vector regression and extreme gradient boosting showing the best performance for accuracy and stability prediction, respectively. Feature importance analysis revealed applied voltage and thermoplastic polyurethane concentration as the dominant parameters. A Monte Carlo-based optimization strategy was employed to identify process parameter sets that achieve target shrinkage ratios while minimizing output variability. The proposed approach enables multi-objective optimization in low-data, high-variability manufacturing environments, offering practical insights into precision fabrication of stimulus-responsive membranes.

In this paper, an improved fabrication method is presented for fabricating carbon nanotube (CNT) based multi-functional bucky paper (CNT-BP) sensors that will be primarily used for adaptive sensing in structural health monitoring applications. A large number of BPs were fabricated using multi-walled CNTs with varying methanol-CNT compositions, sonication times, temperatures, curing durations, membrane thicknesses, and electrode placements to determine the optimal configuration for large-scale production. The obtained optimal configuration of the ingredients that yields an adequate sensitivity and ductility of the CNT-BP was then employed for measuring the crack propagation behavior in the fatigued samples. Further, a long short-term memory (LSTM)-based neural network was proposed for prognosis in a metallic plate with fatigue crack propagation. The actual crack lengths of the fatigue crack obtained by the high-speed digital camera were correlated with that predicted by the CNT-BP-based model and LSTM, showing good agreement. Thus, the present study demonstrates that the proposed improved method of CNT-BP is highly efficient in the diagnosis and prognosis of fatigue cracks in metallic structures.