Laser-assisted automated fiber placement (LAFP) enables high-speed consolidation of thermoplastic composites, yet unstable energy input often leads to weak interlaminar bonding and porosity. Existing optical-thermal models remain limited in resolving three-dimensional energy temperature coupling and supporting process-oriented optimization. This study develops a highly parameterized Monte Carlo ray tracing (MCRT) framework coupled with a transient thermal model to predict spatially resolved heat flux and temperature evolution during LAFP. The optical model resolves tow-scale anisotropic absorption and incorporates key process parameters, including laser incidence angle, spot geometry, energy distribution ratio, and placement speed. Model predictions are validated using eight-tow T700/PEEK layup experiments with synchronized infrared thermography and embedded thermocouples. The coupled model reproduces measured temperature histories with a relative deviation of 1.8% for surface peak temperatures and interlaminar RMSE values ranging from 5.9 to 10.8 degrees C. Parametric analysis reveals that the energy distribution ratio governs interfacial melting behavior, while spot geometry and placement speed control peak temperature and thermal dwell time. The proposed framework provides a quantitative basis for robust process window design in thermoplastic composite placement.

This study presents a novel 3D interlayer contact model based on fractal theory to quantify the evolution of intimate contact during thermoplastic prepreg processing. The rough surface is modeled as a 3D Cantor set embedded in a 2D plane, and the 3D deformation of asperities under processing conditions is described using a creeping squeeze flow model in polar coordinates. Key parameters, including fractal dimension, scaling factor, and a dimensionless surface roughness ratio, are calibrated and used to predict interlayer contact in three-layer prepregs under various thermoforming conditions. Experimental validation using metallographic and C-scan analysis confirms the model's accuracy, with most prediction errors within 5%. Compared to conventional 2D models, the 3D model provides improved accuracy by accounting for the 3D flow of asperities, particularly under higher temperature and pressure. Finally, the independent impact mechanisms of the processing parameters and surface geometry on intimate contact are analyzed, with temperature and fractal dimension exerting the greatest impact on the rate of contact development. This model serves as an effective tool for predicting intimate contact evolution in thermoplastic composite processing and provides theoretical support for interlaminar strength prediction and manufacturing process optimization.

Enhancing manufacturing efficiency of composite components has remained a pivotal research frontier in automated fiber placement technology, where multi-robot collaborative systems have gained prominence as an effective implementation strategy. However, for rotary components with complex geometric features, collaborative multi-robot placement faces challenges such as intricate processing paths, unbalanced task allocation, and poor synchronization. This study proposes a dual-robot task allocation and feature-guided hierarchical multi-objective motion optimization method for redundant systems involving “rotary positioner, mobile platform, and dual robots.” First, a dual-robot collaborative task allocation strategy is established based on kinematic reachability. A hierarchical motion optimization framework is then designed for the redundant system, enabling stepwise trajectory planning for the rotary axis, mobile platform, and dual-robot collaboration. Key innovations include a bilateral path curvature-guided segmented constant-speed motion planning method for the rotary axis, which coordinates path features with rotary axis rotation; a mobile platform following strategy based on spatial optimization analysis of the rotary system; and a dual-robot collaborative motion optimization method combining simulated annealing genetic algorithms to address multi-objective constraints. Experimental validation on complex rotary surfaces demonstrates that the proposed curvature-guided rotary axis strategy enhances placement efficiency and accuracy compared to existing methods. The method ensures smooth rotary axis rotation while significantly improving joint motion stability.

To evaluate the impact of triangular gap defects induced by automated fiber placement (AFP) on the mechanical response of composite laminates, this study proposes a multilayer defect unit cell (MDUC) modeling framework. Firstly, tow-path data are employed to map the gap geometry onto the finite element (FE) model. Subsequently, under periodic boundary conditions (PBC), multi-case loading responses combined with layer-wise back-mapping are utilized to identify the equivalent stiffness and strength of each ply, while accounting for coupling effects between adjacent plies. Then, the macroscale FE model is executed in Abaqus using the 3D Hashin failure criterion via a user-defined material subroutine (VUMAT). Finally, open-hole uniaxial tensile tests with digital image correlation (DIC) measurements demonstrate that the proposed MDUC model can capture the main trends of the macroscopic load-displacement response, full-field strain distribution, and the progressive damage evolution and failure process. Compared to the defect-free baseline, the experimental and numerical peak loads for the defective specimens containing the defect in the −45° ply and 90° ply exhibited reductions of approximately 1.45% (1.80%) and 0.90% (1.21%), respectively. The proposed framework provides an efficient sequential multiscale approach for defect-sensitive analysis, defect-tolerant design, and process optimization of composite laminates.

Automated fiber placement (AFP) of complex curved structures is frequently hindered by steering-induced tow wrinkling. Existing wrinkle prediction approaches typically involve a trade-off among geometric applicability, process fidelity, and computational efficiency. To address this issue, this study proposes a multiphysics-coupled framework for wrinkle assessment that integrates material properties, path geometry, and process-parameter effects within a unified analytical formulation. A physics-informed wrinkle criterion is introduced to characterize wrinkle initiation as a function of process variables and interfacial tack evolution. An FE-calibrated analytical contact pressure model applicable to arbitrary curved surfaces is also developed to provide the process-state quantities required for wrinkle assessment. These components are integrated into a unified framework for wrinkle prediction and mitigation and are applied to representative winglet and S-shaped surfaces, followed by AFP experimental validation. The results show that the proposed approach not only identifies geometry-driven wrinkling risks induced by excessive path curvature, but also captures process-induced risks associated with locally insufficient interfacial tack under suboptimal process conditions. The computation time is only 9.5 s for an 8-tow course with an approximate length of 1400 mm. For the winglet surface, reducing the number of tows within a course eliminates the predicted wrinkling risks at the course edges in the local convex region. For the S-shaped surface, process optimization reduces the predicted wrinkling risk from approximately 20% to zero. Experimental results further confirm high prediction accuracy and effective wrinkle suppression, supporting manufacturability assessment and process planning for engineering-scale AFP applications.

Mechanical property analysis in defect-containing composites is essential to composite design, yet existing methods suffer from an inability to simulate randomness, limitations in scale, or high computational resource consumption. Therefore, a Transformer model integrated with principal component analysis (PCA) is proposed for the first time to accurately predict stress-strain (S-S) curves in mesoscale carbon fiber reinforced polymer (CFRP) laminates with random gaps. Based on automated fiber placement (AFP), a laminate with random gaps finite element model is created. The model's reliability is verified experimentally, and a Python script for parameterized modeling is developed to build the dataset. Subsequently, PCA is applied to reduce the dimensionality of the S-S curves, lowering model complexity and computational demands. A Transformer model is then constructed, with its sample size sensitivity and stability analyzed through different dataset sizes and 5-fold cross-validation. The model's performance is comprehensively evaluated by comparing it with four other deep learning models designed to predict S-S curves. The results indicate that S-S curves predicted by the Transformer model correlate well with results from finite element simulations, with a correlation coefficient above 0.9999. The relative percentage errors in tensile strength and elastic modulus derived from the predicted curves remain within 0.28%, and the Transformer model demonstrates significant advantages in both accuracy and computational resource consumption. The proposed model combines high accuracy with computational cost-effectiveness, which characterizes mesoscale defects and predicts their impact on mechanical properties, promising a new and highly potential tool for composite material design optimization and uncertainty quantification.

Continuous ultrasonic welding (CUW) is a promising technology for assembling large-scale thermoplastic composite structures; however, achieving robust joint quality remains challenging due to the complex dynamic thermal equilibrium involved. Unlike static welding, CUW operates under a moving thermal equilibrium in which energy density and residence time must be precisely balanced to ensure effective polymer melting. In this study, the CUW process of continuous carbon fiber-reinforced PEEK (CCF/PEEK) laminates was systematically investigated using Response Surface Methodology (RSM), integrated with in-situ infrared thermography and instantaneous power-signature analysis. A pronounced non-linear interaction between vibration amplitude and welding speed was identified. An optimized processing regime produced cohesive failure and a maximum lap shear strength of 33.14 MPa. Excessive vibration amplitude applied without sufficient welding pressure was found to induce dynamic acoustic decoupling, leading to intermittent contact and inefficient energy dissipation as localized impact heating. This phenomenon is explained by the hammering effect, which accounts for deterioration in joint quality at high nominal energy inputs. It also demonstrates that joint formation is governed not solely by nominal energy input but by the stability of acoustic coupling at the welding interface. The corresponding instantaneous power waveform serves as a sensitive process fingerprint for distinguishing effective coupling from decoupled welding states. By elucidating the coupled roles of energy density, acoustic coupling stability, and thermal history, this work defines a robust processing window for CUW of CCF/PEEK and highlights the potential of power-signature monitoring for in-situ control.

Continuous ultrasonic welding (CUW) is a promising technology for assembling large-scale thermoplastic composite structures; however, achieving robust joint quality remains challenging due to the complex dynamic thermal equilibrium involved. In this study, the CUW process of continuous carbon fiber-reinforced PEEK (CCF/ PEEK) laminates was investigated using Response Surface Methodology (RSM), in-situ thermal monitoring, ultrasonic C-scan inspection, and instantaneous power-signal characterization. A strong interaction between vibration amplitude and welding speed was identified, showing that joint quality depends on the balance between effective energy input and thermal exposure during the moving welding process. The optimized processing condition produced cohesive failure and achieved a maximum lap shear strength of 33.14MPa. Quantitative analysis of temperature and power curves further revealed that high nominal energy input does not necessarily lead to high joint strength. Under high amplitude and insufficient pressure conditions, intermittent acoustic contact caused hammering-related instability, in which part of the input energy was dissipated through impactdominated mechanism rather than interfacial heating. The coefficient of variation of power, CVP, was introduced to evaluate power-signal stability: a low CVP corresponded to stable acoustic coupling, whereas pronounced power drops and higher CVP values indicated intermittent decoupling. These findings demonstrate that CUW quality is governed by effective energy transmission, thermal history, and acoustic coupling stability rather than nominal energy input alone. The identified process window provides guidance for robust CCF/PEEK continuous ultrasonic welding and supports the use of power-signal monitoring for in-situ process assessment.

Automated fiber placement (AFP) quality control requires defect localization under limited annotation and diagnostic feedback grounded in process knowledge. This paper proposes a knowledge-enhanced multimodal agent that couples depth-map-based localization with ontology-constrained graph retrieval in a tri-space framework. To adapt Qwen3-VL-8B to AFP depth maps, we develop a two-stage post-training strategy. Supervised fine-tuning establishes modality alignment, defect-term grounding, and structured-output learning, while group relative policy optimization refines the language-side policy with detection-oriented rewards. On a production dataset of 1403 depth maps, the proposed method achieves 0.842 class-agnostic recall with only 40 labeled images and remains competitive in the 20-shot setting, surpassing single-stage baselines trained with 40 labels. Repeated experiments show stable performance under subset-sampling and optimization randomness. For diagnostic question answering on 50 multi-turn defect cases, graph-structured retrieval-augmented generation (RAG) improves claim-level evidence grounding over a text-chunk vector RAG baseline. It increases fully supported diagnostic claims from 70.6% to 80.4%, reduces unsupported claims from 13.4% to 7.9%, and improves citation precision from 84.6% to 91.4%, with 3.2 s average latency per question. These results show that metric-aligned post-training and ontology-constrained evidence retrieval support accurate AFP defect localization and auditable, mechanism-consistent diagnosis under limited supervision.

With the widespread application of deep learning algorithms in materials science, this paper proposes a hybrid Convolutional Neural Network-Multilayer Perceptron (CNN-MLP) model for efficiently predicting the transverse elastic properties of unidirectional carbon fiber-reinforced composites (UD-CFRP) containing microvoids. A two-step method is introduced to obtain the dataset required for the CNN-MLP model quickly. In the first step, a greedy algorithm establishes a fiber-resin Representative Volume Element (RVE) model, treating the homogenized results as a new material. In the second step, a new material-void RVE model is constructed, with its images and homogenized results serving as the required dataset. Comparisons with conventional modeling homogenization and Mori-Tanaka calculations show that the two-step method significantly improves the efficiency of large-scale modeling and homogenization while maintaining a difference within 5

Efficient detection of elongated defects (gaps and overlaps) in Automated Fiber Placement (AFP) is critical for enhancing the quality of carbon fiber composite components. To address limitations of traditional 2D inspection methods, including high false-negative rates and poor mask continuity, this paper proposes a real-time defect detection framework integrating 3D laser scanning and enhanced instance segmentation. First, high-precision point cloud data acquired via a 3D Laser Scanner is converted into depth images using a noise-resistant mapping algorithm, enabling submillimeter defect discrimination while correcting contour distortions. Second, we enhance the YOLOv11 instance segmentation model by introducing an Elongated Focal Loss (EFL) that emphasizes endpoint features and longitudinal continuity constraints, coupled with a three-stage dynamic training strategy for progressive performance optimization. Finally, a Dense Conditional Random Field (DenseCRF) post-processing module refines mask boundaries using depth image semantics. Experiments on an industrial dataset of 1030 images demonstrate 99.1

Annotated 3D defect data remain a major bottleneck for automated fiber placement (AFP) inspection, especially for rare defects with complex geometries and limited process coverage. We propose S2G-Net, a semantic-to-geometry framework that generates controllable AFP laser-profile point clouds by separating defect semantics from geometric rendering. Instead of perturbing existing scans, S2G-Net enables the synthesis of diverse defect configurations while preserving AFP-specific tow structures and surface-profile characteristics. Experiments on ZJU-AFP-ScanSeg3D show that synthetic-data quality and selection are more important than scale alone: a utility-guided subset of 1500 synthetic samples achieves an mIoU of 0.842, outperforming the full 40000-sample synthetic pool (0.832). External AFP-line validation further demonstrates cross-line adaptation capability, where synthetic pre-training with only 10% target-line labels achieves 0.890 mIoU, reaching 98.2% of full supervision (0.906 mIoU). S2G-Net provides a controllable 3D synthetic-data foundation for AFP quality assurance and supports scalable defect data generation for intelligent composite manufacturing.

The structural integrity of composite laminates manufactured via automated fiber placement is often compromised by tow-drop gap arrays and their complex overlaps. This paper presents a study of the effects of overlapping of tow-drop gaps in composite plies with different fiber orientations under low-velocity impact (LVI) events. Three multi-directional overlapping configurations (biaxial, tri-axial, and quad-axial) are designed with gaps incorporated into the (+45 degrees), (- 45 degrees/90 degrees/45 degrees), and (0 degrees/- 45 degrees/90 degrees/45 degrees) plies, respectively. Additionally, the impact sensitivity of specific locations within triangular gap arrays, i.e., the interconnecting vertex and the triangle centroid, is examined. Relative responses of specimens are determined through LVI tests with the impact energy of 30 J and compression after impact (CAI). After the LVI tests, the damage of specimens is characterized and analyzed using a combination of visual inspection, ultrasonic C-scan inspection, and numerical simulation. It is concluded that the damage resistance degrades progressively as the number of off-axis plies containing triangular gaps increases. Compared with the gap-free specimen, the peak force of LVI decreases by 0.9%-8.7%, the absorbed energy increases by 18.6%-54.5%, and the CAI strength decreases by 2.2%-20.2% in the gap-embedded specimens. These observations are attributed to the localized thickening effect of gaps. And the triangular gap vertex, acting as a convergence singularity, exhibits higher impact sensitivity. In the currently designed laminates, its presence within the unidirectional gap alignment induces a more significant performance knockdown than even the overlap with quad-axial gaps.

This study concerns the effect of tow-to-tow gaps and their distribution induced by automated fiber placement on the mechanical performance of large composite structures. A gap volume element (GVE) model is first presented for cross-scale analysis of the mechanical behavior of composite panels with tow gaps under realistic engineering conditions. In the GVE model, the mesh elements containing gap defects can be homogenized to account comprehensively for the effects of the geometric volume fraction and spatial distribution of gaps within the solid elements, along with the influence of tow angle deviation. Based on simulated tests under elastic property identification loading and micromechanical theory, the equivalent in-plane elastic stiffness matrix and strength matrix of the elements containing gap defects were reconstructed. Subsequently, the GVE model was validated against the available uniaxial tensile tests on specimens containing triangular gaps, and excellent agreement was obtained. Finally, based on the GVE model, a sequential hierarchical multiscale evaluation framework was established to assess the influence of different gap distribution schemes on the mechanical behavior of composite panels. The evaluation results indicate that a more uniform distribution of gaps within the panel is beneficial to the structural load-bearing capacity.

Automated fiber placement (AFP) technology is widely employed in the manufacturing of large-scale carbon fiber reinforced polymer (CFRP) components with complex geometries, such as wing skins and engine inlets. However, inherent process limitations lead to the formation of triangular gaps at tow-drop locations, and the stacking of these gaps during layup can significantly compromise the mechanical performance of the final components. This study experimentally investigates the effects of staggering on the low-velocity impact (LVI) response of CFRP laminates containing such triangular gaps. Laminates with five different staggered configurations were fabricated using AFP technology, along with two types of baseline laminates: one without gaps and one without staggering. All specimens were subjected to LVI testing, and post-impact damage was characterized using ultrasonic C-scan, digital camera, and optical microscopy. The results demonstrate that properly selecting the staggered parameters, specifically the staggered interval and staggered distance, can effectively enhance the impact performance, including increasing peak force, reducing delamination area, and altering crack propagation behavior. Compared to the defect-free scheme, the fully stacked defect scheme exhibits a 74.41% increase in delamination area, a 17.41% decrease in peak force, and more severe impact damage. For a fixed staggered interval, an intermediate staggered distance, corresponding to one tow width in this study, provides a more favorable balance between damage suppression and energy dissipation compared to smaller or larger offsets. Moreover, for a fixed staggered distance, changing the staggered interval affects the through-thickness distribution of defects and the resulting delamination behavior. These findings provide valuable guidance for improving the damage resistance of AFP-manufactured complex curved components.

The automated fiber placement (AFP) technology has drawn considerable interest to replace the traditional hand laid-up, providing greater productivity, precision, repeatability, reduced waste, and the capability to manufacture complex curved structures. To further advance the application level of AFP, it is essential to exert extraordinary efforts in process planning. In this paper, a collaborative process planning method incorporating process conditions and manufacturing constraints is first developed and described in detail. In it, the mechanism of the key process parameters including pressure, head speed as well as material temperature are analyzed and predicted, respectively. In parallel, the most key manufacturing constraint, critical steering radius, is defined and elaborated. Thereafter, the layup process parameters for a winglet tool is obtained based on the collaborative process planning method. Finally, the method is investigated by the placement experiments of the winglet. Results demonstrate that the method can realize highly efficient and improve the layup quality of the winglet. Such that, the collaborative process planning method can provide an excellent foundation for enhancing the manufacturing precision and efficiency as well as placement quality for complex curvature structures.Highlights A collaborative process planning method is developed and described in detail. FEM simulations and trials of the winglet are utilized to confirm the method.

Automated fiber placement (AFP) technology enhances the structural efficiency in composite manufacturing through advanced steering control. However, its broader adoption is limited by steering-induced defects, particularly the out-of-plane buckling (wrinkling) of prepreg tows. Prepreg tack, a viscoelastic and process-dependent interfacial property, governs wrinkle initiation. This study develops a theoretical model to capture the coupled time-process effects driving wrinkle formation, providing a basis for defect suppression. Sensitivity analysis identifies normal tack as the dominant factor. Analytical solutions for the stress distribution coefficient and evolving minimum steering radius are derived. The model parameters are obtained experimentally. Model validation against steering experiments demonstrates strong agreement, with a maximum deviation of only 6.7% in the steady-state (equilibrium) minimum steering radius. The results underscore the potential of process optimization in enhancing tack performance and suppressing wrinkles. Based on these findings, a time-processinformed wrinkle control framework is proposed to support AFP design optimization, offering a practical approach to improve the quality and reliability of steering deposition.

Automated fiber placement (AFP) enables the efficient fabrication of fiber-reinforced composites. However, the complex geometries of aircraft components often require fiber directions with variable angles, complicating the simultaneous satisfaction of three key manufacturing constraints in AFP path planning: path alignment, path parallelism and path curvature. To address it, a field-based partition framework is developed via singularity construction. First, the vector heat method smooths fiber directions to reduce geodesic curvature. Then, benefiting from the singularities that are constructed by eliminating the vector curl, the ply surface is partitioned from the singularities into patches with improved parallelism of each patch’s vector field. The final laying paths are generated on each partition by parallel offsetting the initial path that comprehensively considers fiber directions over the partition. Compared with the exiting path planning strategy, the proposed method finds higher-quality tow paths with enhanced fiber alignment, lower curvature, fewer partitions and full tow coverage, providing a new paradigm for AFP path planning on complex surfaces.

In automated fiber placement (AFP), addressing the continuous motion planning challenge of redundant layup manipulators in complex environments, this paper proposes an offline redundancy optimization algorithm based on improved RRT* (Rapidly-exploring Random Trees). This algorithm maximizes the utilization of kinematic redundancy to derive smooth joint trajectories devoid of collisions and singularities. Firstly, the algorithm entails constructing a search map by eliminating joint configurations that violate constraints, and subsequently planning and optimizing the joint path by minimizing a multi-objective cost under the map constraint. Furthermore, several strategies are introduced to enhance RRT* for redundancy optimization. These strategies include a piecewise Gaussian sampling strategy (PGSS) to guide efficient tree growth within complex channels and enable joint sampling constrained by task coordinates. Additionally, the improved Steering and Local Optimization method are proposed to plan joint motion while considering intermediate task sequences. The effectiveness of the proposed algorithm is demonstrated in handling complex motion planning scenarios, such as layup involving complex path curves or dense obstacles. Experimental results validate the algorithm's capability to find feasible collision-free and singularity-free paths in relevant scenarios, provided such paths exist. Moreover, trajectory smoothness is optimized with increasing iterations.