I. Core Breakthroughs

Breakthrough 1 : Successfully achieved in-situ simultaneous measurement of the four major sources of fluctuations, quantifying the contribution weights of each defect.

Previous studies have examined only a single type of error, making it impossible to compare their relative magnitudes across different categories. In this paper, we employ a laser tracker, an industrial camera, and a dual-line laser scanner to simultaneously capture four types of variations, quantifying the proportion of variation within each category under standardized operating conditions:

• Tow roller lateral displacement: fluctuation range 1.4mm , accounting for the standard 6.35mm Tow width 22.1% (Primary defect cause);
• Raw material tow width variation: 0.63mm , accounting for 9.9% ；
• Robot And Positioning deviation: 0.54mm , accounting for 8.5% ；
• Tow deformation after compaction: 0.48mm , accounting for 7.5% 。

It has been clearly established that roll‑system slippage is the highest‑priority corrective action for process optimization, fundamentally challenging the conventional belief that “robotic precision determines the extent of defects.”

Breakthrough 2 : Distinguish between position-based and geometric types of wave phenomena, and establish the underlying mechanisms of defect formation.

Clearly distinguish between two types of manufacturing variation:

• Positional drift (robotic deviation) +  Roll-to-roll slip: Directly altering the transverse coordinate of the yarn bundle results in twice the gap for the same amplitude of fluctuation. /  Overlap error;
• Geometric variation (incoming material width) +  Compressive deformation: Only the cross-sectional dimensions of the wire are altered, reducing the impact of defects by half.

This classification model can be directly integrated into path‑simulation software, accurately reproducing the real‑world distribution of ply‑layup defects.

Breakthrough 3 : Monte Carlo simulation based on probability distributions for high-precision gap prediction

Optimal probability functions are fitted separately for each of the four types of errors (robot: Logistic ; Roller slip: extreme value distribution; tow width: normal distribution), a neighboring‑filament simulation algorithm is developed; the mean and standard deviation of the simulated gap are calculated, 90/99 The deviation between the quantile values and the measured data is less than. 10% , validating the predictive model’s effectiveness and enabling the prediction of defect occurrence probabilities and size ranges without extensive trial paving.

Breakthrough 4 : Proposes an online camera-based compensation approach to achieve real-time correction of the layup process.

Leveraging industrial cameras to capture real-time data on filament bundle misalignment on the roller, the system proactively predicts future gap trends and simultaneously dispatches the robot. And Coordinate compensation command, enabling dynamic correction during the lay-up process, for… AFP Closed-loop intelligent placement provides sensing. -  Integrated control implementation plan.

II. Technological Innovation

Innovation 1  Multi-sensor integration and consolidation AFP In-situ measurement hardware platform

• Laser tracker ( Leica AT960 ): Millisecond-level data collection for the placement head TCP Three-dimensional coordinates, calculation And Positioning deviation, with micrometer-level measurement accuracy;
• 500  Wansu High-Speed Industrial Camera: Positioned directly opposite the compaction roller area, it uses grayscale thresholding to detect the offset of the tow edge relative to the roller center, thereby quantifying lateral slip.
• Double 2D Laser line scanning: measures the raw width of the feed material at the front end and the compacted cross-section at the rear end, simultaneously capturing compaction-induced deformation.
• 3D Print a modular sensor bracket compatible with mainstream devices. AFP-XS Head placement is convenient for assembly and disassembly, with no interference with the laying motion.

Innovation 2  Standardized Data Calibration and Parsing Algorithms

• Checkerboard pixels -  Millimeter‑level unified calibration eliminates measurement errors caused by lens distortion.
• Temporal synchronization: Laser, camera, and robot data are aligned using a unified timestamp to eliminate biases caused by desynchronization.
• Distribution-fitting assessment: The log-likelihood criterion is used to select the optimal probability density function, resulting in four standard statistical models for error types.
• Layered statistics: Distinguish between uniform-speed and accelerating/decelerating laying intervals, and analyze the discrete patterns of the robot’s dynamic positioning.

Innovation 3  Neighbor-Simulated Monte Carlo Defect Simulation Algorithm

Five-Step Standardized Simulation Workflow:

• ① Sampling robot position offset; ② Slippage of the tow between sampling rollers, combined to determine the tow center; ③ Original width of the sampled feed; ④ Cross-section corrected for deformation after compaction; ⑤ Gap calculated from the edge-to-edge spacing of adjacent tows. /  Overlap.
• Batch simulations can generate defect‑occurrence probability curves for various preset filament spacing settings, helping process engineers select the optimal lay‑up offset to balance the risks of overlap and gap.

Innovation 4  Grading Process Optimization and Quantitative Assessment Tool

By isolating and eliminating specific types of variability, we simulate the magnitude of reductions in defect probability, providing a clear assessment of the benefits of corrective measures: prioritizing optimization of roll‑system constraints and enhancing tow tension to minimize slippage yields the greatest improvements; followed by refining cut‑to‑length material‑feed accuracy and implementing robotic spatial compensation, thereby furnishing a quantified basis for evaluating the cost‑benefit of production‑line upgrades.

III. Application Scenarios

• Aviation AFP Defect Management on Mass-Production Lines (Core Use Case)

For the fuselage skin and wing-box large‑area tape‑laying lines, the multi‑sensor inspection scheme described in this paper is reused to pinpoint defect origins. Priority is given to optimizing the feed rollers and tow‑tensioning mechanisms, significantly reducing rework man-hours associated with gap and overlap defects and improving equipment utilization.

• AFP Offline Path Simulation and Process Planning

Embed the four types of error probability models. CAM Layup software simulates defect probabilities under various ply‑placement offsets, enabling the pre‑definition of an optimal fiber‑spacing configuration. This eliminates the need for multiple trial layups and shortens the process‑development cycle.

• AFP Equipment Development and Iterative Optimization of the Laying Head

Equipment manufacturers compare the slip‑induced fluctuations of various roller materials, tension systems, and guide slots, quantify the performance gains from hardware improvements, and develop a new compaction module with reduced lateral runout.

• Development of Defect Tolerance Standards for Aerospace Composites

Based on the defect size distribution obtained from simulation, we differentiate allowable gap thresholds for various ply orientations and load‑bearing regions, establishing a tailored acceptance criterion to prevent unnecessary scrapping resulting from a one‑size‑fits‑all approach.

• Development of an Online Closed-Loop Control System

Using real-time data from the on-roller tow camera as a feedback signal, a dynamic compensation control logic for the robot is established to achieve active error correction during the layup process, thereby constructing a “perception”… -  Prediction -  Correction “Smart” AFP Production line.

IV. Summary of the Article

(1) Core Research Findings

• AFP Gaps and overlaps arise from the combined effects of robot positioning, inter-roller tow slippage, incoming material width, and compaction-induced deformation. Among these, lateral slippage of the tow on the rollers is the primary source of defects, with its variation accounting for a significant portion of the total tow width. 22.1% ；
• The impact of positional variations on defects is twice as strong as that of geometric variations; the priority for process optimization is: slip suppression > robotic compensation > incoming material control.
• The four types of errors can be fitted to standard probability distributions, and Monte Carlo simulations predict gap distributions that closely match the measured heights, with a mean relative error of only 0.16% ；
• In industrial settings, visual sensing from compaction rollers can be used to acquire real-time measurements of tow misalignment, enabling online path correction for robots and actively mitigating excessive gaps.

(II) Engineering Application Value

For a long time AFP Defect rectification on the production line often fails to pinpoint the root cause with precision. This paper presents a practical, ready-to-deploy tool for multi-sensor synchronous measurement and statistical simulation. Enterprises can swiftly identify equipment bottlenecks and implement targeted optimizations of roll systems and tensioning mechanisms; meanwhile, during the process‑design phase, they can anticipate defect probabilities and strike an optimal balance between preset wire spacing and resulting gaps. /  Overlapping contradictions, significantly reduced. 32% Quality inspection and rework man-hours are optimized, enhancing the mass-production efficiency of large-scale aerospace composite components. The entire dataset and simulation algorithms are open-source and reusable, providing foundational statistical support for digital twins and closed-loop intelligent ply‑laying control.

(3) Current Research Limitations

Only for 6.35mm Thermoset prepreg flat‑lay unidirectional layup conditions; curved‑surface directional layups have not been validated (turning exacerbates tow slippage); thermoplastic in‑situ processing has not been undertaken. AFP , Multi-width tow testing; a complete hardware closed-loop compensation prototype has not yet been built.

(4) Future Expansion Directions

• Conduct tests on saddle-shaped and variable-curvature surface lay-up, quantifying the increase in slip under steering conditions.
• Expand 12.7/25.4mm Wide‑tape and thermoplastic tape placement multi‑condition database;
• Develop a camera -  A closed-loop real-time communication control system for robots enables online dynamic error correction.
• Coupled mechanical testing was conducted to establish a quantitative model relating defect size distribution to component strength reduction.

Image and Text Guide

Figure 1. AFP Schematic diagram of a layered structure in cross-section. Ideal AFP The mesostructure of composite materials includes adjacent drawn filaments. (a) , gap (b) and overlapping areas (c) 。

Figure 2. AFP Top-view schematic of the laminated structure. Gaps and overlaps arise from deviations in the traction path. a 、 b 、 c ) and differences in manufacturing processes (d) 。

Figure 3. AFP Schematic cross-section of a laminated structure—ideal tendon location (a) , Rod position under positional changes (b) , Rod position under geometric shape changes (c)

Figure 4.  causing gap and overlap defects AFP Manufacturing Process Variation Decomposition Diagram

Figure 5.  Display AFP A multi-sensor layout on the head CAD Model. The stacking direction is positive. X Axial direction.

Figure 6. AFP Layup motion. Program details (top); layup velocity distribution curve (middle); the current step being executed… 17 Next push-block action (bottom).

Figure 7.  Example frame captured by the camera. The measurement window is highlighted in red, and the center of the roller is indicated by a white dashed line.

Figure 8.  By LLS A (above) and LLS B (Bottom) Measured example contour curve.

Figure 9. A 、 B 、 C 、 D (From top to bottom). Measured manufacturing deviations that result in gaps and overlaps. Please note. X The axis uses a uniform scale. Mean and ± 1 Ϭ Values are marked in red, and the distribution‑fitting curve is marked in yellow. (For the meaning of the color codes in the legend, please refer to the online version of this article.)

Table 1 Summary of Manufacturing Variations Affecting Gap and Overlap Measurements

Figure 10.  Schematic diagram of the superposition of experimentally measured gap distribution and predicted gap distribution.

Table 2  Comparison of Measured Gaps and Predicted Gaps

Figure 11.  Probability Prediction of Defects in Adjacent Drawing Segments under Different Programming Gap Values

Figure 12.  When each of the four sources of variation is eliminated one by one, the gap (greater than 0.5 Probability forecast value (in millimeters)

Figure 13.  A single‑cable horizontal comparison between the measured gap and the gap expected from the tow cable’s lateral movement. The example data are derived from the tow cable, respectively. 18 (and its adjacent tow cable) 19 ) and tow cable 25 (and its adjacent tow cable) 26 ).

Figure A.14. This schematic illustrates the step-by-step procedure for simulating drag segments according to the algorithm described in this section. Dashed lines indicate the predefined drag spacing; each double-headed arrow represents a value drawn from the fitted distribution; and the figure shows two simulated adjacent drag segments along with the resulting simulated gaps.

Figure B.15. Example framework displaying a checkerboard scale