Trajectory optimization for composite fiber placement on a dual-arm platform with LSTM-based sequential thermal evaluation

In composite material manufacturing, robotic composite fiber placement (CFP) has significantly improved production efficiency and lamination quality. However, the thermal state of a preheated fiber patch during transfer remains highly sensitive to process variables, including transfer duration, end-effector velocity, gripping orientation, and ambient airflow. Most existing trajectory planning algorithms do not adequately account for history-dependent temperature evolution, making it difficult to meet the stringent requirements of thermally sensitive placement tasks. To address this issue, this paper proposes Sequential Thermal-aware CHOMP (ST-CHOMP), which integrates covariant trajectory optimization with an LSTM-based sequential thermal evaluator. The proposed evaluator predicts transient temperature variations along candidate trajectories, and the resulting thermal loss is incorporated into the CHOMP framework together with collision avoidance and trajectory smoothness. Simulation and physical experiments on a dual-arm placement platform demonstrate that ST-CHOMP improves the thermal retention of the fiber patch compared with benchmark trajectory planners, while maintaining collision-free and smooth motion. Post-layup inspection and procedure-specific mechanical testing of representative laminates further provide supporting evidence that improved thermal retention is associated with more compact visible defect contours and better mechanical responses under the investigated conditions.