Real-Time Evaluation of Molten-Pool Dynamic Stability in Laser-Directed Energy Deposition Based on Interframe Similarity Monitoring

• The Molten-Pool Inter-frame Similarity (MPIFS) based on frame-wise morphology comparison was firstly induced as a novel metric to quantify dynamic stability of molten pool. • A lightweight real-time monitoring framework integrating YOLOv8n and Perceptual Hashing (P-Hash) algorithm is established, enabling real-time quantitative evaluation of molten-pool stability. • The MPIFS achieved 3 × higher sensitivity to molten-pool fluctuations than static geometric parameters (e.g. molten-pool width/height), significantly enhancing anomaly detection capability. • A strong correlation among process, molten-pool stability, and microstructure was confirmed, and the reasonable low power enhances MPSF, yielding smooth interfaces and uniform fine grains. The dynamic stability of the molten pool during laser-directed energy deposition (L-DED) critically affects the forming quality and material properties. However, existing monitoring methods primarily focus on the static geometric features of the molten pool, such as width, height, area, and geometric center, yet fail to capture its instantaneous morphological evolution. This work established a lightweight real-time monitoring framework that integrates YOLOv8n and the Perceptual Hashing (PHash) algorithm to monitor the dynamic stability of the molten pool in L-DED online. Furthermore, the molten-pool interframe similarity (MPIFS) is developed as a novel metric to quantify dynamic stability. The experimental results show that the YOLOv8n-PHash framework achieves a processing speed of 85 FPS (3.15 × faster than U-Net) and reduces computational latency to 10.87 ms/frame, which is 6 × faster than the Structural Similarity (SSIM), satisfying industrial closed-loop control requirements. The MPIFS metric shows three times higher sensitivity to molten-pool fluctuations than the static geometric parameters, with a standard deviation of 2.8% for MPIFS versus 0.15%-0.98% for the width and height. This enhanced sensitivity significantly improves the anomaly detection capabilities. In addition, a strong correlation among the process, molten-pool stability, and microstructure was confirmed. An appropriately low laser power was shown to improve MPIFS stability, resulting in smooth interfaces and uniform fine grains. This work provides a novel approach for the online monitoring of molten-pool stability and microstructure prediction in L-DED additive manufacturing.