Signal enhancement for OCT weld-depth monitoring via adaptive maximum intensity projection

Optical coherence tomography (OCT) has been increasingly explored for real-time weld depth monitoring in laser welding due to its high axial resolution and non-contact operation. In practical welding environments, OCT signals are strongly degraded by vapor, spatter, and plasma, leading to intermittent visibility of keyhole-bottom echoes and unreliable single A-line measurements. Maximum intensity projection (MIP) provides a computationally efficient approach for enhancing signal robustness by aggregating OCT A-lines over time, but its performance critically depends on the choice of the projection window length. Larger windows improve signal-to-noise ratio at the expense of temporal responsiveness and potential motion-induced blur, while small windows often fail under severe signal dropout. In this work, we analyze this trade-off and formulate a closed-form adaptive window selection rule that accounts for A-line sampling rate, welding speed, and required signal quality. The proposed strategy updates the projection window online without iterative optimization and is suitable for real-time implementation. Experiments on steel and aluminum welding demonstrate that the adaptive MIP achieves penetration depth accuracy comparable to large fixed-window processing while preserving high update rates, enabling robust and responsive weld depth monitoring.