Process Phase Estimation and Deviation Detection for Manual Soldering Based on Motion Analysis

Manual soldering requires phase-dependent coordination of posture, hand movement, tool position, and visual attention. We propose a depth-camera framework that estimates the work phase and detects deviations using three-dimensional upper-body and hand features, task-related object positions, and gaze-related approximation features derived from facial orientation and head posture. The system estimates three predefined phases: preparation, active soldering, and cleanup. Windows with insufficient phase confidence are assigned to Uncertain Phase and routed to review rather than treated as deviation labels. In the evaluation, normal trials showed stable process sequences, whereas trials with scripted simulated unsafe-like movements produced local increases in the deviation score and review-required intervals associated with reduced phase-estimation reliability. These findings suggest that the framework may support retrospective safety-related assessment and the identification of process-inconsistent operations in seated manual soldering under controlled laboratory conditions. The bio-inspired contribution is a functional abstraction of phase-dependent perceptual-motor coordination into context-dependent engineering reference patterns and an uncertainty-aware review mechanism.