<p>In high-end CNC machining, the precision and stability of machining processes are determined by the performance of ball screws feed systems. In particular, wear-induced degradation of friction torque (FT) for ball screws can severely affect positioning accuracy, making online FT monitoring essential for enhancing machining performance. However, existing monitoring approaches cannot capture the position-dependent nature of FT and heavily rely on sophisticated architectures, which impose high computational demands and slow update speeds. To address these challenges, this paper proposes a computationally efficient shallow feature-driven dynamic friction torque monitoring method (SF-DFTMM) for indirect, in-situ monitoring of FT degradation. The method establishes a mapping between high-frequency multi-sensor signals and FT values at various nut positions through a structure comprising three modules: node generation and initial learning (NG-IL) for lightweight model construction, fast dynamic update (FDU) for efficient parameter adaptation, and timely condition monitoring (TCM) for online prediction. An innovative dual-screw torsional loading test rig was developed to reveal the wear process of ball screws. A 500&#xa0;h degradation test demonstrated the efficacy of the proposed method, which achieved a mean absolute percentage error below 3.62% for FT prediction across different positions and degradation stages without disassembly, and reduced the model update time to 38.8% of the initial training duration. The proposed method achieves timely, position-dependent modeling of FT degradation with high accuracy and low computational demand, offering a novel solution for intelligent monitoring of CNC motion component health.</p>