Scholarlink

<p>While multi-phase bumpless feedforward (BFF) controllers can adapt to varying dynamics in high-precision motion, their manual parameter tuning is labor-intensive and prone to instability, especially in multiple-input-multiple-output (MIMO) systems. This work addresses the lack of automated solutions by proposing a data-driven tuning methodology integrated within a feedback-feedforward framework. First, the gradient of the BFF signal with respect to its parameters is derived, revealing its dependence not only on the current reference signals but also on the final values of the reference and BFF signals from the preceding phase. Subsequently, the MIMO system’s sensitivity is estimated with impulse response through sequential injection of step inputs as artificial disturbances. The methodology avoids the complex inversion of switching controllers by estimating system sensitivity via sequential step disturbances. Within each optimization cycle, the cost function gradient is then iteratively updated using data from a single normal tracking experiment. This approach solves for optimal parameters with minimal experimental overhead. Experimental validation on an XY stage of a wire-bonding machine demonstrates that the proposed method achieves at least a 55.7% improvement in tracking performance over existing data-driven tuning methods for non-switching feedforward control.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Data-driven tuning for bumpless feedforward in MIMO system with application to wire-bonding machine

  • 期刊论文

  • 发表日期：

    2026年7月16日

查看全文
  • Yifan Xu,
  • Silu Chen,
  • Zhuang Xu,
  • Chi Zhang,
  • Weizhen Wang,
  • Dunant Halim

摘要

While multi-phase bumpless feedforward (BFF) controllers can adapt to varying dynamics in high-precision motion, their manual parameter tuning is labor-intensive and prone to instability, especially in multiple-input-multiple-output (MIMO) systems. This work addresses the lack of automated solutions by proposing a data-driven tuning methodology integrated within a feedback-feedforward framework. First, the gradient of the BFF signal with respect to its parameters is derived, revealing its dependence not only on the current reference signals but also on the final values of the reference and BFF signals from the preceding phase. Subsequently, the MIMO system’s sensitivity is estimated with impulse response through sequential injection of step inputs as artificial disturbances. The methodology avoids the complex inversion of switching controllers by estimating system sensitivity via sequential step disturbances. Within each optimization cycle, the cost function gradient is then iteratively updated using data from a single normal tracking experiment. This approach solves for optimal parameters with minimal experimental overhead. Experimental validation on an XY stage of a wire-bonding machine demonstrates that the proposed method achieves at least a 55.7% improvement in tracking performance over existing data-driven tuning methods for non-switching feedforward control.