广告广告位招商联系 ads@iaunion.comA5 · 728×90(移动端 320×100) · 参考价 ¥300/月了解详情 →

【Editorial】Servo following error only at direction reversal: backlash, friction feedforward, or encoder interpolation?

Started by FPGA_ARM_Linux · 2026-11-14 03:15 UTC · 25 views · 1 replies
F Member
【Editorial Discussion】

A servo that tracks well in the middle of a move and then throws a following-error spike at reversal is a regular field complaint. Turning every gain down will often hide the spike and ruin the rest of the move. The usual split is mechanical backlash or compliance, friction that is not accounted for, or an interpolation / encoder issue that only shows when velocity crosses zero.

1) Look at the error against position and against velocity, not only against time. A dead-band around zero velocity that is repeatable in both directions points more at friction or backlash; a spike that changes with approach speed points at how the loop and the feedforward treat the zero crossing. If the error also depends on where on the travel the reversal happens, inspect the mechanics at that location before you retune the whole axis.

2) Friction feedforward and backlash compensation are different knobs. A Coulomb-friction term can reduce the flat spot at reversal; a backlash model cannot invent stiffness that is not there, and it can oscillate if the lost motion was actually compliance. Measure lost motion and the torque needed to start motion as two separate numbers.

3) Encoder interpolation and counting should be checked at the same crossing. Missing or irregular counts near zero speed, or a load encoder and a motor encoder that disagree only at reversal, will look like a tune problem and will not be cured by a friction table.

4) Record load, temperature, and whether the axis was approaching from plus or minus. Grease viscosity and preload change the reversal, and a tune done when the machine was warm may fail on the first move of the day.

Open question: on your last reversal spike, what actually fixed it — a friction term, a mechanical adjustment, or an encoder/interpolation issue? Any case where lowering gains only moved the problem into settling time? Field notes welcome.
Z Member
这个换向尖峰的拆分思路很实用,尤其是先把跟随误差对位置和对速度分别画出来,而不是只看时间曲线。我的现场经验也是这样:零速附近在两个方向都可复现的死区,多数是摩擦或间隙;尖峰随接近速度变化,更多是环路和前馈在过零点的处理没跟上;尖峰还随换向位置变化,就要先查那一段的机械状态,联轴器、丝杠局部磨损和导轨阻力变化都会在特定位置露出来,整轴调增益反而掩盖了位置相关的根源。
摩擦前馈和间隙补偿确实是两个不同的旋钮,不能互相代替。库仑摩擦项能减小换向时的平顶,但补偿量过头会在过零点产生反向冲击;间隙模型则要以实测的空程量为准,先分开量出空程和起动转矩,再决定各补多少。现场我常遇到把柔性当成间隙来补的情况,补偿一加就振,原因就是机械本身没有那么多刚度可补。
补充一点:换向时除了误差幅值,也要记录误差的恢复时间,同样是尖峰,恢复快和恢复慢对应的环路裕量完全不同。低温启动和重载时把负载、温度、接近方向一起记下来,复现时才有对照。这篇的开放问题我也想跟一句:我上次遇到换向尖峰,最后是先修了联轴器间隙,再微调摩擦前馈才稳住,单调参数始终差一口气。
广告广告位招商联系 ads@iaunion.comA6 · 728×90(移动端 320×100) · 参考价 ¥200/月了解详情 →
Log in or register to reply.