Context
A servo axis follows its command well: position and speed stay close to what was asked for even as the mechanics change underneath. What changes instead is the effort required. Shifts in load, changes in friction, and process forces such as cutting or clamping show up in the axis current rather than in the motion itself. Reading that signal is easy; separating it is not. Most of the current at any moment is simply the cost of moving the axis along its commanded path. The part worth knowing is what is left over, and isolating it needs a reference for how much torque the motion alone should have taken.
What we tested
We built a model that estimates axis current, the quantity proportional to torque, from the motion signals a drive already holds: commanded and actual velocity, acceleration, and position. No additional sensor is involved. The gap between the measured current and the estimate is then the torque the motion command does not account for. The model is built on the family of lightweight machine-learning techniques we work with across our projects (reservoir computing). Validation used publicly available machining data.
What it shows
Axis current could be reproduced on cutting runs the model had never seen. What is left over is the quantity of interest: it is where cutting and clamping forces, changes in friction, and unexpected load appear, without a dynamometer in the machine.
This may apply to your equipment
If you build or run machinery where the process force matters but a torque sensor or dynamometer is impractical, such as machine tools, clamping axes, presses or robots, the same idea may be worth exploring. The signals it needs are ones your drive already reports. A short conversation is normally enough to tell whether your case fits.