Introduction
In high-precision manufacturing, friction is a primary source of tracking error, particularly in CNC systems and robotic arms. Traditional friction compensation methods often rely on motor torque or current data to estimate friction forces. However, many industrial "black box" controllers do not provide access to this internal information, limiting the applicability of advanced models like the LuGre model.
This paper, published in IEEE Access, introduces a novel friction error model that operates solely on position information. By modeling friction error as a function of reference and actual positions, the researchers provide a practical solution for commercial CNC systems where torque data is unavailable.
The Position-Based Friction Model
The core of the proposed approach is a two-stage friction regime model that distinguishes between presliding and sliding behaviors. This conceptual framework allows the system to predict how friction translates into position error ($e_f$) without needing to calculate force directly.
Presliding and Sliding Regimes
Friction behaves differently depending on the displacement of the system:
- Presliding Stage: At micro-displacements (typically 1–7 microns), friction acts like a spring. The model represents the resulting position error as a linear gain applied to the theoretical position.
- Sliding Stage: As motion increases, the system enters the sliding regime. Here, the model incorporates nonlinear functions to account for the Stribeck effect—where friction decreases slightly at the onset of motion—and viscous friction.
Online Parameter Identification
To account for system changes over time, such as lubricant degradation or mechanical wear, the authors employed a Recursive Least Squares (RLS) algorithm for real-time parameter identification.
A significant innovation is the "pre-compensation" of the identification process. By predicting the friction error and subtracting it from the actual position data before the RLS calculation, the researchers prevent friction-induced "noise" from biasing the identification of the system's physical constants. This ensures that the identified transfer function accurately reflects the hardware dynamics rather than the disturbances.
Experimental Results
The model was validated using a custom experimental rig consisting of a Hiwin servo drive linear table and a high-resolution encoder. The results demonstrated a substantial improvement in tracking accuracy:
- Model Accuracy: The predicted friction error closely matched experimental observations, with a Mean Square Error (MSE) of $1.27 \times 10^{-7}$.
- Tracking Performance: In a 2Hz sinusoidal test, the proposed method reduced the maximum identification error from 0.0028mm (using standard methods) to 0.0006mm—a nearly five-fold improvement.
- Mean Absolute Error (MAE): The MAE was slashed from 0.0007mm to 0.0002mm.
Conclusion
The proposed position-based friction error model offers a robust and practical alternative to torque-dependent methods. By requiring only position data, it can be easily integrated into existing commercial CNC systems without hardware modifications. While future work will address handling discontinuous motion paths and simplifying initial nonlinear parameter tuning, this approach represents a significant step forward in friction compensation for closed industrial systems.
Summary updated