On Machine Measurement of Probe Error Research Based on Bayesian Network

Liwen Yan, Huaiqiang Wu2016 · DOI

Abstract

Probe calibration is carried out under specific conditions; most of the error caused by the change of speed parameter has not been corrected. In order to reduce the measuring error influence on measurement accuracy, analyze the relationship between speed parameter and probe error, and using Bayesian network to establish the model of probe error. Model takes account of prior knowledge and sample data, with the updating of data, which can reflect the change of the errors of the probe and constantly revised modeling results.

Introduction

In modern manufacturing, "on-machine measurement" allows a CNC machine to use a probe (like the Renishaw OMP60) to measure a workpiece while it is still on the table. This data is used to modify the machining process or compensate for errors in real-time. However, probe calibration is typically performed under a very specific, static set of conditions. In practice, the probe moves at different velocities ($V_1$ for fast travel, $V_2$ for the actual measurement) and from different approach distances ($D$). These "Direct Computer Control" (DCC) parameters introduce mechanical deformations and trigger delays that change the probe's "effective diameter." Traditional fitting models often struggle to capture the complex, probabilistic relationship between these variables.

Methods

The authors propose using a Bayesian Network to model the probe error. Unlike a simple linear regression, a Bayesian Network uses a Directed Acyclic Graph (DAG) to represent variables as nodes and their causal dependencies as edges.

Bayesian Network structure for probe error modeling

The model identifies three parent nodes—Fast Moving Velocity ($V_1$), Probing Velocity ($V_2$), and Approach Distance ($D$)—which all influence the child node, the Effective Probe Diameter ($d$). The network calculates the joint probability density of these variables, incorporating prior knowledge and updating itself as new sample data is collected. Using the Bayesian formula and the maximum likelihood principle, the system can predict the most likely effective diameter ($d$) for any given set of motion parameters.

Experimental Results and Insights

The researchers generated experimental data using a Dalian Machine Tool (FANUC0i-MD) and a Renishaw OMP60 probe. They collected 18 sets of test data by varying the DCC parameters to observe how the functional diameter changed.

Experimental results showing the impact of velocity and distance on functional diameter

The experimental results reveal that while fast moving velocity ($V_1$) has a relatively flat effect, the actual measurement velocity ($V_2$) has a significant, near-linear impact on the functional diameter. The Bayesian model achieved an average relative error of only 7.2%, with an overall prediction accuracy of 83.3%. In a practical test measuring a 25.0030 mm standard ball, the error was reduced from 25 microns (before compensation) to 14 microns (after compensation).

Conclusion

The Bayesian approach is superior to traditional fitting because it can constantly revise its modeling results as more data is gathered, making it more robust for varying shop-floor conditions. While the paper demonstrates a clear improvement in accuracy, future work should explore larger datasets and account for environmental factors like thermal expansion or probe wear.

Summary updated

Abstract

Probe calibration is carried out under specific conditions; most of the error caused by the change of speed parameter has not been corrected, In order to reduce the measuring error influence on measurement accuracy, analyze the relationship between speed parameter and probe error, and using Bayesian network to establish the model of probe error. Model takes account of prior knowledge and sample data, with the updating of data, which can reflect the change of the errors of the probe and constantly revised modeling results.
BibTeX
@misc{yan2016machine,
  author = {Liwen Yan and Yan Yan and Huaiqiang Wu},
  title = {{On Machine Measurement of Probe Error Research Based on Bayesian Network}},
  year = {2016},
  doi = {10.2991/icmit-16.2016.78},
}

Loading related content...

Figure