Abstract: The expectation and variances of the three distance function based models is derived. Examples demonstrate that the linear model tends to underestimate the workpiece deviation while the two-sided quadratic model provides the most accurate precision prediction. However this precision gain is at the cost of complex matrix calculation. In comparison, the linear model is the most simple and straightforward one. To strike the right balance between precision gain and model complexity, the linkage between the relative error and the locator error for the three models is built. And this leads to model selection method. Given a relative error level (i.e. to ensure the required precision requirement is met), the proposed method makes the model selection based on the covariance of locator errors. Several examples verified the validity of the proposed method.

Key words: Error statistical characteristic, Signed distance function, Workpiece positioning analysis, flexible carbody, high-speed EMU, suspended system, wheel profile wear