Generative Model-Driven Sampling Strategy for High Efficient Measurement of Complex Surfaces on Coordinate Measuring Machines

Abstract

Coordinate measuring machine is widely used in precision measurement of industrial parts. However, the nature of the point-by-point probing characteristic limits its efficiency in the measurement of complex parts which normally requires dense sampling pints for fully evaluating the machining errors with high fidelity. To address this problem, this paper proposes a generative model-driven sampling strategy to reduce the number of the sampling points while maintaining the measurement accuracy. Specifically, the surface error reconstruction under sparse sampling is transformed as an image super-resolution task, which adopts a generative model to estimate accurate dense results from under-sampled data. A multi-scale neural network architecture is designed to achieve reconstruction, and the Fractional Brownian Motion is applied to synthesis large-scale simulated error datasets for model training. The generalized neural model could use sparse measurements to reconstruct global machining error, which dramatically reduces the sampling time and increases measurement efficiency. Both computer simulation and actual measurement are carried out to verify the effectiveness of the proposed method.