Implicit residual approximation for multi-sensor data fusion in surface geometry measurement

Multi -sensor measurement of complex products provides the opportunity for real-time digitising of the product geometry, thus becoming an enabling technology for the digital twin establishment of the manufacturing process. Data fusion of multi -sensor measurement results could improve the accuracy and efficiency of measurement because of the complementary characteristics of different sensors. The classical multi -sensor fusion method, Residual -approximation (RA) has been developed and demonstrated to be an effective solution in fusing heterogeneous point clouds. However, existing RA -based methods rely on the explicit z -direction residual function, which is inapplicable for complex surfaces with varying normals or implicit functions that widely exist in the modern manufacturing industry. Therefore, this research proposes an Implicit Residual Approximation (IRA) method that can represent the residual between different data sets implicitly as subresidual models. By constructing local clusters of measurement data, the complex residual function in the original space can be conveniently represented by the Gaussian mixture of the estimated sub -residual models. Both the simulation case and real measurement experiments are carried out to show the effectiveness of the proposed method. The experimental results demonstrate the superior performance of the proposed compared to the existing RA -based in both residual modelling and data fusion.