Spatio-Temporal Adaptive Sampling for effective coverage measurement planning during quality inspection of free form surfaces using robotic 3D optical scanner

Abstract

Abstract In-line dimensional inspection of free form surfaces using robotic 3D-optical scanners provide an opportunity to reduce the mean-time-to-detection of product quality defects and has thus emerged as a critical enabler in Industry 4.0 to achieve near-zero defects. However, the time needed to inspect large industrial size sheet metal parts by 3D-optical scanners frequently exceeds the production cycle time (CT), consequently, limiting the application of in-line measurement systems for high production volume manufacturing processes such as those used in the automotive industry. This paper addresses the aforementioned challenge by developing the Spatio-Temporal Adaptive Sampling (STAS) methodology which has the capability for (i) estimation of whole part deviations based on partial measurement of a free form surface; and, (ii) adaptive selection of the next region to be measured in order to satisfy pre-defined measurement criterion. This is achieved by first, modelling spatio-temporal correlations in the high dimensional Cloud-of-Points measurement data by using a dimension reduced space-time Kalman filter; then, dynamically updating the model parameters during the inspection process by incorporating partial measurement data to predict entire part deviations and adaptively choose the next critical region of the part to be measured. The developed STAS methodology enhances the current free form surface inspection models, which are mostly based on spatial analysis; into spatio-temporal model, which uses (i) the spatial analysis to model part deformation; and, (ii) temporal analysis to model autoregressive behaviour of the manufacturing process for prediction of next part deviations. This provides capability to predict the whole part deviation based on partial measurement information and consequently reduces measurement cycle time. The industrial case study using a robotic 3D-optical scanner for the measurement of an automotive door inner part demonstrates the STAS methodology, which resulted in (i) a 3 Sigma error of prediction of whole part deviations within 0.27\u202fmm based on measurement of 33% of the part surface; and, (ii) a corresponding CT reduction of 42.2% from 510.5\u202fs required by current best practice to measure the whole part to 295.18\u202fs required to partially measure the part.