Quality inspection in the industrial production field is experiencing a strong technological development that benefits from the combination of vision-based techniques with artificial intelligence algorithms. This paper initially addresses the problem of defect identification for circularly symmetric mechanical components, characterized by the presence of periodic elements. In the specific case of knurled washers, we compare the performances of a standard algorithm for the analysis of grey-scale image with a Deep Learning (DL) approach. The standard algorithm is based on the extraction of pseudo-signals derived from the conversion of the grey scale image of concentric annuli. In the DL approach, the component inspection is shifted from the entire sample to specific areas repeated along the object profile where the defect may occur. The standard algorithm provides better results in terms of accuracy and computational time with respect to the DL approach. Nevertheless, DL reaches accuracy higher than 99% when performance is evaluated targeting the identification of damaged teeth. The possibility of extending the methods and the results to other circularly symmetrical components is analyzed and discussed.
16 February 2022 in IEEE Transactions on Instrumentation and Measurement
This work proposes a framework for the optimization of the metrological performances of a system that measures the relative position and orientation between two rigid bodies using the ISO Guide to the expression of Uncertainty in Measurement as a reference for the analysis. The developed method identifies a non-linear measurement model of the instrument. The procedure uses the Monte Carlo method and the Design of Experiments techniques for determining the instrument uncertainty and the uncertainty sensitivity versus the geometrical and metrological instrument’s characteristics. The innovative result of the proposed approach lies in the derivation of non-linear numerical models of the measurement uncertainty and of the bias error components. A case-study related to a misalignment measurement system based on a universal joint is presented. The measurement uncertainty computed with the proposed method is compared with the one obtained in fit-to-purpose experiments performed with a robotic manipulator. An optimal calibration procedure is then developed to identify the parameters of the system minimizing the overall uncertainty. Results evidenced that the uncertainty of the misalignment, in the tested configuration, mainly depends on the uncertainty of the rotational measurements. The model has also been used to improve the design of the system by increasing the mounting and manufacturing tolerances of the most critical components.
This work investigates the uncertainty in 3D pose estimation of DenseFusion, a Deep Learning-based 3D Object Pose Estimation (3DOPE) framework. The accuracy and precision of object spatial localization is evaluated on the benchmark Yale Carnegie Berkeley-Video Dataset. The Average Euclidean Distance metric is employed as a figure of merit for the quantitative data analysis. Moreover, the Average Rotational and Translational Distances among the estimated and the ground-truth poses are analyzed. Results gathered on the YCB- Video Dataset are validated on a newly collected 3DOPE dataset, the ISS Dataset. The latter is populated by harnessing a 3D scanner, an RGB-D camera, and an anthropomorphic robot. Outcomes highlight an Average Euclidean Distance value of 1.8 mm on the best-processed item in the YCB-Video Dataset. Similarly, an Average Euclidean Distance value of 1.9 mm is achieved on the ISS Dataset. Eventually, the measurement uncertainty of the six components of the 3D poses is inspected. The latter assessment is performed on the ISS Dataset by relying on the ISO GUM [1].