An embedded machine vision system for an in-line quality check of assembly processes

In the context of the Industry 4.0 and Smart Manufacturing paradigm, this paper proposes a machine vision system for a flexible, precise and low-cost in-line geometric inspection of assembly processes. As a case study, the system has been targeted to the catalytic converter assembly process, in order to be easily integrated in the consolidated manufacturing flow. The system is based on developed algorithms to be applied on the image of the interfaces with the complete exhaust system the catalytic converter will be assembled into. An image segmentation procedure is described to robustly identify the region of interest (ROI) in the image. Afterwards, a geometrical model is proposed to detect any possible geometrical defects due to planar and/or rotational shifts of the interfaces around their expected positions. For the sake of validation, the proposed system has been implemented on a Raspberry Pi 3 Single Board Computer (SBC). It showed a sub-millimeter precision for planar movements and a maximum error in detecting the rotation angle lower than 1 degree, respectively. The modularity of the proposed approach makes it suitable to be realized also on different computational platforms, such as the modern heterogeneous System-on-Chips (SoC) hosting a general purpose microprocessor and a Field Programmable Gate Array (FPGA) on the same chip. Indeed, the most time-consuming computational steps can be efficiently realized on the FPGA, exploiting the parallel computing capability offered by a hardware implementation, thus accelerating the overall computation