Utilizing machine learning and AI algorithms for inspection of airbags manufacturing processes

Abstract

Manufacturing processes of airbags products involve different and complex phases. An important one is the quality check using optical metrology methods both during the process and in its final phase. The aim of the present work is to study the advantages and limitations of utilizing a machine vision system for the above purposes. Thus, we report the development of a machine vision system for inter-phase and final quality check related to shape, dimensions, patterns, holes diameter, orientation, as well as the right sequence of processing airbag components. We utilized Basler/Cognex cameras with machine learning (ML)/artificial intelligence (AI) algorithms for inspection and CO2 lasers for cutting the components. Also, we developed (and further on optimized) in-house processes such as ultrasonic welding and dynamic positioning of components using cameras-guided robots. In order to validate the concept of the studies, several methods and tools have been utilized: virtual simulations programing tool (RobotStudio), CAD 3D mechanical design program (SolidWorks), Design of Experiments (DOE), Statistical Process Control (SPC), problem solving DMAIC (Define, Measure, Analyze, Improve, Control), as well as Process Failure Mode & Effects Analysis (PFMEA). The initial rejection rates (i.e., before applying these methods) were >30%, including false rejections. Overall, we reached an accuracy of 0.5 mm for dimensional measurements for textiles parts with different shapes in a field of view of 300 to 400 mm; rejection rate and falls rejects <1.3%; an appropriate stability and repeatability of the manufacturing process.