Smart factories have led to the introduction of automated facilities in manufacturing lines and the increase in productivity using semi-automatic equipment or work auxiliary tools that use power sources in parallel with the existing pure manual manufacturing method. The productivity and quality of manual manufacturing work heavily depend on the skill level of the operators. Therefore, changes in manufacturing input factors can be determined by analyzing the pattern change of power sources such as electricity and pneumatic energy consumed in work-aid tools or semi-automatic facilities used by skilled operators. The manual workflow can be optimized by modeling this pattern and the image information of the operator and analyzing it in real time. Machine learning operations (MLOps) technology is required to respond to rapid changes in production systems and facilities and work patterns that frequently occur in small-batch production methods. MLOps can selectively configure Kubeflow, the MLOps solution, and the data lake based on Kubernetes for the entire process, from collecting and analyzing data to learning and deploying ML models, enabling the provision of fast and differentiated services from model development to distribution by the scale and construction stage of the manufacturing site. In this study, the manual work patterns of operators, which are unstructured data, were formulated into power source consumption patterns and analyzed along with image information to develop a manufacturing management platform applicable to manual-based, multi-variety, small-volume production methods and eventually for operator training in connection with three-dimensional visualization technology.
