Detection of fluid level in bores for batch size one assembly automation using convolutional neural network

Increased customization and shortening product life cycles pose a challenge for automation, especially in assembly. In combination with the nature of assembly tasks, which may require high level of perception, skill, and logical thinking, these tasks are often conducted manually, especially in certain industries (e.g. furniture, power tools) or small and medium-sized enterprises. One of such tasks is the liquid level monitoring in gluing processes. Existing non-manual solutions are based on conventional and less flexible algorithms to detect the current liquid level. In production environments with highly individualized products, a need for more performant models arises. With artificial intelligence (AI) it is possible to deduct decisions from unknown multidimensional correlations in sensor data, which is a key enabler for assembly automation for products with high degree of customization.

In this paper, an AI-based model is proposed to automate a gluing process in a final assembly. Images of a gluing process are taken with a camera and a convolutional neural network is used to extract images features. The features are applied to train a support vector machine classifier to identify the liquid level. The developed model is tested and validated with a Monte-Carlo-simulation and used on a demonstrator to automate a gluing process. The developed model classifies images of liquid levels with over 98% accuracy. Similar results are achieved on the demonstrator.