Artificial Intelligence Based Robotic Automation of Manual Assembly Tasks for Intelligent Manufacturing

Increasing product customization and shortening product life cycles in an ever-changing world is challenging for automation. This is especially true for assembly tasks, requiring a high level of perception, skill, and adaptability. With the rise of smart manufacturing, intelligent manufacturing, and other aspects related to Industry 4.0, the hurdles for automation of the aforementioned tasks are getting reduced. Especially Artificial Intelligence (AI) is expected to enable smart and flexible automation since it is possible to deduct decisions from unknown multi-dimensional correlations in sensor data, which is critical for the assembly of highly customized products. In this research paper, three different conventional and AI-based glue detection models are proposed with the target to automate a gluing process in a manual assembly of highly customized products in a batch size one production scenario. A conventional, one-dimensional rule-based model, and two hybrid models using a support vector machine image classifier (SVM) and either Tamura features or convolutional neural network (CNN) feature extraction are presented and compared. The obtained results demonstrate the efficiency and robustness of AI-based algorithms, as the CNN and SVM hybrid model outperforms the other two approaches achieving a prediction accuracy of >99% at the fastest classification speed.