Subjective Assessment of Operator Responses for Mobile Defect Identification in Remanufacturing

In defect detection, the codification of operator expertise is vital for the successful deployment of machine learning as an assistant in the determination of the next manufacturing process step. While Artificial Intelligence, specifically within Computer Vision, has radically changed the human role in the automatic identification of defects, human intervention is likely to remain crucial in the verification of decisions made by Computer Vision algorithms. This study presents a subjective assessment of operator responses that have been compared to expert responses where significant subjectivity can exist regarding the nature and type of the next process step that is required. The case study in question is taken from the mobile phone defect detection within the remanufacturing process, a key evolving step in the emerging circular economy issue of extending phone life. Using state-of-the-art Natural Language Processing techniques for short text similarity tasks, the findings indicate that models incorporating contextual understanding and vocabulary awareness significantly outperform techniques with limited or no contextual understanding. This study employs Sentence-BERT, Word2Vec, and Dice similarity techniques to compare operator and expert responses, aiming to determine similarity/dissimilarity between them. This comparison helps identify levels of expertise and establishes new, improved guidelines for the use of AI in operator training.