HUMAN-AI INTERACTION BASED ON COMPUTER VISION AND DATA ANALYTICS FOR EXCAVATOR MAINTENANCE

Sammanfattning

With the rapid advancement of AI technology, Artificial Intelligence (AI) is increasingly integrated in real-life work environments. Computer Vision-based AI systems are one example of AI systems that can be used to automate visual inspection tasks and help decision-making in different processes such as machine parts maintenance inspections. Human-AI Interaction (HAII) techniques and design choices can be considered during the development of these systems to improve the overall user experience. Our work is carried out in a partner industry and this thesis describes the process of integrating a trained AI model, designed to identify and segment the sprocket teeth of an excavator, with a HAII centered user interface. We gather HAII design features through interviews and a survey with participants of different roles in our partner industry, selecting the most valuable features based on their use case, and implementing these features in a front-end interface. Our system is then evaluated both in terms of user acceptance and ML performance. In this research, we find that HAII integration can improve the interaction between humans and AI by making a system simple and easy to use. We also find that a guided image capture mechanism supported by an image quality assessment method can involve users in the model operations stage of the ML life cycle, but that the HAII integration, in this case, does not show a measurable impact on MLperformance.