Even today, many maintenance activities are still done manually because maintenance is one of the most difficult areas to be automated in manufacturing. Many technicians spend their time on non-technical activities such as retrieving instructions from manuals. If AI (Artificial Intelligence) can alleviate some of these tasks, the time to diagnosis and repair can be shortened. However there are limited works about the effects of using AI during maintenance activities on a technician's cognitive load. Therefore, as an initiative, we conducted a pilot experiment with 10 participants to analyze the effects of the AI-based support system on diagnosis tasks in the manufacturing. In the experiment, participants were divided into two groups: the group used an AI-based support system and the other group used a Fault Tree (FT) based support system; two groups' mean task completion time and task load of participants using NASA Task Load were measured. According to the experiment results, the group which used the AI-based support system to diagnose the model completed task 53% lesser time than the group which used the FT-based support system. In addition, participants who used the AI-based support system reported relatively lower task loads compared to participants who used the FT-based support system. This experiment results imply that maintenance time and a variability can be reduced if an AI-based support system supports maintenance technicians

Typically, maintaining a machine requires two different distinct tasks: to select which components to focus their attention; and the subsequent task is to check, repair or replace the selected components. For both tasks, the experience of technicians plays a critical role. An experienced technician is likely to select fewer components and requires less time for the subsequent task compared to an inexperienced technician. As a result, the maintenance time will be varied depending on the experience of the technicians. Extant research for maintenance has predominantly used exponential distribution family for modeling primarily because of its analytical tractability but at the cost of fidelity and inability to capture important characteristics such as technicians' experience. With the growing adoption of networked sensors based on Internet of Things (IoT), big data, and real-time machinery diagnostics using artificial intelligence it is imperative to develop models with better fidelity for maintenance operations. Therefore, in this paper, we explore a model based on using the negative-hyper geometric distribution for maintenance time that varies based on the technicians' experience. Our proposed approach requires more inputs such as (1) number of components, (2) number of components not in working state (3) technician's experience level, and (4) time to fix a component based on the technicians' experience. For instance, input for (2) could be obtained from IoT sensors and diagnostics. We study the efficacy of the proposed model using computer simulations and statistically characterize the possible impact of technician experience on the parameters of the maintenance distribution.

No abstract