Predicting “Maintenance Priority” with AI

In Tiipras oil refineries, an average of 100 thousand maintenance requests are created annually for more than 140 thousand pieces of equipment. These requests are prioritized manually by chief experts with over 25 years of experience and classified as urgent or planned. If maintenance requests that need to be solved urgently in the refining industry are mislabeled and delayed, they may cause process upsets leading to health & safety hazards, environment problems or big asset damage. To minimize this risk, we think that supporting the decision mechanism with algorithms and cross checking/replacing human decisions by using today's AI technologies is the right approach that reduces the possibility of human error. In this study, our main goal is to automate maintenance prioritization process with supervised and unsupervised ML algorithms, deploy an AI system and achieve high accuracy. Our study was carried out basically in 4 main steps: • Exploratory Data Analysis • Clustering - Feature Addition - Feature Selection • Model Selection and Results • Additional Studies With this study, we aim to explain our AI study, share our experience with other partners that have similar needs and provide them an effective tool and systematic approach about management of transition from human to machine with a real industry case. We believe that the transfer of priority selection process from human to algorithms ensure consistent decisions, reduce costs and tolerate experience losses.