Artificial intelligence for predictive maintenance

Abstract

Maintenance constitutes an important share of modern industrial activities. Reliable operations rely on the adequate application on maintenance. However, in the present competitive environment, maintenance processes must be optimized so that they will be performed only when needed, otherwise resources will be needlessly wasted. This is in contrast to the conventional approach where maintenance is scheduled according to a time plan regardless of it is needed or not. This paper presents the application of artificial intelligence to create a model that can successfully predict the condition of a machine in terms of the probability of failure occurrence. This work uses a synthetic dataset that reflects a realistic scenario where sensors are connected to a machine to monitor its health condition and record failure incidents. The dataset consists of 10,000 records. Each record consists of five numerical measurements: air and process temperatures, machine rotational speed and torque and finally a measurement of machine wear. This in addition to the type of product the machine is producing makes six input variables. The output response was considered as the state of machine failure that was represented as true or false. An artificial neural network was trained and it successfully predicted the state of the machine. The capabilities and limitations of applying artificial intelligence were discussed. In addition, a brief overview of other predictive techniques was given.