Predictive Maintenance Method Using Machine Learning; Comparing Classifiers

The focus of the world on the Fourth Industrial Revolution brings many benefits beyond ease of interaction between the customers and the manufacturers; with this revolution, quick turnaround times are required, and thus, machine reliability is of utmost importance. The challenge for the maintenance engineer thus becomes that when orders are placed and timelines are given, the plant must always be available to produce the ordered goods. Research has shown that the more data you have, the easier it is to use the statistics to accurately predict the time a piece of equipment will fail and accurately prescribe the component that will fail. The proposed system involves data acquisition and analytics, using different modelling techniques to accurately predict the likelihood of equipment failure and react before it fails, thus increasing equipment reliability. Many smart sensors have been developed and installed in plants in different industries, and thus, a lot of data is being collected. However, most of this data is being underutilised as, in most instances, as much as the data is available, it is used to figure out what happened and may happen due to what happened in the past. The most prevalent form of maintenance currently is predictive maintenance, which, through a number of configurations, produces a lot of accurate predictions as to when a machine will fail, the downside being that it does not give information as to what to do. Here comes prescriptive maintenance, which fills that gap and can be configured to act and help in decision support. Though not widely used, it is the future of maintenance.