Kris Isfeld, Rithmik Solutions; Pat Simpson, Consultant
Maintenance practices have evolved over time, ranging from “fix it when it fails” to Condition-Based Maintenance (CBM). CBM leverages past failure and maintenance experience alongside a current understanding of the equipment condition to schedule maintenance that maximizes uptime and minimizes cost. CBM is predictive in that understanding past events, the equipment’s current condition and how it has changed over time aids in the prediction of impending failure modes. Condition Monitoring (CM) has become the go-to method to obtain equipment condition insights. CM continuously streams data from sensors on equipment, allowing maintenance folks to analyze data that otherwise would be obtained by local equipment inspection to sense heat, vibration, pressure, visual damage, sound, etc. The goal behind CBM is a good one. CM provides value both by continuously exposing indicators of the equipment condition and by keeping an accurate history. However, no matter how well-intentioned maintenance folks are, they struggle to keep up with every piece of equipment and vast amounts of data. Machine Learning (ML) algorithms are predictive by nature; they aim to predict the actions an expert will take. In essence, they learn the decision framework of the expert. Autonomous cars provide a clear example. AI monitors the current and changing conditions via images of the surroundings, along with the speed and direction of the vehicle, and then leverages what experts have done in the past to predict the action the expert would take in the current circumstances. ML is a perfect fit to bolster predictive maintenance. ML augments our maintenance experts by learning their decision framework and in turn informs them where to focus their time. This amplifies the effectiveness of human experts and brings us closer to truly maximizing uptime and minimizing cost.
