Equipment failure imposes an enormous burden on industry. It is estimated that unplanned downtime reduces plant productive capacity by between 5 and 20 percent and costs industrial manufacturers $50 billion annually. The cost to repair or replace equipment may be significant, but the true cost of unplanned equipment failure is its consequences. The effects of a failed machine ripple through disrupted downstream operations and heighten exposure to safety hazards and subsequent failures.

Machine lifetime can be extended, and the costs of unanticipated failure can be mitigated by an efficient maintenance strategy. In a cross-sector survey of more than 450 companies, more than 70 percent reported they lacked awareness of when equipment was due for maintenance. A common solution is to routinely replace parts and service equipment at planned intervals. However, all maintenance incurs cost, and routine tasks may be wasteful if performed when they are not warranted. In contrast, predictive maintenance seeks to target maintenance to need by using data-driven analysis to assess equipment condition.

In this post on the AWS for Industry Blog, we will build a machine learning model to address a predictive maintenance task. We will demonstrate how the application of SpeedWise ML (a commercial AutoML solution) can be used to predict the remaining useful life (RUL) of turbofan jet engines. The objective is to quickly propose a predictive model that, using sensor information describing the engine’s present and past performance, can forecast how many timesteps remain until the engine will fail. When applied to a fleet of engines, the model could help the operator’s direct maintenance to those engines most prone to failure, improving the efficiency of a maintenance routine and, most importantly, the fleet’s reliability.