When most people think of robots, they either envision humanoid servants or big one armed machines on automotive production lines. At MEUNGO we think about robots as being autonomous systems, things that happen by themselves. One of these things that could happen by itself is planning and dispatching preventive maintenance activities. Preventive maintenance prevents unscheduled downtime, it prevents having to stop production, call for service and potentially have unhappy customers.

So how do you prevent unscheduled downtime? The obvious answer is to plan downtime before the machine requires immediate service to continue operating. For this we need to know what the conditions are such that the machine no longer can operate and when these conditions will happen. So if we know when the machine is going to breakdown we know when we have to plan scheduled downtime. The rather dramatic scientific name being used for determining when the machine is going to breakdown is determining the Remaining Useful Life, the RUL of a machine. Determining the RUL helps you in predicting when you need maintenance and we do this to prevent unscheduled downtime. Predicting maintenance is input for your planning and scheduling tool to schedule downtime at a convenient time for you and your customers.

There are many methodologies to determine the remaining useful life of a machine. Almost all of them have one thing in common, they need data as input. Data about previous breakdowns of similar machines, parts, environment variables etc to use in some mathematical model to predict the RUL. Here one get can get pretty fancy pretty quick and use artificial intelligence, filtering techniques and stochastic processes to take into account as much as possible. One thing is critical, these advanced techniques typically need a lot of clean reliable data.

A more pragmatic and definitely a good intermediate step before invoking sophisticated RUL methodologies is to predict your maintenance using the condition of your machine. Let’s use your car as an example, you observe how many miles you have driven but you also observe if your steering wheel is shaking. The earlier you detect the shaking the earlier you can schedule a maintenance to check your wheels, brake rotors and brake caliper. So instead of predicting when your machine breaks down you act as soon as you observe an anomaly or reach a certain condition such as mileage or a date. You prevent unscheduled downtime but probably are sub-optimal because your RUL could have been months. Once again what is critical here is that this methodology also needs data and also continuous monitoring.

At MEUNGO we think you need to start collecting data now independent of what fancy technology you are going to use to predict the remaining useful life of your machines. Also start robotizing now, make sure conditions and conclusions on your data are translated into actions. For example make sure when your data tells you there is a problem your field service software auto-magically gets a work order. The trigger to when to auto-magically create a work order might change but the infrastructure to do so does not have to change. Once you do this you have the basis building blocks in place to continue improving your predictive maintenance.