This piece is a simple example of how predictive maintenance can be done with machine learning. This is not something you have to be a tech to understand. It's very simple and fun, and it's easy to learn. 

Let's start with something easy. Let's say we have two devices that can detect vibrations. Most of the time, vibration monitors are used on things that move, like motors, fans, pumps, gearboxes, and so on.

Let's say we've already put our two shaking sensors on an electric motor. We can measure how much this motor shakes with these instruments.

When the motor is running normally, it vibrates normally. But if something is wrong with the motor, we'll feel a strange shaking. Pretty simple, right?

Now the question is, how can we tell if the sound is normal or not? How should we think about a strange vibration?

Well, since Artificial Intelligence (AI) has come a long way in recent years, we can use simple machine learning methods to figure this out. But don't be scared by words like AI and machine learning. The idea is pretty simple! I'll show you how to do it.

Again, I'm trying to figure out what should be called an unusual vibration by using a simple machine learning method.

I can do this by measuring the sound for sensor A at a random time and writing it down. Let's say that at this point, the shaking measured by sensor A is 2.

Next, I can measure how sensor B is shaking at the same time. Let's say that sensor B is vibrating at a rate of 3. 

So at one point, I measured how much both devices were shaking. Sensor A had a value of 2, and sensor B had a value of 3. It doesn't matter what units are used to measure vibrations in this case.

I'm going to do this test again. This time, I see that sensor A has a value of 3 and sensor B has a value of 5. As you can see, the numbers are a bit higher, but they are still in the same ballpark.

One thing to keep in mind is that it doesn't matter when I take these measurements. I mean, I could take these measurements at any time.

But in this case, the only thing that counts is that both sensors measure at the same time. That means I need to measure the numbers of both sensor A and sensor B at the same time.

Normal method of operation
Ok, I'll do this process again and again to measure the numbers of both sensors. By doing this, I can find out more about how the motor shakes under normal conditions. Now, you may be wondering, why do I measure these numbers or data points?

I measure these numbers so that I can make some kind of model for how the motor vibrates when it is working normally.

Using these data points, I can now get a pretty good idea of how much the motor might shake when it is working normally and there are no problems.

Mode not normal
Now, let's say that one day, at a random time, I see that sensor A has a value of 8, and sensor B has a value of 2 at the same time. This is a very strange number. What should I say?

Because I've already measured how much this motor shakes, and I KNOW that when the motor is working normally and without any problems, the vibration values should usually be in the yellow area, right?

The yellow area in the picture is a model I've made for when the motor is running normally and there are no problems.

Now that I see a value outside of this area or outside of this model, I can easily say that this new value is not normal and could mean that something is wrong with the motor.

This is how I can use machine learning to figure out when a machine is acting in a strange way. This means that I can use simple machine learning techniques to make a simple model of how a machine or application should normally work and then find the values that are outside of that usual range.

So, this was a simple example of how predicted maintenance can be done with machine learning.