Neural network for predicting failure, which one to choose?

Question

Neural networks

Good afternoon and straight to the point. There are data taken from the machine such as the vibration of the machine under different operating conditions. There are vibration indicators of various parts of the machine (in my sample there are 8 of them). The essence of the whole process is to identify when the machine enters a malfunction state for this, some malfunctions or assumptions were made in operation (for example, there is not enough lubrication in the links or not tight contact of the gear teeth, etc.). During the operation of the machine, vibration indicators were taken. And the indicator where the failure occurred is marked accordingly.
The bottom line is that it is necessary to use a neural network to try to predict when it is necessary to make machine diagnostics with only vibration indicators of the machine.
Of course, some elements of the machine (parts) are selected (isolated) from the vibration indicators, which can make a little louder noise if they malfunction.
I ask for help to throw the information in which direction to dig or on which forums to sit.
PS Perhaps the description is not so clear, but the idea seems to have been conveyed.

Answer 1

What you wrote is a separate branch at the intersection of technical diagnostics (in your case, most likely vibration diagnostics) and modern machine learning. It's called Predictive maintenance.
(You write " PS Perhaps the description is not so clear, but the idea seems to be conveyed ." - according to the answers given, I suspect that most did not understand what it was about. If anyone is interested, you can start familiarization, for example, from here:
https://towardsdatascience. com/how-to-implement-ma... ).
In fact, the direction is very promising in terms of both economic feasibility and practical application. I'll tell you a secret - once I had to do an analysis of the literature on this topic and saw projects that brought millions and not rubles of profit (I remember that the customers were Caterpillar, Shell, Boeing, etc. monsters). And there are a lot of literature and Internet sources. Another thing is that the use of neural networks is perhaps not the most promising direction in this problem, simply because often traditional machine learning methods give the type of decision trees give more efficient solutions. ​In addition
to the specified search query, if you enter from the side of machine learning methods, you can also use queries like "Anomaly Detection", "Novelty detection", "
Well, in a separate "line" - a few examples of using deep learning networks specifically in tasks close to yours (just from what is at hand now)
https://www.researchgate.net/publication/341615540...
https:// www.sciencedirect.com/science/article/pii/...
https://comtradedigital.com/wp-content/uploads/201...
Good luck with your research. There will be specific questions - ask.

Answer 2

There is a maintenance schedule for the machine, follow it and it will serve you happily ever after.
Everything that you described is kroilovo leading to popadalov.

Answer 3

when it is necessary to diagnose the machine having only vibration indicators of the machine

there, even without a neural network, it will most likely be clear that the amplitude or frequency of oscillations will change due to wear and, accordingly, changes in the vibrating material

Answer 4

Any, the question is, do you have data after how much after a certain vibration (Or something like that (not a physicist)) you need to do diagnostics? If there is, what is the problem?