Predictive maintenance - data required for generalizing the model

I have gone through the video for doing predictive maintenance. (https://www.mathworks.com/videos/predictive-maintenance-part-1-introduction-1545827554336.html). I understand that for doing predictive maintenance in that video, sensor input from 100 machines were used.

I would like to know to generalize a model to do predictive maintenace using user sensor data, data of how many machines I need to use to train/validate the model? How to find how many machine data required, so that model is generalized. Can you please support

Answer 1

[Udit06]

Hi Dhiya,

In order to create a robust model, you will need a large set of sensor data healthy and faulty operations operating in different conditions.

If you do not enough data, as suggested in the video you can build mathematical model and estimate its parameters from the sensor data and simulate a model with different fault states in different operating conditions to generate synthetic failure data

However, the number of machine data required so that the model is robust, depends on multiple factors like model complexity, the number of features that you are using to train the model. The more important thing is to have a better quality data covering different operating conditions since a predictive model is as good as the data that is feeded while training the model.

You can refer to the following resources to understand the amout of data required to train a model:

• https://keras.io/examples/keras_recipes/sample_size_estimate/
• https://www.akkio.com/post/how-much-data-is-required-to-train-ml

I hope this helps.