Summary
This Repo will analyze the sensor data of several pumps. You can find the data here: https://ga-data-cases.s3.eu-central-1.amazonaws.com/pump_sensor.zip
After a first analysis, we will use a simple LSTM to predict 10min into the future to detect any upcoming failures.
Files:
Sensor_analysis.py : Here we will analyze and manipulate the sensor data. This incorporates several plot functions to visualize the data.
Sensor_learning.py : Here we create the time series for prediction, set up and train the LSTM. The model is saved into a /model folder. Result plots are generated
Sensor_learning_OOP.py :This file has the same functionallity as the Sensor_learning.py file, however, it is coded object oriented. This is meant for better implementation, but also for people learning OOP to compare both files and see the differneces. The Sensor_learning.py might be easier to read/understand for some.
printing_functions.py : This file is used by the Sensor_learning_OOP.py file
Quick Peak:
We have this amount of data:
The Data is composed like this:
We use the following simple LSTM to predict the classes:
With this simple model, we achieved a prediction that should be sufficient to warn the teams 10min in advance about a pump failure.