MAFAULDA_SEP769

This is a work based on the MAFAULDA project.

The main dataset files shall be found on Kaggle - Machinery Fault Dataset

Description:

Based on the MAFAULDA project datased available also on Kaggle - Machinery Fault Dataset, this project's main idea is to create a Deep Neural Network which will classify when the machine is running on its normal status or imbalanced status.

This database comprises on samples taken from a rate of 50 kHz scanning A/D device using the SpectraQuest Inc. Alignment/Balance Vibration Trainer (ABVT) Machinery Fault Simulator (MFS) as shown below:

Source: Machine Fault Simulator

For more details, reach the MAFAULDA project as mentioned on the link above.

Dataset format:

This database is composed of 1951 multivariate time-series acquired by sensors on a SpectraQuest's Machinery Fault Simulator (MFS) Alignment-Balance-Vibration (ABVT). The 1951 comprises six different simulated states: normal function, imbalance fault, horizontal and vertical misalignment faults and, inner and outer bearing faults. This section describes the database.

The database is composed by several CSV (Comma-Separated Values) files, each one with 8 columns, one column for each sensor, according to:

• column 1 - tachometer signal that allows to estimate rotation frequency;

• columns 2 to 4 - underhang bearing accelerometer (axial, radiale tangential direction);

• columns 5 to 7 - overhang bearing accelerometer (axial, radiale tangential direction);

• column 8 - microphone.

And making a simple exploratory analysis, that is what the data extract from the columns 2 to 7 looks like:

Instructions:

Generating data files

If running the code on a local machine:

1 - Download the dataset from Kaggle - Machinery Fault Dataset. The files expected 'normal' and 'imbalance' datasets only.

2 - After cloning this repo, execute "data_generation.py" under the Data Generation folder. This will produce the .txt files containing the data used to train and test the models. Ensure you specify the correct directory when generating the data.

If running on Google Colab, simply open the model .ipynb files in colab and the organized data file will be pulled from a drive folder: Google Drive

Running Models

1 - Open either the LSTM or CNN model in Goolge Colab

2- If you wish to test the model without training, simply change the "saved_model" variable from 0 to 1. This will load the models from the .h5 files.

Group:

Name*	email
Amir Kamaleddine	[email protected]
Ernani Fantinatti	[email protected]
Mohammed Ibraheem	[email protected]

*Names are sorted alphabetically.