Deep Q Learning to classify Bearing Fault Modes using Continous Wavelet Transform Scalograms

RL Scheme

.

Episode

1. One episode is basically a set of scalogram images of bearing faults.
2. The Question to the agent when looking at an image is "To which fault does the scalogram belong to ?"

RL Training

Description:

1. In this project, bearing fault classification is done with the help of Deep Q Learning method.
2. A Deep Reinforcement Learning Agent is used to predict whether vibration data of a bearing state is in which fualt mode.
3. The deep reinforcement learning agent has a 2 Convolution Layer with 2 Fully connected layers and a output layer to predict q-values.
4. Vibration data is converted to 2D Continous Wavelet Transform Scalogram for the Training and Testing the agent.
5. The training takes place as quiz game where the agent will predict N number of scalograms in an episode. The scalograms in an episode contains Bearing state of 4 different conditions.
6. If the agent predicts correctly, then it gets a reward of +1 else -1.
7. The goal of the agent is to maximize rewards, thus the goal is to train the agent to predict the state of the scalogram.

Reference Paper:

             Intelligent fault diagnosis for rotating machinery using deep Q-network
             based health state classification: A deep reinforcement learning approach by:
             Yu Ding, Liang Ma, Jian Ma, Mingliang Suo, Laifa Tao, Yujie Cheng, Chen Lu

Data: Case Western Bearing Data

  https://engineering.case.edu/bearingdatacenter/download-data-file