qawl987 / dast-self

About The Project

This project aims to use Transformer instead of RNN/CNN to imporve model performance on bearing RUL(Remain Useful Life) prediction.

(back to top)

Built With

python 3.7.15 CUDA Version: 10.1

(back to top)

Getting Started in Colab

Prerequirement & Installation

No!

Dataset

1. Download zip from https://github.com/wkzs111/phm-ieee-2012-data-challenge-dataset
2. upload zip to your google drive, don't extract and upload it will take long time.

Pull github repo

1. create a colab notebook in your drive
2. run below command

from google.colab import drive
drive.mount('/content/drive')
%cd /content/drive/MyDrive/
!git clone https://github.com/qawl987/DAST-self

3. close this notebook

Run the autorun notebook

1. open /content/drive/MyDrive/DAST-self/notebook/FEMTO-st/DAST_run.ipynb
2. unzip the dataset zip in workspace & copy to drive for future reuse

注意

1. 資料集部分只要跑一次即可，會存到google drive中
2. 以下指令複製PHM 2012資料集後在當下這個colab notebook中的workspace可見，但實際上複製到drive的過程很慢，如果關掉或結束這個colab執行階段，需要等大概1或2小時才可在google drive中實際見到資料。所以解決方法1. 不要關掉這個colab跑完 2. 先跑到複製的指令後，等1,2小時後回來直接跑下面的實驗即可。 !cp -r "/content/phm-ieee-2012-data-challenge-dataset-master" "/content/drive/MyDrive/TMP-for-DAST"

!unzip /content/drive/MyDrive/phm-ieee-2012-data-challenge-dataset-master.zip -d /content/
!cp -r "/content/phm-ieee-2012-data-challenge-dataset-master" "/content/drive/MyDrive/TMP-for-DAST"

3. move TMP folder dataset to DAST-self project folder

!mv /content/drive/MyDrive/TMP-for-DAST/phm-ieee-2012-data-challenge-dataset-master/Full_Test_Set/* /content/drive/MyDrive/DAST-self/data/10FEMTO/raw_data/Validation_set
!mv /content/drive/MyDrive/TMP-for-DAST/phm-ieee-2012-data-challenge-dataset-master/Learning_set/* /content/drive/MyDrive/DAST-self/data/10FEMTO/raw_data/Validation_set

4. preprocess Choose dataset & run the preprocess block
5. DL Set hyperparameter & run the DL block
6. TL Set hyperparameter & run the TL block

Getting Started in Local

This is an example of how you may give instructions on setting up your project locally. To get a local copy up and running follow these simple example steps.

Prerequirement

You need to intall appropriate torch version with GPU according to your CUDA version

Installation

1. Install package

   pip install -r requirements.txt

Dataset

1. Download dataset from https://github.com/wkzs111/phm-ieee-2012-data-challenge-dataset & extract zip

move all folder under Full_Test_Set/ and Learning_set/ to /data/10FEMTO/raw_data/Validation_set/ 
All raw data path /data/10FEMTO/raw_data/Validation_set/

Usage

Preprocess

1. Create folder /data/10FEMTO/processed_data/
2. run all block in notebook/FEMTO-st/data_processing/main.ipynb

Train

Deep learning

# model hyperparameters
selected_indices = [1, 3, 5, 7, 9, 10, 14]
# selected_indices = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15]
FEATURE_LEN = len(selected_indices)
FEATURE_SIZE = 20
EMBEDD = 20
HYPER_PARAMETERS = {
    # model parameter
    'batch_size': 256,
    'dim_val': FEATURE_SIZE,
    'dim_attn': EMBEDD,
    'dim_val_t': FEATURE_SIZE,
    'dim_attn_t': EMBEDD,
    'dim_val_s': FEATURE_SIZE,
    'dim_attn_s': EMBEDD,
    'n_heads': 4,
    'n_decoder_layers': 1,
    'n_encoder_layers': 2,
    'lr': 1e-3,
    'epochs': 100,
    'time_step': 40,
    # limit how many last input used, important!
    'dec_seq_len': 6,
    'output_sequence_length': 1,
    'feature_len': FEATURE_LEN,
    'debug': True
}

# Other Hyperparameters
DATA_PATH = '../../../data/10FEMTO/processed_data/' # set the data path after preprocess
TRAIN_DATASETS = ['Bearing1_1', 'Bearing1_2']
TEST_DATASET = ['Bearing1_3', 'Bearing1_4', 'Bearing1_5', 'Bearing1_6', 'Bearing1_7']
# TRAIN_DATASETS = ['Bearing2_1', 'Bearing2_2']
# TEST_DATASET = ['Bearing2_3', 'Bearing2_4', 'Bearing2_5', 'Bearing2_6', 'Bearing2_7']
# TRAIN_DATASETS = ['Bearing3_1', 'Bearing3_2']
# TEST_DATASET = ['Bearing3_3']

MODEL_SAVE_NAME = f'Bearing{TRAIN_DATASETS[0][7]}_pretrain_{len(selected_indices)}'
NORM_TYPE = NormType.NO_NORM # choose norm method before feed into model, no_norm or batch_norm or layer_norm
TRAIN_TYPE = TrainType.DL # set train mode to deep learning
MODEL_SAVE_PATH = '../../../model/norm' # save the model save path
IS_SAVE_MODEL = False # set if save the model

• Run DL block in notebook/FEMTO-st/model_training/main.ipynb

Transfer learning

DATA_PATH = '../../../data/10FEMTO/processed_data/'
PRETRAIN_DATASET = 'Bearing2' # set pretrain dataset name
FINETUNE_DATASET = 'Bearing3' # set finetune dataset name

TRAIN_DATASETS = ['Bearing1_1', 'Bearing1_2']
TEST_DATASET = ['Bearing1_3', 'Bearing1_4', 'Bearing1_5', 'Bearing1_6', 'Bearing1_7']
# TRAIN_DATASETS = ['Bearing2_1', 'Bearing2_2']
# TEST_DATASET = ['Bearing2_3', 'Bearing2_4', 'Bearing2_5', 'Bearing2_6', 'Bearing2_7']
# TRAIN_DATASETS = ['Bearing3_1', 'Bearing3_2']
# TEST_DATASET = ['Bearing3_3']

MODEL_SAVE_PATH = '../../../model'
MODEL_SAVE_NAME = f'{PRETRAIN_DATASET}_pretrain_{FINETUNE_DATASET}_finetune_7'
PRETRAIN_MODEL_PATH = '../../../model' # set the pretrain model path
PRETRAIN_MODEL_NAME = f'{PRETRAIN_DATASET}_pretrain_7'
IS_SAVE_MODEL = False
NORM_TYPE = NormType.NO_NORM
TRAIN_TYPE = TrainType.TL

• Run TL block in notebook/FEMTO-st/model_training/main.ipynb