idaly 1.0.2

A library and a software that support various algorithms for industrial data augmentation.

IdaLy

IdaLy provides a python library and a software that support various algorithms for industrial data augmentation, which can promote the effiency of your industrial machine learning projects(such as fault diagnosis and soft sensor).You can use any algorithms in your script files by importing the python library.Certainly, if you are not familiar with coding of python or just want to finish augmentation tasks conveniently, you can use our software directly.

Augmentations

Because of the sampling difficulty and data privacy of industrial data, data we can acquire from real industry process sometimes is insufficient, thus causing problems such as overfitting and class-imblance while training intelligent agencies. Nowadays, more and more data augmentation algorithms are applied in the field of industrial big data. We collect some algorithms often used and classify them as follows:

• Noise Injection

  • GNI: Gaussian Nosie Injection

• Interpolation

  • SMOTE, Synthetic Minority Over-sampling Technique. Reference: SMOTE, SMOTE in industrial data.

  • LLE, Locally Linear Embedding. Reference: LLE.

  • MTD, Mega Trend Diffusion. Reference: MTD.

  • KNNMTD, k-Nearest Neighbor Mega-Trend Diffusion. Reference: KNNMTD.

• Probability Model

  • GMM, Gaussian Mixture Model. Reference: GMM, GMM in industrial data.

• Deep Learning

  • GAN, Generative Adversarial Network. Reference: GAN, GAN in industrial data.

We collect these algorithms in a python library.You can download it in idaly directly. Or you can download the library by pip instrustion: pip install idaly. It should be noted that the library is Python-based and requires at least Python 3.6, and the envrionment should satify requirement.txt.

Here we give an example about how to use the library.

 import idaly.augmentation as aug

 

 ori_data_path = "your_ori_industrial_data_path.npy"

 out_data_path = "your_output_path.npy"

 ori_data = np.load(ori_data_path)

 

 # aug_method = aug.method(parameter)

 # data_gen = aug_method.fit(ori_data)

 aug_smote = aug.Smote(N=the numer generated, k=15)

 gen_data = aug_smote.fit(ori_data)

 

 # visualization()

 np.save(np.concentrate((ori_data, gen_data), axis=0),out_data_path)

Industrial Data Augmentation Platform

To fininsh industrial data augmentation tasks more conveniently, we intergrate algorithms mentioned above into a software developed by pyqt5. You can download the software here directly.