Download PDFOpen PDF in browser

Application of Auto-Encoder for Time Series Classification with Class Imbalance

EasyChair Preprint no. 1471

6 pagesDate: September 5, 2019


Prognostics and health management (PHM) is important to increase the reliability of production equipment and to detect failure events of equipment in advance. In order to model the equipment status, data-driven approaches have been used to extract key feature from various sensors installed on the equipment and build a model to diagnosis the occurrence and end time of the plant failure and the sort of the fault. In particular, the data from each sensor were recorded by specific time period and are also called time series data. To consider the condition of monitor equipment, it is defined as time series classification problem. In recent years, machine learning methods have been widely used to detect plant failure events. Each sensor data typically are separated into several indicators based on their process steps. However, the variation need to be incorporated into time series classification model. In addition, although auto-encoder are widely used to image classification, it still has a challenge to use to time series data. This paper presents an auto-encoder (AE) method of time series classification to distinguish different time series pattern for failure diagnosis. Auto-encoder is a fault detection way of identifying the normal or abnormal of each sensor data. In order to get better results, deep learning model for multivariate time series classification is used to extract the time sequence characteristics. To evaluate the performance of proposed model, the data collected from PHM 2015 were used to compare with the Random Forest, Xgboost and LSTM-based model for performance evaluation. In particular, a plant with minimum proportion fault type was used to examine the effect of imbalanced class. According to the experimental results, the proposed AE outperforms better than other machine learning classification models.

Keyphrases: anomaly detection, Auto-encoder, Classification, Long Short-Term Memory (LSTM), Prognostics and Health Management (PHM)

BibTeX entry
BibTeX does not have the right entry for preprints. This is a hack for producing the correct reference:
  author = {Yi-Ching Yeh and Chia-Yu Hsu},
  title = {Application of Auto-Encoder for Time Series  Classification with Class Imbalance},
  howpublished = {EasyChair Preprint no. 1471},

  year = {EasyChair, 2019}}
Download PDFOpen PDF in browser