A One-Stage Ensemble Framework Based on Convolutional Autoencoder for Remaining Useful Life Estimation.

Area of Science:

• Industrial Engineering
• Data Science
• Energy Management

Background:

• Increasing legislative pressure necessitates accurate energy consumption data for manufacturing production planning.
• Traditional power estimation focuses on single, continuous operations, which is insufficient for complex, modular production lines.
• Modular factories with discrete operations present challenges in signal interpretation due to mixed data and discrete events.

Purpose of the Study:

• To propose a comprehensive data pipeline for estimating future energy consumption in modular manufacturing.
• To develop an automated system for generating labeled datasets for power estimation models.
• To improve the accuracy of power usage predictions in complex factory environments.

Main Methods:

• A data pipeline encompassing data collection, preprocessing, conversion, synchronization, and deep learning classification was developed.
• The system integrates data from various sources, including machine controllers via standardized protocols.
• An auto-labeling mechanism for individual operations was established to eliminate manual data preprocessing.

Main Results:

• The proposed pipeline successfully estimates total power usage for future process plans in modular factories.
• An auto-labeled dataset was created, facilitating the development of a power estimation model without manual intervention.
• Application to a robot arm cell demonstrated synchronized power profiles with the robot program.

Conclusions:

• The developed data pipeline provides an effective solution for energy consumption analysis in modular manufacturing.
• Automated data labeling and deep learning classification enhance the accuracy and efficiency of power estimation.
• This approach supports informed production planning and contributes to reduced energy consumption in factories.