Deep Learning Approach for Behavior of Piston of Linear Compressor

Deep learning–based sensorless control of linear compressor pistons with over 90% performance improvement.

2022.12 – 2023.12
Time-Series Representation Learning
Refrigerator CompressorMLPSensorless TechnologyPrecision ControlDNN

Motivation

  • 80% of the total refrigerator power consumption is consumed by the compressor
  • Cooling is determined by distance and position of piston movement

Goal

Precision control of piston and sensorless technology required

Data

Build an automated data collection system which can measure current, voltage, and stroke

Methodology

  • Modeling a Multi Layer Perceptron (MLP) consisting of four layers (with drop-out)
  • Contribution: More than 90% performance improvement