Domain Knowledge-Informed Functional Outlier Detection for LQC

An ST-based method using failure pattern knowledge to detect tiny anomalies in manufacturing time-series data.

2022.01 – 2022.12
LG ElectronicsSystem Monitoring & Anomaly Detection
ManufacturingLQCSTDomain Knowledge

Motivation

  • In the manufacturing process, time-series data are collected from multi-sensors and used for quality control.
  • In the line quality control system (LQC) process, weak failures that are difficult to detect with conventional detection methods occur.

Goal

  • To develop a methodology for detecting tiny anomaly patterns in manufacturing time-series data using Sequential Transformation (ST) and domain knowledge of failure patterns.

Methodology

  • The ST maximizes the time-series pattern of a tiny anomaly sample through various calculation.
  • We utilize domain knowledge of failure patterns to define new derivatives and combine them with ST to improve their performance.