Solving the puzzle: Leveraging machine learning for effective root cause analysis

Reduce complex root cause analysis from days to under two hours. This talk shows how to do it with explainable AI.

#1about 3 minutes

The challenge of root cause analysis in complex manufacturing

High-precision manufacturing, like semiconductor lithography, generates complex data that makes finding the root cause of failures extremely difficult.

#2about 3 minutes

Why traditional data analysis methods fall short

Standard statistical methods like linear regression fail with real-world manufacturing data due to non-linearity, missing values, and high class imbalance.

#3about 2 minutes

Using explainable AI to understand black box models

Explainable AI (XAI) provides methods to understand how a machine learning model makes predictions, which builds trust and helps discover hidden patterns in data.

#4about 9 minutes

A three-step approach using LightGBM and SHAP

A practical workflow for root cause analysis involves training a LightGBM model, explaining it with SHAP to find feature importance, and forming hypotheses from the results.

#5about 2 minutes

Building a self-service tool for domain experts

To scale the impact of data science, an internal self-service tool was built to empower engineers to perform their own root cause analysis without data science expertise.

#6about 3 minutes

Key principles for successful tool adoption by engineers

The tool's success relied on being human-centric, transparent about limitations like correlation vs causation, robust for real-world data, and simple to use.

#7about 1 minute