Knockoff procedure for false discovery rate control in high-dimensional data streams.

Area of Science:

• Statistics
• Industrial Engineering
• Data Science

Background:

• Root-cause identification in high-dimensional data streams is crucial for fault detection.
• Limited samples of out-of-control (OC) data streams pose challenges for traditional methods.
• Controlling false discoveries is essential for reliable fault identification in statistical process control (SPC).

Purpose of the Study:

• To propose a novel Knockoff procedure for multiple testing in multivariate SPC.
• To develop a method that combines with existing fault detection techniques without altering stopping times.
• To control the false discovery rate (FDR) in identifying OC data streams.

Main Methods:

• A Knockoff filtering procedure is proposed for multivariate SPC.
• The procedure is designed to integrate with other fault detection methods.
• Theoretical guarantees for FDR control are provided via a theorem.

Main Results:

• The proposed Knockoff procedure effectively controls the false discovery rate (FDR).
• The method demonstrates high statistical power in identifying faulty data streams.
• Simulation studies validate the performance and FDR control.

Conclusions:

• The Knockoff procedure offers a robust solution for fault identification in SPC with limited OC samples.
• This method enhances the reliability of fault detection in high-dimensional data streams.
• The approach is applicable to real-world scenarios, such as semiconductor manufacturing.