Fault Detection Using PCA at a Municipal Wastewater Treatment Facility

• Activity Number: 390 - Advanced Fault Detection and Attribution in Large and Complex Data Streams
• Type: Topic Contributed
• Date/Time: Tuesday, July 30, 2019 : 2:00 PM to 3:50 PM
• Sponsor: Quality and Productivity Section
• Author(s): Kathryn Blair Newhart* and Tzahi Cath and Amanda S Hering
• Companies: Colorado School of Mines and Colorado School of Mines and Baylor University
• Keywords: fault detection; PCA; biological processes; sensors

Abstract

Fault detection at municipal wastewater treatment plants (WWTP) in the United States is almost universally univariate due to the complex nature of the wastewater treatment processes, which include physical, chemical, and microbiological treatment units. The resulting data exhibit nonstationarity, autocorrelation, and non-normality. To address these features, a multi-state adaptive-dynamic variation of principal component analysis (PCA) is used to detect out-of-control (OC) process conditions at a WWTP. Mechanical, biological, and sensor faults are induced and analyzed using the PCA program in real-time at a demonstration-scale WWTP in Golden, Colorado for the purpose of early fault detection. Preliminary results demonstrate that the modified PCA program can detect faults hours to days prior to the traditional approach. Different configurations of process treatment units (i.e., multi-state) are compared for their time to detection and false alarm rate. We also investigate the potential of the results from PCA program to be used to diagnose problems.