![Masters Theses](../../assets/md5images/ebc0648ded2cd429f55e0a2fe4407546.gif)
Masters Theses
Date of Award
12-1999
Degree Type
Thesis
Degree Name
Master of Science
Major
Chemical Engineering
Major Professor
Tsewei Wang
Committee Members
Paul D. Frymier, Duane D. Bruns
Abstract
This research analyzes the feasibility of developing a Multivariate Statistical Process Control (MSPC) framework for monitoring and diagnosing a biological wastewater treatment plant. MSPC makes use of historical database of past successful operations as a reference to judge the normality of future operations. The projection method, Principal Component Analysis (PCA), is utilized not only to compress the originally correlated data but also to extract statistically meaningful information, by projecting the multivariate trajectory data onto a lower dimensional space, spanned by the Principal Components (PC s) retained. From the established 'normal' operation domain, departure of new operating points from that of 'normal' domain can be detected by the use of several MSPC monitoring plots.
The proposed methodology generates monitoring charts by analyzing the process variables gathered in a reference database; new observations are analyzed by contrasting their projections onto the reference PC s space against that of normal, using a variety of monitoring charts. Possible root causes can sometimes be identified when abnormal deviations have been detected. The capability of such MSPC scheme in monitoring and assessing the behavior of new wastewater treatment operations against the reference is illustrated through simulations of the bio-wastewater treatment plant under a variety of operating conditions.
The research first reviews the concepts and techniques of MSPC and the Activated Sludge Model No. 1. It then utilizes these techniques in creating the monitoring and diagnosis framework for a wastewater bio-treatment plant using the activated sludge model No. 1 description as the process model. Simulation is carried out using the Matlab (version 4.2c) and Simulink"^" as the programming platform. The MSPC framework is able to detect abnormal process deviations by comparing the projection of new observations onto the principal component subspace to the 'normal operation' region established from base case data. If current operating points fall inside this region, it implies that the current operation is 'normal'; If they fall or show a trend of migrating toward outside of the region, it implies emergence of abnormal operations. Usually, it is possible to trace back from the abnormal behavior to their assignable causes by analyzing contribution plots.
In this study, a reference database is generated based on the simulation of a large number of variations in the process operating conditions in the neighborhood of a nominal operating condition. These variations include: -21% to +21% changes in the influent nitrate concentration, [NO3"], in the maximum growth rate of the heterotrophic biomass, pm, h, in the half-saturation constant of COD, Kg, [cod] and - 15% to +15% changes in the influent ammonia concentration, [NH4'^]. These deviations are defined as 'normal operation' deviations. Monitoring charts are obtained based on this simulated database. Acceptable regions are identified in these charts as the standards for monitoring all future processes. Three abnormal cases are simulated to validate the established base case PGA model. They represent 1) bigger than normal amount of changes in the operating conditions not affecting the biological model; 2) bigger than normal amount of changes in the bioprocess parameters altering the process model; 3) new biological event causing plant/model mismatch. Analysis results show that the indication of the migration, over time, toward a state of abnormality is clear and direct. Diagnosis is carried out by analyzing the contribution plot for each of the three abnormal cases. Results show that the PCA method can also identify the possible root causes for the observed abnormality. In addition, the interpretation of the principal components provides more insights to the behavior of the process variables. However, important implementation issues remain that must be addressed before it can proved to be effective when brought on line.
Recommended Citation
Fu, Bei, "Development of Multivariate Statistical Process Control for an industrial prototype wastewater bio-treatment plant. " Master's Thesis, University of Tennessee, 1999.
https://trace.tennessee.edu/utk_gradthes/9825