Song Won Park

Multivariable identification of an activated sludge process with subspace-based algorithms

Abstract This paper is aimed at identifying a linear time-invariant dynamical model (LTI model with lumped parameters) of an activated sludge process. Such a system is characterized by stiff dynamics, nonlinearities, time-variant parameters, recycles, multivariability with many cross-couplings and wide variations in the inflow and the composition of the incoming wastewater. In this simulation study, a discrete-time identification approach based on subspace methods is applied in order to estimate a nominal MIMO state-space model around a given operating point, by probing the system in open-loop with multi-level random signals. Six subspace algorithms are used and their performances are compared based on adequate quality criteria, taking into account identification/validation data. As a result, the selected model is a very low-order one and it describes the complex dynamics of the process well. Important issues concerning the generation of the data set and the estimation of the model order are discussed.

Software sensor for on-line estimation of the microbial activity in activated sludge systems

This paper considers the design of a software sensor (or soft-sensor) for the on-line estimation of the biological activities of a colony of aerobic micro-organisms acting on activated sludge processes, where the carbonaceous waste degradation and nitrification processes are taken into account. These bioactivities are intimately related to the dissolved oxygen concentration. Two factors that affect the dynamics of the dissolved oxygen are the respiration rate or the oxygen uptake rate (OUR) and the oxygen transfer function (K(l)a). These items are challenging topics for the application of recursive identification due the nonlinear characteristic of the oxygen transfer function, and to the time-varying feature of the respiration rate. In this work, OUR and the oxygen transfer function are estimated through a software sensor, which is based on a modified version of the discrete extended Kalman filter. Numerical simulations are carried out in a predenitrifying activated sludge process benchmark and the obtained results demonstrate the applicability and efficiency of the proposed methodology, which should provide a valuable tool to supervise and control activated sludge processes.

A simulation benchmark to evaluate the performance of advanced control techniques in biological wastewater treatment plants

Wastewater treatment plants (WWTP) are complex systems that incorporate a large number of biological, physicochemical and biochemical processes. They are large and nonlinear systems subject to great disturbances in incoming loads. The primary goal of a WWTP is to reduce pollutants and the second goal is disturbance rejection, in order to obtain good effluent quality. Modeling and computer simulations are key tools in the achievement of these two goals. They are essential to describe, predict and control the complicated interactions of the processes. Numerous control techniques (algorithms) and control strategies (structures) have been suggested to regulate WWTP; however, it is difficult to make a discerning performance evaluation due to the nonuniformity of the simulated plants used. The main objective of this paper is to present a benchmark of an entire biological wastewater treatment plant in order to evaluate, through simulations, different control techniques. This benchmark plays the role of an activated sludge process used for removal of organic matter and nitrogen from domestic effluents. The development of this simulator is based on models widely accepted by the international community and is implemented in Matlab/Simulink (The MathWorks, Inc.) platform. The benchmark considers plant layout and the effects of influent characteristics. It also includes a test protocol for analyzing the open and closed-loop responses of the plant. Examples of control applications in the benchmark are implemented employing conventional PI controllers. The following common control strategies are tested: dissolved oxygen (DO) concentration-based control, respirometry-based control and nitrate concentration-based control.

Fault Detection and Diagnosis in the DAMADICS Benchmark Actuator System – A Hidden Markov Model Approach

Abstract Early fault detection and diagnosis in chemical process monitoring represents a challenge to be overcome. Another one concerns the spatial overlapping problem among distinct fault classes, once some events may only be distinguished from the others by taking into account its order of occurrence. The hidden Markov model (HMM) technique is capable of providing information about the tendency of the process and of modelling ordered data. Hence, the goal is to investigate the contribution of this technique to both aspects related to process monitoring activities. The case study is based on the DAMADICS benchmark actuator system. Both abrupt and incipient faulty events were investigated. To the former, detection and diagnosis tasks were immediately satisfied; and to the latter, they were carried out in a progressive and correct course.

Are Automated Planners up to Solve Real Problems

Abstract It is a well known fact that the AI planning community is very committed to apply the developments already achieved in this area to real complex applications. However realistic planning problems bring great challenges not only for the designers during design processes but also for the automated planners during the planning process itself. In addition, it is quite common to face issues about whether the available planners will be up to solve the problem being modeled during the initial design stages. In this paper we present the experience, results and issues that emerged from testing the performance of the recent planners when solving a real and complex problem such as the planning of daily activities of a petroleum plant for docking, storing and distributing oil. Due to the complexity of this real planning problem, the KE tool itSIMPLE was used in order to support all the design processes such as specification, modeling and domain model analysis that resulted in a PDDL model, automatically generated by the tool, which was used as input for planners. In addition, we present the main modeling process performed for the domain model construction.

Fault detection in continuous industrial chemical processes: a new approach using the hidden markov modeling. case study

The development of automatic and reliable monitoring systems is an open issue in continuous industrial chemical processes. The challenges lay on simultaneously managing multiple normal modes of operation as well as the transitions among them with reasonable false alarm rates, and in reaching early fault detection. This work explores and attests the capacity of the signal processing method called hidden Markov model (HMM) in contributing to overcome these issues. After presenting the motivation for its use in this engineering field, the methodology is introduced and an application is illustrated. Here, the HMM ability of directly learning from process historical data both desired features system dynamics and structure of correlations is shown. Aiming to reach practical insights a real case study based on operations of an industrial boiler is used. A comparison with Principal Components Analysis (PCA) and Self-Organizing Maps (SOM) shows the effectiveness of the proposed HMM-based fault detection system.

GRAPHICAL REPRESENTATION OF CAUSE-EFFECT RELATIONSHIPS AMONG CHEMICAL PROCESS VARIABLES USING A NEURAL NETWORK APPROACH

The visualization of relevant information from numerical data is not a natural task for human beings, mainly in case of multivariate systems. In compensation, graphical representations make the understanding easier since it explores the human capacity of processing visual information. Based on that, this study constructs a cause-effect map relating effects of operating process variables over the steam generated by a boiler. This is done after the identification of a neural predictive model for this response. The use of such data-driven technique is due to its capacity of performing a non linear input-output mapping given a reliable database. The case study is based on the operations of a chemical recovery boiler belonging to a Kraft pulp mill located in Brazil. The utility of the obtained map is clear, once the visualization of the contributions of each process variable over the output steam, from this graphical representation, is more intuitive.

Multivariable Identification of an Activated Sludge Process with Subspace-Based Algorithms

Abstract This paper deals with the use of subspace-based identification methods, to obtain multi variable linear dynamic models in state-space form of an activated sludge process around an operating point. Different subspace algorithms (such as CVA, N4SID, MOESP, DSR) are used and compared, based on performance quality criteria. The selected model is validated with a data set not used in the identification process and it describes well the complex dynamics of the process. This model is asymptotically stable and it can be used for control and monitoring purposes.

Variables Selection for Neural Networks Identification for Kraft Recovery Boilers

Abstract The recovery boiler plays a decisive role for the economic and environmental viability in the Kraft process. The residual black liquor from the wood chemical pulping is concentrated and burned in this equipment. The steam generated in the boiler is then used in heat transfer operations and as electrical energy by the mill. This study aims at predicting this energy performance by neural networks. As there are many input variables available for the steam modeling, it leads to variables selection by some techniques. The historical process data was taken from the operation of a hardwood pulping mill located in Brazil.

Fault Detection in Continuous and Periodic Industrial Chemical Processes with Hidden Markov Models

The development of automatic and reliable fault detection systems is still a challenge nowadays. Chemical processes are complex by nature by presenting non linear dynamics, multiple modes with constant interchanges, and spatial and serial correlations, to mention a few. To address these issues, this work explores the hidden Markov model (HMM) technique to construct a fault detection system for continuous and periodic processes. The DAMADICS actuator benchmark, with thirty four abrupt fault scenarios, was used for evaluation purposes. Abrupt faults of low magnitude are challenging and of great interesting in practice. The results obtained with the proposed methodology were compared to classical multivariate statistical process control (MSPC) techniques. They show a significant higher performance leading to earlier fault detection given a fixed false alarm rate of 1%.