(146a) Multiblock Process Monitoring and Agent-Based Control of Spatially Distributed Processes

Authors: 
Perk, S. - Presenter, Illinois Institute of Technology
Cinar, A. - Presenter, Illinois institute of technology


Some chemical processes consist of a single step, whereas many others include multiple phases due to operational or phenomenological regimes or multiple stages where many different processing units are employed. Having a single model for all different phases or stages with different covariance structures may not give a sufficient explanation of the system behavior and fault detection and diagnosis can be more challenging with increasing model size. Multiblock methods have been recently proposed to improve the capabilities of the existing statistical process monitoring models. The integration of multiblock process monitoring techniques and an agent-based control method and the evaluation of this combined monitoring-diagnosis-control system on spatially distributed reactor networks is reported in this paper. Traditionally, principal components analysis (PCA) is used to form the statistical model based on the covariance structure of the normal operating data and the new observations are tested against this model. Monitoring large-scale distributed systems is a challenging process since many different structures constitute the data and treating the data as if it is coming from a single structure and trying to come up with a single statistical model that perfectly explains the system behavior is usually not possible. Even if the single model is powerful enough to detect any variation from in-control observations, the detection of fault is delayed. For the real- time simulation of a highly nonlinear system, where a small disturbance may have instant drastic effects over the system behavior, timely detection of faults is crucial. Moreover, detecting the parts of the system that were most affected and the variables that contribute to the fault the most is equally important. Here, the system of interest is a CSTR network. We use autocatalytic reactions in these networks to formulate surrogates for predator-prey, virus propagation in a distributed population or chemical manufacturing problems. The network consists of interconnected CSTRs, each having feed and exit streams, as well as multiple connections to their neighbors. The control problem considered is the restoration of normal process operation after a disturbance hits the system. Several disturbances such as a step change in feed flow rate, a sensor fault on a valve controlling the inter flow between reactors or the introduction and propagation of a different species that competes for resource utilization, in one of the reactors or a part of the network are introduced to test the effectiveness of the multiblock monitoring methods in detection and diagnosis and the control structure in restoring the system to its expected behavior. Multiblock monitoring based on consensus PCA algorithm is used in the statistical model formation where each CSTR in the network is represented as a separate block in the model. The multiblock PCA method was especially suitable for use since the system is a network of spatially distributed reactors where each reactor is dominated by different species and monitoring of each reactor is of special interest. The block data consists of concentrations of species in the reactor and the resource concentration. Multiblock monitoring enables the monitoring of each reactor as well as the monitoring of the system at the super level. Considering CSTRs in the network as different entities with different structures helps to build a more reliable, realistic model, prevents loss of necessary information during the dimension reduction stage of single PCA model and more importantly, reduces the challenge of identifying the parts of the system that were affected the most by localizing the fault. The agent-based control structure is called after the monitoring part detects a fault in the system and finds the cause. After the part of the network or the single CSTR that is affected is identified, that part of the system is isolated from the network; the affected reactor(s) are washed out and then are reconnected to the network. The problem becomes an agent based optimum control problem, where the difference between the current states and the normal operation before is minimized by agents via adjusting the flow between reactors, giving minimum disturbance to the other parts of the network. As a whole, the reactive monitoring agents report if there is a fault in the system and the proactive control agents take action to reestablish the normal operation throughout the network. The framework is built in Java using RePast toolkit and COLT distribution. The reactor network model and agent-based control system is implemented with the open source agent modeling and simulation environment RePast. The RePast toolkit is a java-based framework for agent simulation and provides features such as an event scheduler and visualization tools. The control agents created with RePast interact with virtual representations of the physical reactor network. The virtual network objects map the states of the physical system to objects that can be manipulated by the control objects. The ordinary differential equations that describe the autocatalytic reactions in each CSTR are solved numerically using the CVODE solver. The solver code is written in C and linked with RePast via the Java Native Interface (JNI). COLT distribution is used to implement the statistical model building and statistical testing methods and is connected to the agent simulation for online monitoring using RePast event scheduler. The COLT distribution consists of several free Java libraries, for user convenience bundled up under a single name and provides an infrastructure for scalable scientific and technical computing in Java. It contains, among others, efficient and usable data structures and algorithms for off-line and on-line data analysis, linear algebra, multi-dimensional arrays, statistics, histogram computation and Monte Carlo simulation and parallel and concurrent programming. A case study is performed in the network of reactors by introducing a disturbance to the network. The effectiveness of the methods to detect the fault and take action to control the problem is demonstrated.