Self-tuning ID Controller Design for Continuously Stirred Tank Reactor - Term Paper Example

The chemical reactions of various compounds in a reaction tank are affected by various factors by which the operator of the machine could control in order to hasten or slow down the reaction time thus the design of a Proportional Integral Derivative controller to continuously stirred tank reactor. The underlying theory as well as how the reaction could be improved can be gleaned indirectly through the results that followed. It is from this understanding that this report is based on an attempt of communicating these findings.

The design considered changes in reaction rate and chemical levels in the tanks that were nonlinear. Therefore, a slight change in these parameters could bring an effect on the processing unit. How these changes could be coordinated to give a more desired aftermath was integral in this investigation. Of these is the output originating from the controller, its ability to coordinate with the desired reaction rate was investigated, how the process could be made more effective from the controller point of view was important, it was therefore mandatory when investigating this to apply control measures, and such effects of introducing any step changes on parameters like stirring rate, the effects of integral actions on the system’s response was key too.

The plant will have a feedback loop that measures the variable, SetPoint that converts feedbacks to a voltage. There is also an Error Signal that alerts about differences between SetPoint and measured variables. The disturbances and Controller to supply power, read SetPoint, process the error, and give the output to the Plant. Problem statement Most industries with chemical reactions of various compounds in reaction tanks encounter problems with the quality of performance in continuous stirring.

This is made difficult when considering that different chemicals have different reaction rates. The reaction of the different chemical compounds will be measured using the conductivity they exhibit after a series of time-lapse.

The reasoning behind is the change of the compounds composition as they react to each other. Since different chemical compounds will have unique conductivity properties, we will be able to infer the rate of reaction and the amount of end substance produced. There are several other factors that affect the rate of reaction time but they all follow and complement each other. The aim of paper is design Matlab PID controller that can monitor stirring and send feedback for correction action. This will improve the application of CSTR in industries.

Approach to problem Matlab Simulink will be used to design PID controller and develop Simulink program to test CSTR. This will include a closed loop control system with fine-tune PID parameters to be simulated. Eventually user-friend program manual will be designed for reference by users of the system. Literature Generally, a neural network can be described as a mathematical model of the biological neural systems (Turchetti, 2004). Within the context of an AI framework, neural networks can be defined as, “A basic component of any system that is intended to exhibit learning and autonomous behavior in a complex environment.

Their primary application is non-linear regression- a way to identify statistical relations between variables” (Wawrzynski and Papis, 2010, p 1). There are a large variety of combinations that neural networks can be arranged in, but they all have five common characteristics which include: 1. Multiple computer processing elements with each individual one consisting of a local memory. The processors can be implemented at different locations such as the hardware, special neural chips, as well as within software or parallel computers. 2. Connection lines that are responsible for transmitting information between the processors.

Similar to the various locations that the processors can be implemented at, the lines of hardware can also be implemented in different locations. 3. Although a single processor is capable of receiving multiple inputs fromother processors, it can only have a single output leading to other processors. 4. The processor output can take on any type of mathematical form depending on its input. 5. Within the network there needs to be a learning law that is responsible for modifying the memory of the processors so it can be adapted to the environment (Braunschweig, 1995).

Given the fact that neural networks are non-linear and flexible, they can be adapted to and used within various activities such as prediction and forecasting, system identification, classification, optimization and decision support (Hanrahan, 2011). Due to their ability to generalize and learn, these systems are better suited for control applications than traditional controls. As well, since they are highly parallel, they can easily process large quantities of sensory information. There are four primary reasons why artificial neural networks are well suited within the control system of an autonomous robot.

They include: 1. Since neural networks provide a link between sensors and motors, they are well equipped to handle a constant flow of input and output signals. 2. They can adapt to noisy environments given the fact that their units are constructed based on several weighted signals. The changes in the individual values of the signals do not impact the behavior of the entire network. 3. Neural networks can be adapted to an array of learning algorithms. 4. Neural networks are flexible so learning techniques can be applied to various levels within the neural network (Khattak et al., 2003). Depending on what needs to be achieved from the network, neural networks can be arranged in different architectures.

Feed forward neural networks are the simplest example of an artificial, neural network. They are described as network connections whose units do not form a full circle. Information flows in a singular direction from the input layer to the output layer. What happens within the input layer is representative of what occurs within the individual networks.