Artificial Neural Networks Controller vs Proportional Integral Derivative Controller - Report Example

ANN (Artificial Neural Networks) Controller vs PID (Proportional Integral Derivative) Controller al Affiliation Industries implement various techniques with several parameters in their operations to control the work of actuators. DC motor is a very common machine that implements the use of control systems. The most common controller for such functions has been Proportional, Integral and Derivative (PID) controller. However, with the dynamism of technology, Artificial Neural Network (ANN) has been introduced. In this paper, ANN is examined in order to understand the reasons for its preference to PID in the current century. The first part provides an introduction of general motor application, then followed by the definition of both the Artificial Neural Networks and Proportional, Integral and Derivative controllers. The main part of the paper brings out the description of ANN controllers and eventual comparison to PID controller. The application of Artificial Neural Networks in controlling motors has proved to provide answers to where classic mathematic is erroneous. Its application will reign into the unforeseeable future. Meanwhile, it has proved to replace PID controllers. Keywords: Neural, Networks, Proportional, Derivative, Integral, Artificial, Artificial Neural Network (ANN) Controller, Proportional, Integral and Derivative (PID) Controller Introduction The need for high performance motor drives is very essential in industrial applications. A good performance motor drive is a system that possesses speed regulating mechanism and a need for a stable system. In this case, a DC motor is considered a system requiring a mechanism with torque and speed characteristics that is compatible with mechanical loads, the reason being that DC motors normally function with the help of drives. There are various problems faced during motor control mechanism, which include presence of variable and erratic inputs, noise propagation and variation in load dynamics (Atri, 2012). Artificial Neural Networks (ANN) are processing algorithms modeled simulating the structure of the mammalian neural system though on a much smaller scale. The difference is that ANN algorithm has thousands or less number of processing units whereas the human brain contains billions of neurons which correspondingly increase depending on the emergences. Neural networks are arranged in layers; each layer is made up of interconnected nodes with activation functionality. It contains an input layer, hidden layers and the output layers (Atri, 2012). ANN implements the learning rule whereby the nodes have to be trained on some inputs similar to what is expected to be doing, just as a child of a mammal is taught to recognize a donkey by being shown some examples of donkeys. As suggested by the name, Proportional Integral Derivative has three modes of control. These are proportional, integral and derivative functions. There are three basic modes to be used with this algorithm, namely Proportion (P), Proportion and Integral (PI) and Proportion, Integral and Derivative (PID). Proportional mode is used to make the adjustment of the output signal by way of controller input. Integral mode is used to offset the sensed error; therefore, it’s used for resetting the system. The response is in a way oscillatory and can destabilize the system; therefore, derivative function is introduced so as to stabilize the system. Applying Artificial Neural Networks to replace Proportional Integral derivatives Mathematical model of ANN Artificial neurons have similarity with the mammalian neurons. ANN neurons replicate the performance of biological neural networks and hence have been classified as very significant tools to be used by control engineers (Makableh, 2011). They are commonly used to solve non-linear issues. The structure of an Artificial Neural Networks is composed of a very long connection of neurons with similar structures. The output can be represented as a = f(Wp + b) while the input represented as n = wp + b, where a is output, p input, b bias, f is transfer function and w is the weight (Makableh, 2011) Industrial processes implement various techniques through application of variable parameters used to operate the working of actuators in the field, the most common application being in the DC motors. The angular positioning of DC motor gives it an advantage to be applied in several controls of processes (Muhammad, 2013). For such an application, PID controllers are used; the signal can be controlled before being sent to the plant unit. However, Artificial Neural Network has been introduced to serve the purpose of controlling the angular positioning of a DC motor. ANN has the ability to cope with the dynamic environment. It works with a supervised learning mechanism whereby data is presented to provide network training before simulation is run. According to Atri, “the approach of neural network basically works on the provided priorities information and makes suitable decision for a given testing input based on the provided training information” (2012, p. 210). Neural networks as applied to the computer world work like the human brain during problem solving. A human brain involves the use of neurons. Its network is through axons and synapses which provide a communication flow generated in form of an impulse. The controller block of ANN controller is trained by inputting the desired values before the actual training starts. The training of ANN controller consumes some time depending on the available parameters and processing speed (Muhammad, 2013). Neural Network controller controls the plant input as per the training data, the inputs are then processed according to the training data. The error that occurs is due to the difference between the actual and desired output, while the output is due to the controlled error and the input. The output is taken back as a feedback to the controller. The process of back propagation begins from the measurement of the actual output in comparison with the desired output. The difference is registered as an error, which through back propagation process, is taken back as an input. The ANN controller process constitutes training and learning of neural networks, whereby the neural networks are adjusted to minimize the error until a desired output is achieved. According to Muhammad (2013), more accuracy is achieved by increasing the number of hidden layers. A careful selection of hidden layers is done with respect to the cost benefit analysis. Although PID controller has good characteristics and response dynamics, it has some limitation. In applications where accuracy is required, the controlling factor needs a specialized concentration to achieve the desired accuracy. Most applications using PIDs have constant gain PID controllers which need an accurate model to describe the dynamics of the system. It is hard to achieve accurate models and therefore PID controllers are limited unlike ANN controllers. The advantage offered by a neural network is its ability to represent both linear and non-linear relationship; furthermore, it has the ability to learn the relationship from the data being generated (Kumar, Ali & Sinha, 2014). Neural networks have processing units and junctions. The junction enables the ordering of a number called weight. The weights are modified so that their characteristics correspond to those of the surrounding where the data is received. Kumar, Ali and Sinha (2014) compare ANN controllers to PID controllers as follows: • ANN controllers have better performance because they have the ability to reduce the steady state error, maximum overshoot, rise time, and settling time. • ANN controller has better sensitivity response to load disturbances as compared to the classical PID controllers. ANN is able to combine artificial neurons so that it can process information. If the weight of the artificial neuron is high, then the input will be stronger. Govindaraj and Dhivya (2014) provide further advantages of ANN system as to have “knowledge acquisition under noise and uncertainty, flexible knowledge representation, efficient knowledge processing, fault tolerance, and learning capability” (p. 789). ANN has different response as compared to PID. ANN controller input signal has no overshoot; this means that the controller is able to function without experiencing any risk of burning itself. Furthermore, absence of overshoot enables the controller to maintain the level of power consumption (Makableh, 2011). The output speed signal generated by ANN has the same shape and value as the controller; this helps the system eliminate available non-linear effects originating from the motor components or any environmental instability causing a disturbance to the motor load. PID controller does not match its signal to that of the output, therefore not so much reliable. There is big difference in time response between PID controller and ANN controller. In ANN, the time taken by the signal to reach the final value is half the time value in the case of PID controller. This property is very important in various applications where any time variation can cause unwanted results in the system which could also destabilize the whole system. Conclusion The best way of improving PID controller is integrating it with an ANN controller to form one combined controller referred to as PID-ANN controller. Madheswaran and Muruganandam point out that “the dynamic speed response of PMDC motor with PID-ANN controller was estimated for various speeds and found that the speed can be controlled effectively” (2012, p. 323). This means that PID-ANN controller has improved speed regulation for a given load disturbance as compared to conventional PID controller. Other proved characteristics of PID-ANN controller according to Madheswaran and Muruganandam (2012) are that there are reduced steady state errors, settling time, and a drop in maximum speed. Reference List Atri, A. (2012). Speed Control of DC Motor using Neural Network Configuration. International Journal of Advanced Research in Computer Science and Software Engineering, 2(5). http://www.ijarcsse.com/docs/papers/May2012/Volum2_issue5/V2I500421.pdf. ISSN: 2277 128X Govindaraj, T. & Dhivya, N. M. (2014). Simulation Modeling on Artificial Neural Network Based Voltage Source Inverter Fed PMSM. International Journal of Innovative research in Electrical, Electronics, Instrumentation and Control Engineering, 2(1). http://www.ijireeice.com/upload/2014/january/IJIREEICE5E_s_DRTG_Dhivya_simulation.pdf. ISSN (Online) 2321 – 2004. ISSN (Print) 2321 – 5526 Kumar, S. B., Ali, M. H. & Sinha, A. (2014). Design and Simulation of Speed Control of DC Motor by Artificial Neural Network Technique. International Journal of Science and Research Publications, 4 (7). http://www.ijsrp.org/research-paper-0714/ijsrp-p3144.pdf. ISSN 2250-3153 Madheswaran, M. & Muruganandam, M. (2012). Simulation and Implementation of PID- ANN Controller for Chopper Fed Embedded PMDC Motor. ICTACT Journal on Soft Computing, 2 (3). http://ictactjournals.in/paper/IJSC_Vol2_Iss3_2_Paper_319_324.pdf. ISSN: 2229-6956(ONLINE) Makableh, Y. (2011). Efficient Control of DC Servomotor Systems using Back Propagation Neural Networks. Georgia Southern University. http://digitalcommons.georgiasouthern.edu/cgi/viewcontent.cgi?article=1771&context=etd. Muhammad, A. (2013). On replacing PID Controller with ANN Controller for DC Motor Position Control. International Journal of Research Studies in Computing, l2 (1), 21-29. DOI: 10.5861/ijrsc.2013.236 Read More