Application of ANN in control systems

The development of Artificial Neural Network (ANN) models for prediction of wheat crop evapotranspiration
using measured weather data and lysimeter measured crop evapotranspiration (ETc) for Delhi is described. Eleven
meteorological variables were taken into consideration for this study. ANN models were developed in MATLAB©
with different network architectures using Feed Forward Back Propagation (FFBP) and Elman Back Propagation
(EBP) algorithms. The total length of data record used was 744, out of that 60% was taken for model training, 20% for
model testing and remaining 20% for model validation. Training and testing data sets were used for model development
purpose, while validation data set was used for model evaluation. The ANN modelling strategy having back
propagation learning algorithm, log-sigmoid transfer function and model input strategy-1 exhibited better results
with Nash-Sutcliffe Coefficient (E) and Root Mean Square Error (RMSE) of 0.972 and 0.498 mm for development
data set and 0.776 and 1.334 mm for evaluation data set, respectively. FAO Penman-Monteith method was also
used to estimate evapotranspiration. Comparison of the ANN predicted ETc and FAO Penman-Monteith estimated
ETc with lysimeter values showed that the ANN predicted ETc was more close to the lysimeter measured values.

Diabetes mereupakan penyakit yang diakibatkan oleh tinggi nya glukosa atau gula yang ada didalam darah dan tidak adanya insulin. Di Indonesia penyakit diabetes biasanya sering di sebut dengan penyakit kencing manis. Penyakit diabetes berbeda dengan penyakit – penyakit yang lain yang bisa dirasakan oleh para penderita. Sering kali para penderita penyakit diabetes atau kencing manis ini tidak merasakan adanya gejala dari penyakit diabetes itu sendiri tiba –tiba mereka telah terserang penyakit diabetes. Oleh karena itu dibutuhkanlah sistem yang bisa membantu mendiagnosa penyakit diabetes.Dalam penelitian ini menggunakan Artificial Neural Network. Adapun parameter yang digunakan dalam Artificial Neural Network. Parameter training cycle sama dengan 500, learning rate sama dengan 0,3 serta momentum 0,2. Dengan mengkombinasikan parameter tersebut menghasilkan RMSE yang kecil yaitu 0,3831 dengan tingkat akurasi 81,8 %

Multi level inverters are gaining attraction because of the inherent advantages like low switching losses and less voltage stress which results in low filter cost. The common techniques that are available for switching the multi level inverters are based on sinusoidal pulse width modulation and using conventional PI based controllers, hysteresis based controllers. These controllers suffer with slow response time this makes usage of multi level inverters in custom power devices difficult. Because custom power devices require fast acting controller action which can be achieved by intelligent controllers. In this project artificial neural network based modulation scheme is designed and implemented for a cascaded H bridge inverter. The response time of controller for different operating power factors of the load are compared with conventional PI controllers and are presented. The developed control technique is developed by using Sim Power Systems Block set of MATLAB/SIMULINK Release R2015a.

One of the most important components in hydraulic circuits is the pump that generates hydraulic supply at constant pressure. The performance and the versatility of a variable displacement piston pump are, to a large extent, determined by its controller. Upgrading of existing controllers is considered one of the major means of improving the characteristics of the pump. The dynamic performance of hydraulic systems with a demand flow supply not only depends on the performance of the flow modulation valve, but also on the performance of the pump. In this study, the performance of soft computing methodology, trained Neural Network (NN) based on the conventional PID controller, is used for the control of a swash plate displacement and compensates the effect of the backing pressure by implementing an inverse NN model. The feasibility of system is simulated and issue of implementation such pumps control is established. It is seen that the use of the proposed methodology results in some desirable characteristics.

The objective of the is work is the study of the vectorial command by neural networks of a three-phase asynchronous machine, fed by a three level voltage inverter of tension. This study was started with a modeling of the three-phase asynchronous machine and its supply (three level voltage inverter). Then we used the vectorial command, which allows to command independently the flux and the electromagnetic torque. Finally, we applied the characteristics of the networks of neurons to identify the behaviour of the system (regulating PI plus the block of the field oriented control FOC), and to replace it by only one neural network

Literatures show that there are several structured integration frameworks which emerged with the aim of facilitating application integration. But weakness and strength of these frameworks are not known. This paper aimed at reviewing these frameworks with the focus on identifying their weakness and strength. To accomplish this, recommended comparison factors were identified and used to compare these frameworks. Findings shows that most of these structure frameworks are custom based on their motives. They focus on integrating applications from different sectors within an organization for the purpose of eliminating communication inefficiencies. There is no framework which guides application's integrators on goals of integrations, outcomes of integration, outputs of integration and skills which will be required for types of applications expected to be integrated. The study recommended further study on integration framework especial on designing unstructured framework which will support and guide application's integrators with consideration on consumer's surrounding environment.

The research was motivated by the extremely altered weather condition of Indonesia within the last two decades. In forecasting the country's weather, recently, we have been built an improvement system of back propagation performance by using Conjugate Gradient (CG) on forecasting of air temperature and humidity in Indonesia using Artificial Neural Network (ANN). The research used weather data taken from the Indonesian Agency for Meteorology Climatology and Geophysics (BMKG). The data included air temperature, rainfall, air humidity, length of sun radiation, air pressure, and wind speed in Kemayoran Jakarta within five years from 2007 to 2012. Through this piece of research, ENN (Evolving Neural Network) algorithm was compared to ANN optimalized with CG (ANN-CG) in order to discover the accuracy of weather forecasting performance between them. Genetic Algorithm was utilized with ENN to discover the ANN's optimal weight and architecture. The result revealed that ENN could fore...

Abstract: In marine application, marine grade steel is generally used for haul and superstructures. However,
aluminum has also become a good choice due to its lightweight qualities, while rusting of aluminum is minimal
compared to steel. In this paper a study on friction stir welding of aluminum alloys was presented. The present
investigation deals with the effects of different friction stir welding tool geometries on mechanical strength and
the microstructure properties of aluminum alloy welds. Three distinct tool geometries with different types of
shoulder and tool probe profiles were used in the investigation according to the design matrix. The effects of
each tool shoulder and probe geometry on the weld was evaluated. It was also observed that the friction stir
weld tool geometry has a significant effect on the weldment reinforcement, microhardness, and weld strength.

An Artificial Neural Network (ANN) is a mathematical model inspired by the biological behaviour of neurons and by the structure of the brain, which is used to solve a wide range of problems. A network is reached by connecting several neurons with a specific architecture (Hopfield Networks, Kohonen Networks, Perceptron, and so on), in which neurons learn through a process of self-organization. 1 During the learning process of the ANN, when a data is introduced into the network, just the neuron that has a positive activity inside the proximity will be activated at the exit stage. 1, 2 There is a wide variety of ANN models, which depend on the objective by which they were created, as well as the practical problem they solve. 3 During these last decades, several inconveniences about ANN applications have been found in the literature, 4, 5, 6, 7 as the performance of the ANN neither have much relation to the amount of acquired information, nor the way the algorithm detects the information, 8 but also with issues like: the selection of the network model, the variables to incorporate on it and the pre-processing of the information that will form the training group. 9 The ANN depends essentially on the exact information of the system under study, and the methods of training that must be used, as the algorithm of ANN during their training have the ability of identifying unnecessary data. 10 Besides, the routines of training require a huge amount of data to make sure that the results can be statistically precise 11, but a lot of algorithms have been proposed to improve the performance of the ANN. 12, 13 Therefore, it can be seen that it is important to know more about the precision and the sturdiness of the ANN. 14 Given the wide range of applications of the ANN towards different areas of science (financial and economic modelling, market profiles - customer, medical applications, management of knowledge and discovering of data, optimization of industrial processes and quality control, among others) generate excellent results. 2, 15, 16 Given the great versatility of the applications and the uses of the ANN, it is possible that can be used in system for which not necessarily has an optimum performance, they may be redundant or inefficient. So, that would be advisable assess the own characteristics of the nature of the problem to solve, in order to assess the application of an ANN.