Adaptive Neuro-fuzzy Inference System Network Research Articles

Neuro-fuzzy-based space vector modulation for THD reduction in VSI fed induction motor drive

Space vector modulation is an optimal pulse width modulation technique for variable speed drive application. This paper proposes adaptive neuro-fuzzy inference system (ANFIS) based space vector modulation (SVM) technique for voltage source inverter. The proposed ANFIS network is independent of the switching frequency and uses hybrid learning algorithm for training. Due to this learning algorithm, SVM algorithm can be implemented very fast and the desired training error can be obtained with less number of iterations compared to other optimisation techniques like neural, fuzzy and genetic. The performance of ANFIS controlled drive is compared with the conventional SVM-based drive. The simulation results of inverter phase voltages obtained are verified experimentally using a Dspace kit (DS1104). The % THD value of simulation and experimental waveforms of inverter phase voltages for 3 kHz switching frequency is presented.

Research on Integrated Navigation Algorithm Based on ANFIS

Combination of Global Positioning System (GPS) and Inertial Navigation System (INS) can improve the navigation performance that is superior to either one. This paper proposed and discussed an INS/GPS integrated navigation method based on adaptive neuro-Fuzzy Inference System (ANFIS) to fuse INS and GPS data. In this method, an ANFIS network was trained to mimic the error dynamical model of INS when GPS signals were available. If GPS outages occur, the trained ANFIS network is utilized to bridge the GPS outages. Simulations in INS/GPS integrated navigation system show the proposed method can reduce the positioning error during GPS outages.

Prediction of Surface Roughness in End Milling Process Using Intelligent Systems: A Comparative Study

A study is presented to model surface roughness in end milling process. Three types of intelligent networks have been considered. They are (i) radial basis function neural networks (RBFNs), (ii) adaptive neurofuzzy inference systems (ANFISs), and (iii) genetically evolved fuzzy inference systems (G‐FISs). The machining parameters, namely, the spindle speed, feed rate, and depth of cut have been used as inputs to model the workpiece surface roughness. The goal is to get the best prediction accuracy. The procedure is illustrated using experimental data of end milling 6061 aluminum alloy. The three networks have been trained using experimental training data. After training, they have been examined using another set of data, that is, validation data. Results are compared with previously published results. It is concluded that ANFIS networks may suffer the local minima problem, and genetic tuning of fuzzy networks cannot insure perfect optimality unless suitable parameter setting (population size, number of generations etc.) and tuning range for the FIS, parameters are used which can be hardly satisfied. It is shown that the RBFN model has the best performance (prediction accuracy) in this particular case.

Sensitivity of total charge and maximum charge per delay on ground vibration

Frequency is among the most used properties for determining the detrimental effect of ground blasting on the surrounding areas. In this paper, an attempt has been made to calculate the frequency of the blast by developing an adaptive neuro-fuzzy inference system (ANFIS) model. Blast design and explosive parameters are incorporated in the intelligent model. The sugeno type fuzzy inference system was generated using the subtractive clustering partition method. Two separate ANFIS networks have been generated, each with five inputs (maximum charge per delay and total charge, respectively in two different cases) and one output parameter (frequency) was trained using a hybrid parameter optimization method. A set of 160 data obtained from blast monitoring at a major surface coal mine in India was used for training the network. A different set of 27 data was used for validating and testing the designed network. It has been observed that for a given distance the frequency of the vibration is affected not only by the maximum charge per delay but by total charge also. The results obtained from ANFIS were able to predict that the frequency of ground blast operation is obtained with higher precision when the maximum charge per delay is used as an input parameter instead of total charge.

Full Measuring System for Copper Electrowinning Processes Using Optibar Intercell Bars

An evolved Optibar intercell bar for copper electrowinning processes with current sensing capabilities is presented. This technology upgrades the advantages of the conventional Optibar by providing a complete measuring system using magnetic sensors inside the capping board. To enhance reliability and simplicity, only half of the intercell currents are physically measured. This is accomplished using adaptive neurofuzzy inference system networks to calculate the current flowing through nonsensored connecting Optibar segments using virtual sensors. This way, process cathode currents are available online enabling myriad of key process computations; i.e., cathode harvest time, weight at harvest, current dispersion among cathodes, energy consumption and process efficiency, and online setting of optimum process current level based on the ability to detect and locate metallurgical short-circuits events. In this paper, this task is accomplished using a short-circuit diagnosis algorithm based on the recognition of current distribution patterns that characterizes the phenomena using artificial neural networks. The technology employed to implement the system is completely embedded in the bar to ensure compatibility with the process environment. From the outside, Optibar add-ons are hidden and do not disrupt the operation. On site industrial data proved that cathode current measurements exhibit an average absolute error lower than 2% with a dispersion lower than 1.6%. Finally, the algorithm developed for short-circuit diagnosis exhibits a success rate of 98% or better.

A genetically trained adaptive neuro-fuzzy inference system network utilized as a proportional-integral-derivative-like feedback controller for non-linear systems

This paper presents a genetically trained PID (proportional-integral-derivative)-like ANFIS (adaptive neuro-fuzzy inference system) acting as a feedback controller to control non-linear systems. Three important issues are addressed in this paper, which are, first, the evaluation of the ANFIS as a PID-like controller; second, the utilization of the GA (genetic algorithm) alone to train the ANFIS controller, instead of the hybrid learning methods that are widely used in the literature; and, third, the determination of the input and output scaling factors for this controller by the GA. The GA, with real-coding operators, is used to adjust all of the ANFIS parameters, which include the input and output scaling factors, the centres and widths of the input membership functions (MFs), and the consequent parameters. To show the effectiveness of this controller and its learning method, several non-linear plants, including the CSTR (continuous stirred tank reactor), have been selected to be controlled by this controller through simulation. Moreover, this controller's robustness to output disturbances has also been tested and the results clearly indicated the remarkable performance of this controller and its learning algorithm. In addition, the result of comparing the performance of this controller with a genetically tuned classical PID controller has shown the superiority of the PID-like ANFIS controller.

Impulse activated sparse cell array network in non-linear autoregressive process modeling

This paper introduces new neural network/fuzzy logic architecture called the impulse activated sparse cell array network (IASCAN) architecture. The motivation behind and origins of the new architecture are discussed. Its performance in non-linear dynamic system modeling using the Mackey Glass process which exhibits chaotic behavior is compared with that of Adaptive-Neuro Fuzzy Inference Systems (ANFIS). This is done by comparing the root mean squared errors (RMSE) using training and checking data sets after the networks are trained with data taken from simulations of the Mackey Glass process. The same data sets are used in training and checking of errors for both the ANFIS networks and the new architecture for a direct comparison of performance. The results shows that the RMSEs of the new architecture were lower than that of the ANFIS networks with standard configuration of the problem even after many epochs of training. The errors were shown to approach that of the new architecture for equivalent order models only after increasing the numbers of Membership Functions (MFs) in the ANFIS model after which it took an extremely long time comparatively to produce a solution. The method is also uniquely extensible to the practical solution of high order problems. There are also other features that make this new architecture very promising.

Adaptive neurofuzzy inference systems networks design using hybrid genetic and singular value decomposition methods for modeling and prediction of the explosive cutting process

Genetic algorithm (GA) and singular value decomposition (SVD) are deployed for the optimal design of both Gaussian membership functions of antecedents and the vector of linear coefficients of consequents, respectively, of adaptive neurofuzzy inference systems (ANFIS) networks that are used for modeling of the explosive cutting process of plates by shaped charges. The aim of such modeling is to show how the depth of penetration varies with the variation of important parameters, namely, the apex angle, standoff, liner thickness, and mass of charge. It is demonstrated that SVD can be effectively used to optimally find the vector of linear coefficients of conclusion parts in ANFIS models and their Gaussian membership functions in premise parts are determined by a GA.