Search

In this study, the Field Programmable Gate Array (FPGA) technology is employed to integrate multi-loop controllers for motion systems. To provide precise positioning and trajectory tracking for multi-axis systems, the proportional-integral (PI) control is used in the speed control loop and adaptive PID control in the position control loop. The motion system under consideration comprises an X-Y table driven by permanent magnet synchronous motors (PMSMs), and controlled by two programmable servo systems, each designed to regulate a separate axis. Each axis of this system consists of a motion planning module, a speed PI controller in the inner loop, and an adaptive PID position controller in the outer loop. The adaptive PID controller is specifically designed using a multilayer perceptron (MLP) neural network and parameter tuning methods. The control objective is to enhance trajectory tracking accuracy, especially in the presence of dynamic variations and uncertain disturbances. The Very High-speed IC Hardware Description Language (VHDL) is utilized to implement the desirable features of the control system. The control development is based on an FPGA device using Altera’s Quartus II and Nios II software environment. The VHDL designs are analyzed and synthesized within this software environment. Simulation results demonstrate that the on-chip control system can achieve accurate positioning and tracking performance for the X-Y table motion.

Brain-computer interface (BCI) systems can be used to communicate and express desires from people with severe nervous system damage. Among BCI systems based on evoked responses, steady state visual evoked potential (SSVEP) responses are the most widely used. Canonical correlation analysis (CCA)-based methods have been widely used in SSVEP-based online BCIs due to their low computation and high speed, and many methods have been introduced to improve the results. In this research, a method for constructing reference signals used in CCA based on the amplified evoked response of brain is introduced. In the proposed method, after removing the latency in the training signals, to construct reference signals, multilayer perceptron neural networks of the fitting type are used instead of the usual sine/cosine signals. The results show the success of this method in boosting the evoked responses of brain. The detection accuracy in 100-second time windows was 100%, and the information transfer rate in the same period was 240 bits per minute. Making reference signals similar to the recorded electroencephalogram allowed us to make more similarities in the CCA between the signals under consideration, and the reference signals, and to dramatically improve the results.

Brain-computer interface (BCI) systems can be used to communicate and express desires from people with severe nervous system damage. Among BCI systems based on evoked responses, steady state visual evoked potential (SSVEP) responses are the most widely used. Canonical correlation analysis (CCA)-based methods have been widely used in SSVEP-based online BCIs due to their low computation and high speed, and many methods have been introduced to improve the results. In this research, a method for constructing reference signals used in CCA based on the amplified evoked response of brain is introduced. In the proposed method, after removing the latency in the training signals, to construct reference signals, multilayer perceptron neural networks of the fitting type are used instead of the usual sine/cosine signals. The results show the success of this method in boosting the evoked responses of brain. The detection accuracy in 100-second time windows was 100%, and the information transfer rate in the same period was 240 bits per minute. Making reference signals similar to the recorded electroencephalogram allowed us to make more similarities in the CCA between the signals under consideration, and the reference signals, and to dramatically improve the results. [ABSTRACT FROM AUTHOR]

There are several papers on pruning methods in the artificial neural networks area. However, with rare exceptions, none of them presents an appropriate statistical evaluation of such methods. In this article, we proved statistically the ability of some methods to reduce the number of neurons of the hidden layer of a multilayer perceptron neural network (MLP), and to maintain the same landing of classification error of the initial net. They are evaluated seven pruning methods. The experimental investigation was accomplished on five groups of generated data and in two groups of real data. Three variables were accompanied in the study: apparent classification error rate in the test group (REA); number of hidden neurons, obtained after the application of the pruning method; and number of training/retraining epochs, to evaluate the computational effort. The non-parametric Friedman’s test was used to do the statistical analysis. [ABSTRACT FROM PUBLISHER]