Performance comparison of neural architectures for on-line flux estimation in sensor-less vector-controlled IM drives

The sensor-less vector-controlled induction motor drive requires accurate estimation of speed and flux. The speed estimation depends on the motor flux, which has to be measured or estimated. The flux measurement is difficult and expensive and hence generally estimated. Conventional voltage model equations for flux estimation encounter major drawbacks at low frequencies/speed. Neural network-based estimator provides an alternate solution for on-line flux estimation. The on-line flux estimator requires the neural network model to be accurate, simpler in design, structurally compact, and computationally less complex to ensure faster execution time in real-time implementation for effective control. This in turn, to a large extent, depends on the type of neural architecture. This paper investigates three types of neural architectures for on-line flux estimation and their performance is compared in terms of accuracy, structural compactness, computational complexity, and execution time. The suitable neural architecture for on-line flux estimation is identified and the promising results obtained are presented.     

A novel NN based rotor flux MRAS to overcome low speed problems for rotor resistance estimation in vector controlled IM drives

This paper presents a new neural network based model reference adaptive system (MRAS) to solve low speed problems for estimating rotor resistance in vector control of induction motor (IM). The MRAS using rotor flux as the state variable with a two layer online trained neural network rotor flux estimator as the adaptive model (FLUX-MRAS) for rotor resistance estimation is popularly used in vector control. In this scheme, the reference model used is the flux estimator using voltage model equations. The voltage model encounters major drawbacks at low speeds, namely, integrator drift and stator resistance variation problems. These lead to a significant error in the estimation of rotor resistance at low speed. To address these problems, an offline trained NN with data incorporating stator resistance variation is proposed to estimate flux, and used instead of the voltage model. The offline trained NN, modeled using the cascade neural network, is used as a reference model instead of the voltage model to form a new scheme named as “NN-FLUXMRAS.” The NN-FLUX-MRAS uses two neural networks, namely, offline trained NN as the reference model and online trained NN as the adaptive model. The performance of the novel NN-FLUX-MRAS is compared with the FLUX-MRAS for low speed problems in terms of integral square error (ISE), integral time square error (ITSE), integral absolute error (IAE) and integral time absolute error (ITAE). The proposed NN-FLUX-MRAS is shown to overcome the low speed problems in Matlab simulation.     

Feedforward Neural Network Implementation in FPGA Using Layer Multiplexing for Effective Resource Utilization

This paper presents a hardware implementation of multilayer feedforward neural networks (NN) using reconfigurable field-programmable gate arrays (FPGAs). Despite improvements in FPGA densities, the numerous multipliers in an NN limit the size of the network that can be implemented using a single FPGA, thus making NN applications not viable commercially. The proposed implementation is aimed at reducing resource requirement, without much compromise on the speed, so that a larger NN can be realized on a single chip at a lower cost. The sequential processing of the layers in an NN has been exploited in this paper to implement large NNs using a method of layer multiplexing. Instead of realizing a complete network, only the single largest layer is implemented. The same layer behaves as different layers with the help of a control block. The control block ensures proper functioning by assigning the appropriate inputs, weights, biases, and excitation function of the layer that is currently being computed. Multilayer networks have been implemented using Xilinx FPGA "XCV400hq240." The concept used is shown to be very effective in reducing resource requirements at the cost of a moderate overhead on speed. This implementation is proposed to make NN applications viable in terms of cost and speed for online applications. An NN-based flux estimator is implemented in FPGA and the results obtained are presented     

New membership functions for effective design and implementation offuzzy systems

The paper proposes new membership functions (mfs) for fuzzy modeling. Existing mfs do not simultaneously satisfy ease in optimization and low-end hardware implementation. The proposed mfs satisfy the two contradicting requirements. The algorithm for hardware implementation is detailed. The performance and applicability of the proposed mfs are illustrated using two well-known benchmarks. Optimized fuzzy models are coded and implemented using the Intel 8XC196KC microcontroller. Results show that the proposed mfs simplify offline design and facilitate online implementation