Andrei Dinu

Direct Neural-Network Hardware-Implementation Algorithm

An algorithm for compact neural-network hardware implementation is presented, which exploits the special properties of the Boolean functions describing the operation of artificial neurons with step activation function. The algorithm contains three steps: artificial-neural-network (ANN) mathematical model digitization, conversion of the digitized model into a logic-gate structure, and hardware optimization by elimination of redundant logic gates. A set of C++ programs automates algorithm implementation, generating an optimized very high speed integrated circuit hardware description language code. This strategy bridges the gap between ANN design software and hardware design packages (Xilinx). Although the method is directly applicable only to neurons with step activation functions, it can be extended to sigmoidal functions.

A VHDL Holistic Modeling Approach and FPGA Implementation of a Digital Sensorless Induction Motor Control Scheme

This paper presents a sensorless neural-network-based induction motor control scheme, developed by following a holistic approach to electronic system modeling and controller design. The method uses very-high-speed integrated circuits hardware description language (VHDL), allowing the engineering systems functional description to be combined with a detailed digital controller design, which is then implemented into a field- programmable gate array (FPGA). The VHDL description of the hardware-implemented neural networks is automatically generated by C++ programs, in an adaptable architecture, appropriate to low-dynamic systems such as fans and pumps. The complete system performance is investigated by simulation and validated experimentally. This approach provides advantages such as a unique modeling and evaluation environment for complete power electronic systems, the same environment is used for the digital controller design and compact FPGA rapid prototyping, fast design development, short time to market, a CAD platform independent model, and reusability of the model/design

Compensation of Inverter Nonlinear Distortion Effects for Signal-Injection-Based Sensorless Control

A simple analytical technique, which uses readily available datasheet parameters, is developed to model the low-current switching characteristics of insulated-gate bipolar transistors (IGBTs). Simulation and experimental results obtained using three differently rated inverters are presented to demonstrate the accuracy of the technique. The model is applied to compensate the nonlinear distortion, introduced by IGBT switching action in a three-phase inverter, of high-frequency injected voltage and current signals used in sensorless control of a permanent-magnet brushless ac machine. Experimental results show that the compensation technique improves rotor position estimation by up to 20° electrical.

Rotor Design for Sensorless Position Estimation in Permanent-Magnet Machines

A high-frequency injection sensorless rotor position estimation algorithm is incorporated directly into the finite-element design process to realize a permanent-magnet (PM) machine that is suited to zero- and low-speed sensorless control. The machine design is tightly constrained by an existing stator assembly, only enabling the redesign of the replacement PM rotor, and by the requirement that the manufacture of the resulting machine must be simple while meeting aerospace standards. Experimental results confirm that the resulting machine matches the original performance specification and that it operates under position sensorless control.

A VHDL success story: electric drive system using neural controller

This article describes the successful VHDL design of a digital current control architecture for three phase electric drive systems, based on hardware implemented neural networks. The controller comprises an on-line inductance estimator and a PWM switching pattern generator based on a state space observer. Reconfigurable neural networks implemented with logic gates are involved in both. The complete development of the system has been achieved using VHDL and implemented into a single FPGA chip. The VHDL description of the neural networks has been automatically generated by C++ programs. The VHDL model of the motor controller allows the control over the hardware implementation complexity and therefore it can be readily adapted to the available hardware resources.

A Digital Neural Network FPGA Direct Hardware Implementation Algorithm

An algorithm for compact neural network hardware implementation is presented, which exploits the special properties of the Boolean functions describing the operation of perceptrones (artificial neurones with step activation function). The algorithm contains three main steps: the digitisation of the ANN mathematical model, the conversion of the digitised model into a logic gate structure, and finally the hardware optimisation by elimination of redundant logic gates. A set of C++ programs has been developed to implement the algorithm. The programs generate an optimised VHDL model of the ANN implementation. This strategy bridges the gap between the ANN design and simulation software and software packages used in hardware design (Viewlogic, Xilinx). Although the method is directly applicable only to neural networks composed of neurones with step activation functions, it can also be extended to sigmoidal functions.

The effect of magnetic saturation on sensorless control of a brushless permanent magnet motor under AC and DC excitation

The design and performance of a permanent magnet rotor for a concentrated wound, torque dense machine exhibiting a saliency profile suited to high-frequency injection based zero- and low-speed sensorless control is presented. Rotor design is mechanically constrained by an existing stator. Electromagnetic specifications require the motor to match the torque capability of an existing baseline machine whilst demonstrating a saliency characteristic suited for injection based, low-speed sensorless control. The effects of fundamental current amplitude on flux saturation and saliency profile are investigated for brushless DC and brushless AC operation. The saliency profiles required for sensorless detection show significant differences under these operating modes. For the selected rotor design, brushless AC operation results in a stronger saliency signature at light loads, which quickly deteriorates with the onset of saturation. Brushless DC operation results in more gradual deterioration in saliency signature, suggesting sensorless control would be possible over a greater torque range.

Space vector-based hysteresis current controller with self-monitoring mechanism

This paper presents a robust space vector-based hysteresis current controller (SVBHCC) which does not use the back-emf information directly to produce the optimal pulse width modulation (PWM) switching pattern. The controller operation is based on a generalized reference vector which tracks the fundamental harmonic of the load voltage. This vector is calculated iteratively by monitoring the PWM pattern generated by the current controller itself. The generalized reference vector is a direct replacement for the back-emf vector used by other SVBHCC methods previously published in the literature. The main advantages of this new current control method are simplicity and improved current control quality. Two versions of the new control algorithm are presented and test results are shown for an electrical drive system designed around this type of current controller.

Integrated renewable energy system modelling with direct FPGA controller prototyping

A holistic approach to the modeling of integrated renewable energy systems is presented. The method is using the DK5 modeling / design environment from Mentor Graphics and is based on the new Handel-C programming language. The goal of the work carried out was to achieve a combined model of a photovoltaic energy system and a wind power system, enabling an optimised holistic digital control system design, followed by rapid prototyping of the controller into a single Field Programmable Gate Array (FPGA). This would also allow easy connection (by expanding the model/controller) to the grid or to a fuell cell. Initially, the system was simulated using Matlab / Simulink, to create a reference for comparison. Later on, the model was successfully developed in Handel-C. The holistic functional simulation of the system is performed in the same environment as its controller hardware implementation and timing analysis. The controller design was then downloaded in hardware onto a RC203 development board containing a Xilinx Virtex II FPGA and was successfully tested experimentally. This approach enables the design and fast hardware implementation of efficient controllers for Distributed Energy Resource (DER) hybrid systems.

Simulation Package for a New Sensorless Control Strategy for PM Synchronous Machines and Brushless DC Machines

A novel control strategy, applicable to Permanent Magnet Synchronous Machines (PMSMs) and Brushless DC Machines (BLDCMs), was investigated and modelled by an academic team in collaboration with NEC Electronics Europe GmbH. The company has recently developed a new microcontroller, based on a simple sensorless control principle of PMSMs, which is equally applicable to Brushless DC Machines. The use of the new microcontroller in drive systems was modelled by the research team at the university using a Visual C++ environment and was compared with traditional control methods. Extensive simulations were carried out and the results validated the new control principle, which is simple to implement and cost effective if using the novel NEC Electronics Europe GmbH microcontroller. The performance obtained proves the novel control strategy to be at least as effective as the classical ones, at reduced cost. Typical applications are in the white goods industry.