Czeslaw T. Kowalski

Neural networks application for induction motor faults diagnosis

The paper deals with diagnosis problems of the induction motors in the case of rotor, stator and rolling bearing faults. Two kinds of neural networks (NN) were proposed for diagnostic purposes: multilayer perceptron networks and self organizing Kohonen networks. Neural networks were trained and tested using measurement data of stator current and mechanical vibration spectra. The efficiency of developed neural detectors was evaluated. Feedforward NN with very simple internal structure, used for the detection of all fault kinds, gave satisfactory results, which is very important in practical realization. Experiments with Kohonen networks indicated that they could be used for the initial classification of motor faults, as an introductory step before the proper neural detector based on multiplayer perceptron is used. The obtained results lead to a conclusion that neural detectors for rotor and stator faults as well as for rolling bearings and supply asymmetry faults can be developed based on measurement data acquired on-line in the drive system.

Neural network application for flux and speed estimation in the sensorless induction motor drive

Sensorless field-oriented control (SFOC) of induction motor drives requires the knowledge of instantaneous magnitude and position of the rotor flux as well as the rotor speed. This paper deals with the application of artificial neural networks (ANN) for estimation of the rotor flux vector and motor speed on the basis of phase current measurement only. Various structures of the neural estimators were simulated and their performances were compared. The influence of changing rotor parameters during the drive were tested. The neural network is able to estimate accurately the rotor flux and speed during line-start operation and load torque changes of the motor. The results of simulation experiments indicate that the neural network estimator may be a feasible alternative to other flux and speed estimation methods.

On-line neural network-based stator fault diagnosis system of the converter-fed induction motor drive

This paper deals with the incipient stator-winding fault detection of the converter-fed induction motor drive. The fault level is modeled by change of a number of shorted stator-winding turns. The method based on a relative phase shift between the phase voltages and line currents of the converter-fed induction motor is used for the on-line fault monitoring and diagnosis. The fault indicators obtained for different load torque and supply frequency conditions for the drive system are used for neural network training. The on-line diagnosis system based on such neural detector is described and tested. Obtained experimental results show very good efficiency of the neural detector, which enables not only fault level evaluation (number of shorted turns) but also fault localization under drive system operation.

General Regression Neural Networks as rotor fault detectors of the induction motor

This paper presents the application of the General Regression Neural Networks in the diagnostics of the induction motors. The specific fault symptoms of rotor damages included in measured stator current spectrum are proposed as elements of input vectors of GRNN. The structure and training procedure of such neural detector are described. Diagnostic results obtained by the proposed neural detector of rotor faults are demonstrated.

Sensorless DTC control of the induction motor using FPGA

The high performance sensorless AC drives require a fast digital realization of many mathematical operations concerning control and estimatorspsila algorithms, which are time consuming. Therefore developing of custom-built digital interfaces as well as digital data processing blocks and sometimes even integration of ADC converters into one integrated circuit is necessary. Due to the fact that developing an ASIC chip is expensive and laborious, the FPGA based solution should rather be used on the design stage of the algorithm. In this paper the application of FPGA in high performance DTC induction motor drive is presented. Few issues concerning the implementation of IM drive control structures in FPGA are discussed. The use of CORDIC algorithm for some mathematical operations in the DTC method is described. Experimental test results of this drive control structure realised in FPGA are demonstrated.

Incipient stator fault detector based on neural networks end symmetrical components analysis for induction motor drives

In this paper a computer system for monitoring and diagnostics of the stator windings of the induction motor fed from the frequencies converter is presented. A possibility of stator faults detection using a neural-network detector, possibly at the early stage of the damage, is checked. The experimental research was carried out for a small power induction motor, with different levels of the stator failures (special motor design allowed to test short circuits in each phase of the motor), load torque and converter frequencies. The measurement of stator phase currents, phase-to-phase voltages and their transformation to symmetrical components are used in stator winding condition monitoring. For diagnostic purposes, the Multi-Layer Perceptron (MLP) neural detectors are developed. Preliminary research of the neural network response (offline) showed that the neural networks with simple internal structure, used for the fault level detection, give satisfactory results, that is very important in practical realization. The designed detector was implemented in LabVIEW and tested during the operation of the machine (online).

The influence of sensor faults on PM BLDC motor drive

The correct commutation of a PM BLDC motor requires information about the rotor position, which can be obtained from rotor position sensors built in motor. Every fault of these components reduces the reliability and performance of the drive system. In this work PM BLDC motor control system has been presented and the influence of some sensor faults on phase currents waveforms has been discussed. For the diagnostic evaluation a-p plane hodographs of phase currents vector and rotor position sensors vector have been proposed. Faulty sensor detection and identification algorithm has been presented. A post-fault operation has been proposed and sample results are presented too.

Speed sensorless DTC control of the induction motor using FPGA implementation

Purpose – The purpose of this paper is to present a fixed‐point implementation of a complete direct torque control (DTC) algorithm connected with a rotor speed estimation algorithm for the induction motor drive, using field‐programmable gate array (FPGA).Design/methodology/approach – The parallel processing approach is described, which requires a decomposition of the control and estimation algorithms for the converter‐fed induction motor to several tasks, realised in parallel. The advanced data processing techniques are described, like PIPELINE technique for data streams design, coordinate rotation digital computer algorithm for transformation of stator flux vector components from Cartesian to polar coordinates. Moreover, the method for the qualitative analysis of the full‐order state observers sensitivity to the variations of the induction motor equivalent circuit parameters is presented.Findings – It is shown that the developed FPGA‐based DTC structure enables designing an efficient application for the...

Application of radial basis neural networks for the rotor fault detection of the induction motor

The main stages of the design methodology of the radial basis neural detectors are described. Furthermore, influence of neural networks complexity and parameters of RBF activation function on quality of data classification is shown. Presented neural detectors are tested with data obtained in laboratory setup contained of converter-fed induction motor and changeable rotors with different degree of damages.

Analysis of inter-turn fault symptoms for the converter-fed induction motor based on the phase-shift calculation

In this paper a stator fault detection method based on monitoring of the phase shift between the line currents and phase voltages of the converter-fed induction motor is presented. The phase shift analysis is used as an indicator of the stator-fault-progression level. The mathematical model of induction motor, which includes the stator-winding fault is presented. The fault level is modeled by change of a number of shorted stator-winding turns. Next the new fault indicator is introduced. Comparison of simulation and experimental results of the converter-fed induction motor for different number of shorted turns and different load torque conditions is shown. Proper fault severity is obtained with the proposed fault indicator.