depending on the magnitude of the physical parameters, using a two-step strategy based on the nonlinear least-squares regres- sion techniques. In this paper, a new method is proposed to estimate the parameters in the single-cage and double-cage models of the inductionmotors. The proposed method is based on artificial neural network (ANN) and ANFIS. For this pur- pose, the manufacturer data of 20 inductionmotors with the different synchronous speed (two poles to eight poles) and the line voltage of 400 V are used. Table 1 shows the manufac- turer data of these inductionmotors (these data are obtained from the motor template), where the subscript ‘‘FL ” refers to the full load, ‘‘M ” refers to the point of the maximum tor- que, and ‘‘ST ” refers to the starting point  . The rest of the paper is organized as follows: Section 2 summarizes the dynamic squirrel-cageinduction motor model. The single- cage and double-cageinductionmotors formulation is pre- sented in Section 3 . In Section 4 , the proposed ANN and ANFIS models for the parametersestimation of the inductionmotors are introduced. Finally, Section 5 describes the simula- tion results of the proposed methods and finally, the last sec- tion summarizes the conclusions of this study.
In this paper, three versions of PSO based on their inertia weight have been used for solving our estimation problem. According to estimated values and calculated errors, third PSO estimates the parameters of studied induction motor more accurate, but in other hand, based on average time of simulation second version of PSO do the simulation faster than others. Therefore, we should consider a reasonable tradeoff between time and accuracy of estimation. In this work, because accuracy is more important third version of PSO can be best choice. Although depend on application of parameter estimation process, this tradeoff can be changed. For example, when parameter estimation is used to detect electrical or mechanical faults , faster method with acceptable accuracy is needed, but when it is just used to estimate structural parameters in order to manufacturing purpose and optimization design, using more accurate method with acceptable speed is reasonable.
The modelling of induction machine with ﬁxed parameters is continuously used nowadays. Nevertheless, its operations in the ﬁeld with variable frequency, particularly for motors with high power, which have suﬃciently deep rotor bars, cannot be represented by a constant rotor resistance and leakage reactance. The nonuniform dispersion of the current density, which passes through these bars, induces skin eﬀect phenomenon. The latter is manifested in high frequency and when the height of the slot is of the order of 20 mm at least . It makes the current ﬂow focus on the periphery of conductor, when the frequency is high. Therefore, its section decreases, and its resistance increases, reducing the calling of the current and amplifying the torque at starting. This principle is used in the motors with deep bars, to improve their characteristics and starting performances.
The artificial mushroom from Agaricus and Lepiota family were retrieved from UCI Machine learning respository . This dataset consists of 8124 instances, 22 attributes, 2 possible classes. Mushroom dataset was split for training and testing purposes. Different sizes of training data were used to check the performance of classifier. The performance of different classification algorithm such as ANN, ANFIS , and Bayes Net classifier were compared on the basis of Mean absolute error, Accuracy, Kappa statistic for mushroom datasets. Feature extraction is required to get mathematical transformation of the multivariate time response. These transformations actually reduce the dimension of input data with more informative data .The steps involved in proposed method is shown in fig.1.
To use KF with non linear system models it is necessary to first linearize the model about a nominal or auxiliary state trajectory to produce a linear perturbation model. The basic KF is used to estimate the perturbation states and then these are combined with the auxiliary states to produce the state estimates of the non linear model. When the auxiliary state is made equal to the most recent KF estimate the procedure is known as the Extended Kalman Filter (EKF). The EKF can be applied as a state estimator for non linear systems. It can carry out combined state and parameter estimation treating selected parameters as extra states and forming an augmented state vector. The result is that whether the original state space model was linear or not the augmented model will be non linear because of multiplication of states.
Torque vs speed curve for the motor using skin effect and the motor without skin effect is shown in fig 4. and fig. 5. respectively. For the motor using skin effect the speed decreases quite smoothly with torque where as for the motor without skin effect it is found that the speed decreases very slowly up to full load torque. So, remarkable change has been found in the torque-speed curve using skin effect.
including the weights and the inputs (Figure 1(a)), total function, activation function, and output [10- 12]. There are two types depending on the ow direction of sign in the neural networks, feedforward network, and feedback or recurrent network. The feedforward network, also known as a static network, is the simplest and most primitive structure of ANN. In this network, knowledge only moves to the forward output layer hiddenly and the system does not have memory. However, the feedback network is a network structure that feeds back to the input units or previous layers from the output and intermediate layers. These neural networks have dynamic memories. The output of neurons in this structure not only depends on the current input values, but also depends on previous input values. Therefore, this network structure is particularly suitable for estimation . The least mean square error method (to minimize errors, which adapts the network weight to the mean square error in between the actual output and the model output) is one of the most widely used learning algorithms for feedback [14,15]. The general structure of feedback learning method is the feedback network. This network is multi-layer and feed forward and suitable for clas- sication, projection, and solving interpretation and generalization problems . It consists of many neural cells connected with ANN. Collection of the neural cells is not random. Generally, the network is constituted in such a way that cells are three-layer and are parallel in each layer. There are hidden layers between the input and output layers (Figure 1(b)). Neural cells in the output layer produce the required output as input data in the input layer of the network to process knowledge from the hidden layers .
model with three-phase squirrel-cageinduction generator, the system is connected to the utility grid. It describes the simulation of a common fault (three phase fault) that occurs along the transmission line of the power system. The response of the wind turbine and the three-phase squirrel-cageinduction generator are analyzed and discussed at different fault periods. Finally A transient fault ride through controller is designed to control and improve the system performance during and after fault occurrence, the controller is a simply PID controller tuned by genetic algorithm and is used to control the turbine blade pitch angle.
Abstract.Most of the electrical machines design studies found in literature lie on the concept that the design under investigation (and optimization) focuses mainly on the geometrical aspects of the machine and thus takes into account only a certain ferromagnetic material (i.e. iron) for its parts. These studies, give little or no information about the influence of material alternatives on the same (and optimized) design. From a manufacturers’ point of view though, this information is crucial especially nowadays that there are a lot of commercially available materials in the market. In this context, this paper presents the results of a research project in the design stage of an energy efficient three phase squirrelcageinduction motor (SCIM), by investigating the effects of several soft magnetic materials (adopted for its stator and/or its rotor parts) on multiple quantities of primary concern such as: efficiency, power factor, output torque, losses, weight and cost. After a brief proposed design procedure, a total of twenty-two different materials from recent manufacturers’ data were examined. Also, the main electromagnetic analysis was performed through commercial analysis software. Simple ranking methods are also proposed here for different application areas and the results obtained are then thoroughly discussed and commented.
developed. This model is based on a winding function approach and the coupled magnetic circuit theory and takes into account the stator and the rotor asymmetries due to faults. This paper presents a computer simulation and experimental dynamic characteristics for a healthy induction machine, machine with one broken bar and a machine with two broken bars. The results illustrate good agreement between both simulated and experimental results. Also, the power spectral density PSD was performed to obtain a stator current spectrum.
During normal operation, the triple pole double throw (TPDT) switch is closed to the starter side. With RYB supply sequence given to the motor, the direction of the rotating magnetic field and direction of rotating of the shaft are in the same direction, say clockwise. To develop the braking torque, the TPDT switch is now closed to the auto transformer side. The Induction machine is provide with reverse phase sequence supply PBY. This causes the direction of the rotating magnetic field to get reversed immediately in the anticlockwise direction. Due to this the direction of the torque developed in the rotor becomes opposite to the direction of the rotating of the shaft. The braking torque causes a rapid reduction in speed. The rate of reduction of speed depends on the magnitude of the negative sequence voltage, applied to the motor, which is normally much lower than the rated voltage. When the speed drops to zero, the machine should be disconnected from the supply, failing which the machine will begin motoring in the opposite direction.
Abstract. Squirrelcageinduction machine are the most commonly used electrical drives, but like any other machine, they are vulnerable to faults. Among the widespread failures of the induction machine there are rotor faults. This paper focuses on the detection of bro- ken rotor bars fault using multi-indicator. However, diagnostics of asynchronous machine rotor faults can be accomplished by analysing the anomalies of machine local variable such as torque, magnetic flux, stator cur- rent and neutral voltage signature analysis. The aim of this research is to summarize the existing models and to develop new models of squirrelcageinduction mo- tors with consideration of the neutral voltage and to study the effect of broken rotor bars on the different electrical quantities such as the park currents, torque, stator currents and neutral voltage. The performance of the model was assessed by comparing the simulation and experimental results. The obtained results show the effectiveness of the model, and allow detection and diagnosis of these defects.
ABSTRACT: This paper presents the estimation of speed of a vector controlled squirrelcageInduction motor drive using an Artificial neural network. Reference Stator voltages and the stator currents are given as input to the ANN and rotor speed is taken as output. Neural network is first trained with test data and is finally the algorithm is tested taking various loads into consideration. For the training of artificial neural network, Levenberg-Marquardt algorithm is implemented. By the implementation of Artificial neural network, the speed of the Induction motor can be estimated accurately and is made independent of stator resistance variation. The proposed method is validated through computer simulation using MATLAB/SIMULINK environment.
Abstract: Flywheel energy storage systems (FESS) are one of the earliest forms of energy storage technologies with several benefits of long service time, high power density, low maintenance, and insensitivity to environmental conditions being important areas of research in recent years. This paper focusses on the electrical machine and power electronics, an important part of a flywheel system, the electrical machine rotating with the flywheel inertia in order to perform charge-discharge cycles. The type of machine used in the electrical drive plays an important role in the characteristics governing electrical losses as well as standby losses. Permanent magnet synchronous machine (PMSM) and induction machines (IM) are the two most common types of electric machines used in FESS applications where the latter has negligible standby losses due to its lower rotor magnetic field until energised by the stator. This paper describes research in which the operational and standby losses of a squirrel-cageinduction machine-based flywheel storage system (SCIM-FESS) are modelled as a system developed in MATLAB/Simulink environment inclusive of the control system for the power electronics converters. Using the proposed control algorithm and in-depth analysis of the system losses, a detailed assessment of the dynamic performance of the SCIM-FESS is performed for different states of charging, discharging, and standby modes. The results of the analysis show that, in presence of system losses including aerodynamic and bearing friction losses, the SCIM-FESS has satisfactory characteristics in energy regulation and dynamic response during load torque variations. The compliance of FESS and its conversion between the generating and motoring mode within milliseconds show the responsiveness of the proposed control system.
Simulation (Section IV) and stability study (Section III) have confirmed that the drive is stable in all the four quadrant modes of operation. Here, the same is verified through experiments for an operating point, which is located in the second quadrant. Load torque is applied through a DC motor which drives the induction machine. The torque generated by the induction machine is shown in Fig. 5(d). A reference speed of −5 rad/s is applied to the machine [Fig. 5(a)]. Hence, the induction machine is now running in second quadrant (regenerating mode). The estimated and actual speeds of the machine are observed in Fig. 5(b) which shows that the proposed MRAC can estimate the rotor speed satisfactorily even in the regenerating mode of operation. The flux orientation is available in Fig. 5(c).
In this paper, the speed of an Induction Motor drive is controlled by the hybrid of PSO- ANFIS algorithms. According to the results of the MATLAB simulation, the Adaptive Neuro Fuzzy (ANFIS) controller efficiently is better than the traditional FLC. The ANFIS-PSO is the best controller which presented satisfactory performances. The major drawback of the fuzzy controller presents an insufficient analytical technique design (choice of the rules, the membership functions and the scaling factors).Thus we chose with the use of the Neural Networks and Particle Swarm Optimization for the optimization of this controller in order to control Induction motor speed. Finally, the proposed controller (PSO-ANFIS) gives a very good result.
a) Current: MCSA, usually performed by FFT, is based on the evaluation of the typical current sidebands located at ±2ksf around the fundamental line (k is an integer), and in particular of the previously described USB and LSB; the measure of only one current is needed, but LSB and USB both must be measured and summed to obtain results quite independent from drive inertia . Anyway, sideband amplitude depends on load torque level, , , , , on the particular motor parameters and power ratings , on manufacturing asymmetries , on constructive details (as spidered rotors, ), and eventually on motor feeding frequency . Load dependency, for example, is a physical phenomenon evident enough. Once the load of an induction machine is removed, rotor currents almost vanish. Therefore, the reaction of a rotor fault on the air gap field and the signatures in the stator current almost disappear, too. Theoretical and experimental evidences of some of these drawbacks have been also given in this work. In addition, load torque fluctuations and speed oscillations produce sidebands similar to LSB and USB, so a mismatch is a concrete possibility, especially in drives with mechanical gear couplings. In certain drives, the low-frequency mechanical oscillations arising from a stage of the gear coupling make the correspondent current sidebands to completely superimpose to LSB and USB, . So, an high-resolution spectral analysis is often required, together with particular methods for removing the load torque oscillation effects from the current spectrum, although additional information may be required through multiple acquisition channels (e.g., currents and voltages), . Moreover, it must be remarked that the fault-related sidebands arise in the current if the machine is supplied by a voltage source such as the mains or a Volt- per-Hertz controlled inverter . Current or torque controlled drives may behave as a current fed induction machine , and the sidebands emerge in the phase voltage, instead. However, the entity of this phenomenon strictly depends on the feed-back control loop speed, and the research about this problem is very recent and still not consolidated.
Diagnostic of electric motors, especially large motors, is currently important operations issue from point of view of electric drives reliability. Therefore in last years it is observed great interest of this subject matter. Many publications where diagnostic signals subject for analysis have arisen (diagnostic signals: available waveforms of stator currents, machine vibrations, etc.) on the basis of such analysis estimation of the damaged squirrelcage is made. It is proposed to use band filters, FFT analysis, wavelet analysis, artificial neural network, Kalman filters and other [1, 2, 3, 4, 6, 7, 8]. However, no one of above methods gives information concerning number of broken bars and their distribution. No one method gives information if a damage concerns bars or squirrelcage rings. In the stator current no additional components arise in case of symmetric damages, therefore such waveform may not be used as a diagnostic signal. In practice it is possible by periodical carrying out measurements to deduce about arising damages by comparison of current waveforms.
outer cage must handle the large starting current for the long acceleration time, with limited path for heat dissipation . Cyclic electromagnetic forces, thermal stress, environmental stress, and mechanical stress associated with the large starting current, make the outer cage more vulnerable to fatigue failure. Particularly, in case of rotor outer cage broken bar, the current in the rotor bar adjacent to the faulty one increases remarkably in comparison to the nominal current. More specifically, the circulation of inter bar currents reduce the degree of rotor asymmetry, leading to a reduced sensitivity of the related sideband current components Conventionally extracted by FFT-based spectrum analysis techniques. Recently, it was proved that this sensitivity is more reduced for a double cage separate end ring rotor than for a common end ring one [4-5]. But inter bar currents produce core vibrations, which can be detected using vibration analysis techniques. The situation is more complicated when the frequency components characteristics of the rotor fault are very close to the fundamental one, and specifically under low load operating conditions [6-7]. In order to discern cases in which the presence of inter bars currents decrease the sensitivity of the MCSA, axial and radial vibrations analysis were investigated in , and more recently for double cage motor in . On the other hand, the rotor current mainly flows in the symmetric inner cage under steady state operating condition, and fault signature is insensitive to outer cage damage. A combined use of current and vibration analysis was developed, by correlating the signal spectra to enhance broken bars detection ability under loaded and unloaded operating conditions of the motor in .
Stator current used to generate magnetic flux, Rotor speed, and Electromagnetic torque waveform. At starting the levels of inrush current consume by the induction motor was reduced hence causing limited voltage drop which will reduce voltage sagging. To compare with model A, which takes a significant amount of time for up to 0.3𝑠 before attaining steady state, allowing higher amount of inrush current. model B attain steady state quickly within a time of 0.1𝑠 curtailing the amount of current consume, hence reduce power consumption with a possibility to reduce energy bills. Model B has better torque control with insignificant harmonics component, hence reduce motor vibration. Compare to Model A with significant amount of harmonic component with a possibility of uncontrolled motor vibration. Model B has a faster rise time and attain steady state at 0.1𝑠 for its rotor speed which is operating at half it rated speed.