MULTILEVEL INVERTER AND NEURAL NETWORK INTRODUCTION
Download as PPTX, PDF6 likes1,233 views
This document describes a proposed project on multilevel inverter simulation and fault classification. It begins with an introduction on the importance of fault identification in electrical systems. A literature review is then presented covering previous work on fault detection techniques using methods like Concordia transforms and neural networks. The document outlines the proposed work, which includes simulating different types of multilevel inverters and extracting features to train a neural network for fault classification. Procedures for feature extraction, neural network training and testing are described. Future work is noted to expand the simulations and fault diagnosis to additional inverter configurations.
![PROCEDURE FOLLOWED
For fault diagnosis two neural networks are required for analysis, one for
open circuit fault and one for short circuit fault
Total harmonic distortion of each case is taken for open circuit switch fault
classification.
Neural network creation for open circuit switching faults.
The THD matrix for open circuit switch fault classification is x
=[ 28.43 34.73 18.86 18.15 18.43 18.66 18.39 18.15 18.43]
t = [0 1 2 37 48 5 6 37 48]
44](https://image.slidesharecdn.com/suryakant-151218074820/85/MULTILEVEL-INVERTER-AND-NEURAL-NETWORK-INTRODUCTION-44-320.jpg)
![PROCEDURE FOLLOWED
Neural network creation for short circuit switching fault.
the THD values are stored in the matrix x = [28.43 21.31 21.89 43.89
20.85 35.97 18.02 43.95 18.55]
t = [0 1 2 3 4 5 6 7 8]
50](https://image.slidesharecdn.com/suryakant-151218074820/85/MULTILEVEL-INVERTER-AND-NEURAL-NETWORK-INTRODUCTION-50-320.jpg)
MULTILEVEL INVERTER AND NEURAL NETWORK INTRODUCTION
- 1. MULTILEVEL INVERTER SIMULATION FAULT CLASSIFICATION AND DIAGNOSIS SURYAKANT TRIPATHI 12117081 SUMAN KUMAR 12117080 1
- 2. AGENDA • INTRODUCTION • LITERATURE SURVEY • PROPOSED WORK • FUTURE WORK • CONCLUSION • REFERENCES 2
- 3. INTRODUCTION Nowadays most of the electrical projects are based on fault identification and rectification as the society wants an automated system that can not only run the plant smoothly but also automatically rectifies the fault within it. Today’s demand is to run the system continuously, even at the time of fault so that the production during a particular time interval should be maximum to maximise the profit of any industry. 3
- 4. LITERATURE SURVEY In December 1998 Raphael, Stephen and Jean produced a paper on fault detection on three phase inverters by using Concordia transform or alpha beta transform During switching fault conditions due to unbalance in the three phase currents the relation between the alpha beta currents changes and the type of fault can be classified. The relations for switching fault is given as-: 4
- 5. LITERATURE SURVEY In 2007 Surin Komfoi released his 22nd volume on fault detection technique using neural networks. He implemented his theory on cascaded H-bridge inverter. The basic steps for fault detection and remedies according to him is Step 1 - Feature extraction for different kind of faults(THD). Step 2 - Arranging these features in a matrix form and arrange another matrix (target matrix) in which the type of fault are given. Step 3 - Arrange these data column wise feature extraction of n parameters in first n columns and the target matrix in another column. This is called the training data set. Step 4 - Fed this training data set to a neural network for training . Step 5 - Once the network is trained connect this network to a simulated inverter in which feature extraction data has been taken and test for different kind of faults. 5
- 7. BASIC INVERTER • A basic inverter is able to convert dc to pulsating form of ac • These are basically of two types called CSI and VSI. • CSI or current source inverters are those in which the source current remains constant independent load, a VSI or voltage source inverter are those in which voltage is kept constant. • On the basis of construction inverter is classified as Cascaded H-Bridge, flying capacitor type and diode clamped inverter 7
- 10. BASIC INVERTER 10 NO FAULT
- 12. MULTI LEVEL INVERTER Unlike basic type inverters multi-level inverters have more than one voltage levels They are meant to make the output voltage and current waveform more sinusoidal. Actually we are getting a stair-case waveform, capacitive and inductive filters are used to make the waveform smoothen and the resulting waveform becomes sinusoidal. As the level of voltage level increases the size of the smoothening reactor filter reduced to make the stair case waveform more sinusoidal. The main heart of inverter is its pulse sequence. PWM technique is generally used for firing the IGBTS. 12
- 13. MULTILEVEL INVERTER If there are n level of inverter the no of PWM saw tooth waves required for supplying the pulse in the IGBTs is P and the no of IGBTS are I then:- P = I/2 I = 2(n – 1) P = n-1 13
- 14. Neural Network Neural network is a highly interconnected sets of neurons which can be trained and then can be used as a human brain . Its application is not only in engineering ,mathematics and science but also in medicine ,business ,finance and literature as well. Most NNs have some sort of training rule. In other words, NNs learn from examples (as children learn to recognize dogs from examples of dogs) and exhibit some capability for generalization beyond the training data. Neural computing requires a number of neurons, to be connected together into a neural network. neurons are arranged in layers. 14
- 15. Neural Network Architecture Inputs Weights Output Bias 1 3p 2p 1p f a 3w 2w 1w bwpfbwpwpwpfa ii332211 15
- 16. Learning Methods Supervised learning In supervised training, both the inputs and the outputs are provided. The network then processes the inputs and compares its resulting outputs against the desired outputs. Examples-multi-layer perceptron Unsupervised learning In unsupervised training, the network is provided with inputs but not with desired outputs. The system itself must then decide what features it will use to group the input data. Examples-kohonen ,ART 16
- 17. THREE LEVEL INVERTER 17
- 19. FIVE LEVEL INVERTER 19
- 21. SEVEN LEVEL INVERTER 21
- 23. NINE LEVEL INVERTER 23
- 25. NINE LEVEL INVERTER WITH CAPACITOR IN PARALLEL When a capacitor of suitable value is connected in parallel to the resistive load It produces a sinusoidal voltage waveform. For a resistance of 1 ohm 0.1 F capacitor is required 25
- 26. ELEVEN LEVEL INVERTER 26
- 28. THIRTEEN LEVEL INVERTER 28
- 30. T TYPE INVERTER 30
- 31. T TYPE INVERTER VOLTAGE WAVEFORM 31
- 32. T- TYPW INVERTER PULSES 32
- 33. CYCLOCONVERTER Cycloconverter i is an ac to ac converter by changing the input frequency. There are two types of cycloconverters step up and step down. The step up cycloconverter steps up the frequency of output waveform as compared to input voltage waveform. The step down cycloconverter steps down the frequency of input voltage waveform 33
- 35. CYCLOCONVERTER WAVEFORM STEP DOWN 2:1 35
- 36. CYCLOCONVERTER WAVEFORM STEP DOWN 2:1 36
- 37. INVERTER USED IN FAULT IDENTIFICATION SYSTEM 37
- 38. FEATURE EXTRACTION For feature extraction process we have to separate the output waveform in its frequency components. 1st ,3rd,5th, ………19th harmonics are taken. This can be done by Fourier transformation block. Total Harmonic Distortion of each harmonic has been taken for each and every switch fault conditions. For test purpose one switch at a time get faulted. The simulated results of fault condition of five level inverter is taken. 38
- 39. RESULTS (TRAINING DATA) 39 SWITCH/PARAMETER MI-1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 s1c 19.25 16.97 15.2 10.3 6.63 7.36 8.94 13.8 20.91 48.07 s1o 19.33 17.06 15.25 10.33 6.6 7.2 8.73 13.66 20.49 46.76 s2c 19.03 16.76 15.03 10.3 6.73 7.39 8.99 13.86 21.02 48.5 s2o 19.11 16.83 15.07 10.31 6.59 7.2 8.73 13.55 20.49 46.76 s3c 19.33 17.06 15.25 10.33 6.6 7.2 8.73 13.55 20.49 46.76 s3o 18.05 18.46 22 28.94 35 41.68 40.96 43.43 43.1 48.03 s4c 19.11 16.83 15.07 10.31 6.59 7.2 8.73 13.55 20.49 46.76 s4o 18.18 19.8 22.41 22.41 36.43 42.3 41.72 44.43 44.28 48.45 s5c 18.3 18.93 22.63 29.48 36.13 43.52 42.55 45.02 44.46 48.06 s5o 19.42 19.09 22.78 29.53 36.11 43.11 42.51 43.41 44.95 48.45 s6c 18.18 18.59 22.22 29.41 35.74 42.7 41.8 44.04 43.37 48.49 s6o 18.28 18.76 22.37 29.45 35.69 42.49 41.75 44.22 43.77 48.03 s7c 18.57 19.26 23.03 29.93 36.72 43.94 43.31 46.02 45.64 48.49 s7o 18.05 18.46 22 28.94 35 41.68 40.96 43.43 43.1 48.03 s8c 18.44 18.92 22.63 29.85 36.3 43.31 42.55 45.01 44.45 48.06 s8o 18.18 18.8 22.41 29.01 35.43 42.3 41.72 44.43 44.28 48.46 normal 3.2 3.36 4.53 4.4 5.07 7.2 8.73 13.55 20.49 46.76
- 40. TESTING DATA 40 SWITCH/PARAMETER 0.95 0.85 0.75 0.65 0.55 0.45 0.35 0.25 0.15 0.05 s1c 18.09 16.41 13.31 9.28 6.58 6.4 9.68 13.02 21.66 48.07 s1o 18.18 16.49 13.37 9.27 6.48 6.2 9.36 12.71 21.17 46.76 s2c 19.03 16.04 13.15 9.16 6.5 6.46 9.63 13.1 21.77 48.5 s2o 17.97 16.09 13.18 9.13 6.36 6.2 9.36 12.71 21.17 46.76 s3c 18.18 16.49 13.37 9.27 6.48 6.2 9.36 12.71 21.17 46.76 s3o 19.37 20.28 23.31 30.23 39.29 42.74 41.81 43.59 47.41 48.03 s4c 17.97 16.09 13.18 9.13 6.36 6.2 9.36 12.71 21.17 46.76 s4o 19.66 20.83 23.73 30.69 40.03 43.41 42.61 44.59 48.64 48.45 s5c 19.8 21 23.95 31.22 40.96 44.44 43.43 45.21 48.96 48.06 s5o 19.9 21.14 24.11 31.24 40.79 44.26 42.51 45.4 49.4 48.45 s6c 19.52 20.47 23.55 30.76 40.23 43.78 42.65 44.24 47.83 48.49 s6o 19.61 20.59 23.68 30.77 40.05 43.59 42.62 44.39 48.17 48.03 s7c 20.08 21.35 24.38 31.68 41.5 45.11 44.22 46.21 50.19 48.49 s7o 19.37 20.28 23.31 30.23 39.29 42.74 41.81 43.59 47.41 48.03 s8c 19.79 20.8 23.95 31.22 40.76 44.44 43.42 45.2 48.95 48.06 s8o 19.66 20.83 23.73 30.69 40.03 43.41 42.61 44.59 48.64 48.46 normal 2.74 3.25 4.14 3.58 5.93 6.2 9.36 12.71 21.17 46.76
- 43. PROGRAM FOR CREATING P MATRIX 43
- 44. PROCEDURE FOLLOWED For fault diagnosis two neural networks are required for analysis, one for open circuit fault and one for short circuit fault Total harmonic distortion of each case is taken for open circuit switch fault classification. Neural network creation for open circuit switching faults. The THD matrix for open circuit switch fault classification is x =[ 28.43 34.73 18.86 18.15 18.43 18.66 18.39 18.15 18.43] t = [0 1 2 37 48 5 6 37 48] 44
- 45. PROCEDURE FOLLOWED 45 Number of layers - 4(two hidden layers, one input and one output layer) Input layer - 10 neurons Hidden layer 1- 8 neurons Hidden layer 2 – 6 neurons Output layer - 1 neuron Function used – tangent sigmoid
- 50. PROCEDURE FOLLOWED Neural network creation for short circuit switching fault. the THD values are stored in the matrix x = [28.43 21.31 21.89 43.89 20.85 35.97 18.02 43.95 18.55] t = [0 1 2 3 4 5 6 7 8] 50
- 51. PROCEDURE FOLLOWED 51 Number of layers - 4(two hidden layers, one input and one output layer) Input layer - 11 neurons Hidden layer 1- 6 neurons Hidden layer 2 – 5 neurons Output layer - 1 neuron Function used – tangent sigmoid for input layer and hidden layer 1 Pure linear for hidden layer 2 and output layer.
- 56. FUTURE WORK Simulation of T type inverter Fault diagnosis of both 5 level cascaded inverter and T type inverter 56
- 57. CONCLUSION Our project was the fault diagnosis in the multilevel inverter with the help of ANN. We started with the study of neural network so that we would be familiar with what we have exactly to do. In the neural network we studied about the basic neural network , its biological interpretation, application, architecture, classification perceptron neuron model, training rules and some examples. Then we move towards multilevel inverter we started with the basic inverter knowledge and then we go towards multilevel inverter. In this we studied about three level, five level, seven level and nine level inverter and simulated these inverters to find out voltage waveforms We then found out voltage waveforms in the 1st, 3rd ,5th up to 19th harmonics. 57
- 58. REFRENCES Fault diagnostic system for multilevel inverter using ANN by SURIN KHOMFOI VOLUME 2 2007. Unique fault tolerant design for flying capacitor multilevel inverter by XIAOMI N KUO 58