Loading...

Experimental Modeling of a Transparent Fuel Cell with Aid of Deep Neural Network to Measure Water Coverage Ratio and Fuzzy Control

Mehnatkesh Ghadikolaei, Hossein | 2020

500 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 53570 (08)
  4. University: Sharif University of Technology
  5. Department: Mechanical Engineering
  6. Advisor(s): Alasty, Aria; Kermani, Mohammad Jafar; Boroushaki, Mehrdad
  7. Abstract:
  8. Polymer membrane fuel cell is a rich source of renewable energy. Two problems of short life and low reliability in fuel cells are the main problems of this system. The difference in partial pressure of hydrogen and oxygen causes serious damage to the fuel cell and inefficiency of the electricity production capacity. On the other hand, the life and optimal performance of the fuel cell depends on the moisture content of the membrane because the membrane needs sufficient moisture to pass ions. Transparent fuel cells can be used to study the water in the fuel cell. With direct imaging of this type of fuel cells, the phenomenon of water creation in the fuel cell can be studied with the help of image processing, but this method has a high computational cost and is sensitive to the initial position of the camera and ambient light. In this research, in order to reduce the difference between the partial pressure of hydrogen and oxygen, mathematical modeling of the fifth order of polymer membrane fuel cells will be performed and then we will control this mathematical model with the help of different controllers. Due to the qualitative nature of the processes in the fuel cell, a fuzzy controller has been used to control the model. Finally, we optimize the fuzzy PID controller using the particle swarm optimization method, which shows a 78.66% improvement in reducing the largest partial pressure difference compared to Ziegler-Nichols method. One of the most important parameters in the life time of the fuel cell is the presence of water in it. The water coverage ratio in the fuel cell was investigated by a deep neural network using real data from a transparent fuel cell. For the production of image labels, all data is divided into 6 classes based on the percentage of water coverage ratio, which the number of data in each class is unbalanced. To overcome this problem, random oversampling and undersampling have been used. Images and classes are considered as the input and output of the deep neural network, respectively. The deep network structure used has 4 convolution layers and 2 fully connected layers, which with the help of genetic algorithm optimization, the number of nodes has been obtained 4, 41, 64, 120, 99 and 6, respectively. After network training, the deep neural network model is able to estimate the amount of water in the fuel cell channels by receiving fuel cell images. Accuracy of 96.77% and 94.23% was obtained in training and testing data. In addition, this network is 10 times faster than the image processing method, and the area of water accumulation in the gas channel can be identified by robustness to environmental conditions
  9. Keywords:
  10. Proton Exchange Membrane (PEM)Fuel Cell ; Deep Neural Networks ; Intelligent Optimization ; Fuzzy Control ; Mathematical Modeling ; Drying ; Flooding

 Digital Object List

 Bookmark

...see more