FAULT DIAGNOSIS IN DEAERATOR USING NEURAL NETWORKS
Shankar Srinivasan, Pradeep Mohan Kumar Kanagasabapathy, Narayanasamy Selvaganesan · Iranian journal of electrical and computer engineering · 2007
sets and it overcomes the drawback of the local criteria in nature because it uses more than one neighbor [15]. In engineering, the most straightforward application of SOM is in the identification and abstraction of hidden information from high-dimensional raw data. The SOM converts a complex, nonlinear relationship between highdimensional data into simple geometric relationship on a low-dimensional display [16]. Thus, it compresses the information while preserving the most important topological relationship of the primary data elements. In this paper, SOM, BP and RBF neural networks are used for diagnosing the fault in the deaerator. The faults simulated are leakage in the tank, sedimentation deposition in the tank, positive and negative bias in the water inlet valve, positive and negative bias in the steam valve, steam mixing with water in preheater and decrease in inlet temperature of water. The rest of the paper is organized as follows. Section 2 deals with neural network learning. Section 3, deals with the working of the deaerator. In section 4, fault simulation with different types of neural networks is presented. Finally, conclusions are made in section 5. II. NEURAL NETWORK LEARNING Fault diagnosis technology is a popular method used to identify problems and the causes for the problems. Neural network is one among the tools used for fault detection and identification. Many types of neural networks can be used to accomplish fault identification very well. This paper uses BP neural network, SOM neural network and RBF neural network for fault diagnosis in deaerator. In supervised learning method, training of network (BP, RBF) is accomplished by presenting a sequence of training vectors, or patterns, each with an associated target output vector. The weights are adjusted in BP network to make the actual response of the network move closer to the desired response. The back propagation algorithm used here is conjugate gradient algorithm. The algorithm combines the model-trust regional approach, used in LevenbergMarquardt algorithm, with the conjugate gradient approach.