OPTIMAL BLENDING OF POLANGA OIL WITH DIESEL IN CI ENGINE USING ARTIFICIAL NEURAL NETW
Active In SP
Joined: Jun 2010
27-09-2010, 11:08 AM
OPTIMAL BLENDING OF POLANGA OIL WITH DIESEL IN CI ENGINE USING ARTIFICIAL NEURAL NETWORKS.doc (Size: 802 KB / Downloads: 71)
This article is presented by:
Department of Mechanical Engineering, Indira Gandhi Instutute of Technology, Sarang, Dhenkanal, 759146, Orissa.
This project and implimentation deals with optimization of blending percentage of polanga oil with mineral diesel used in a conventional single cylinder compression ignition diesel engine. The various performance parameters for the engine are chosen to be load and speed as input & BHP, SFC, O2, NOx, CO emissions as the output. The various blending are 0%, 10%, 20%, 30%, 40% & 50% of polanga with an energy wise substitution of mineral diesel. With a variable load & speed combination the corresponding SFC, BHP & emissions of the above three gases in exhaust are to be experimentally found out. The data so obtained is confined to a finite number of observations spread over discrete points. For better understanding of performance the parameters must be obtained for all possible points in the operating range. For this artificial neural network, a soft computing method is to be used. The input parameters to the network are chosen to be load, speed & %blending. Output parameters SFC, BHP & emission of O2, NOx & CO etc are to be predicted at all possible points by the ANN model. Using some of the experimental data for training, an ANN model based on standard back-propagation algorithm for the engine is to be developed. Then, the performance of the ANN predictions are to be measured by comparing the predictions with the experimental results which are not used in the training process. The acceptable error in the process is confined within (3-6) %. From these experimental & predicted values an optimized blending is to be determined on the basis of maximum load with minimum SFC & emissions. The nature of the proposed ANN model is Multilayer perceptron, sigmoid activation function, LMSE with stochastic gradient descent error minimization. Thus the study will show an alternative to classical modeling technique of engine and provide an optimized blending for polanga oil in diesel engine.
Key words- ANN, back propagation algorithm, multi layer perceptron, sigmoid activation function, LMSE, Stochastic gradient descent error minimization.
In these days of energy crisis and global warming renewable energy recourses in general and alternate fuel in specific are gaining momentum worldwide. Although alternate fuels are a viable option but they posses some constraints even more severe than mineral diesel. These fuels are costly to convert to biodiesel form; posses lower energy density, substantially variable self ignition property & non compatible knocking properties. Environmentally also a 100% biodiesel may be less eco-friendly than 100% mineral diesel. More ever the magnitude of production of biodiesel is much inferior as compared to the actual consumption of petroleum products. Hence, it is always recommended to find suitable blending of biodiesel in the conventional mineral diesel so that we can obtain same operational parameters as that of mineral diesel which will be more eco-friendly, economic & reduce the burden on mineral diesel. Again the source of biodiesel is quite diverse.
Starting from Jatropha to rape seed and from sugarcane to alcohol a variety of bio-fuels can be converted to biodiesel. Our project and implimentation deals with a special fuel extensively found in the costal belt of eastern & western India locally known as polanga. The energy density of this seed is too high. Traditionally this has been used as a lightning fuel for years. Our project and implimentation evaluates the optimal blending of this fuel in mineral diesel so as to operate at maximum load with minimum SFC, maximum BHP & minimum emissions of gases like O2, NOx & CO etc. Initially a number of experiments are performed with a variable load & speed combination to obtain corresponding values of other parameters in various blends. Now we can presume that in an engine testing the input & output parameters are nonlinearly related to each other. Depending upon this presumption we can go for a neural network system with the number of nodes equal to the number of input parameters to the engine & the output nodes as the number of output parameters from the engine. The number of hidden nodes is decided by plotting graphs between fractional variance & number of neurons. The maxima point indicates the number of optimal neurons for the system. The nature of the ANN model is a multilayer perceptron, sigmoid activation function, LMSE with stochastic gradient descent error minimization.
After training the system the complete performance characteristics are forecasted with a reasonable error. From both graphical & mathematical analysis of various blending an optimal blending is chosen.