Research Areas

Artificial Neural Networks in Engineering Design and Engineering Applications

Christos N. Schizas and Costas Neocleous, Soteris Kalogirou, Silas Chr. Michaelides.

The power of artificial neural networks to model engineering systems, primarily in engineering design and control, is being applied to various problems. Applications have been developed on the design of marine propellers, concentrating solar collectors, and heating and air-conditioning load estimations, and power load forecasting.


Related Papers

Title : Artificial neural networks in marine propeller design.
Authors : Costas Neocleous, Christos N. Schizas
Abstract : Various neural network systems were developed for examining propeller performance data that were derived experimentally. This study aimed to establish an accurate mapping thus facilitating propeller selection during the design process. Different neural network architectures and learning rates were tested, aiming at establishing a near optimum setup. It is evident from the findings, that this technology can be used effectively in modeling the performance of a series of marine propellers and thus may be used for propeller selection, and for extrapolation to new designs. 


Title : Comparative study of methods for estimating intercept factor of parabolic trough collectors.
Authors : Soteris Kalogirou, Costas Neocleous, Christos N. Schizas
Abstract : One of the parameters used for the evaluation of a parabolic trough collector performance is optical efficiency. This depends on the properties of the various materials employed in the construction of the collector, the collector dimensions, the angle of incidence and the intercept factor (ã). The intercept factor depends on the size of the receiver, the surface angle errors of the parabolic mirror, and on solar beam spread. A ray-trace computer code called EDEP (Energy DEPosition computer code) is used by Guven and Bannerot (1985) to calculate the intercept factor. The intercept factor can also be calculated by a closed-form expression developed by Guven and Bannerot (1985). This expression considers both random and non-random errors. These errors are encountered in the construction and/or in the operation of the collector. An artificial neural network was trained to learn the ã-values based on the input data of collector rim angle, random and non-random errors, and the EDEP results. The output is compared with the EDEP results which are considered to be the most accurate, the results of a simple program developed by Guven (1987) using the trapezoidal integration method, and a multiple linear regression analysis. From all the above it is shown that the results obtained by the artificial neural network system approximates the results of the ray-trace model, extremely well with an R2-value equal to 0.999.



Title : Building heating load estimation using artificial neural networks. 
Authors : Soteris Kalogirou, Costas Neocleous, Christos N. Schizas 
Abstract : A number of commercial software programs are currently available for the estimation of the heating loads of buildings. These programs basically perform simple multiplications between the areas of the various building envelope components with the corresponding U-values and the effective temperature difference. The results of these multiplications are then added to obtain the actual heating load on which a 5% safety factor is usually added. A novel approach is presented in this paper. The objective is to train an artificial neural network (ANN) to be able to predict the building heating loads with the minimum of input data. An ANN has been trained based on 250 actual data varying from very small rooms to large spaces of 100 m2 floor area. The type of rooms varied from small toilets to large classroom halls while the room temperatures varied from 18°C to 23°C. In addition to the above an attempt was made to use a large variety of room characteristics e.g. rooms without windows, rooms without external walls and windows etc. In this way the network was trained to accept and handle a number of unusual cases. The data input were the areas of windows, walls, partitions and floors, the type of windows and walls, and the classification on whether the space has roof or ceiling. The output training data was the actual heating load. Preliminary results on the training of the network showed that the accuracy of the prediction could be improved by separating the input data into two categories one with spaces of floor areas up to 7 m2 and one with spaces of floor areas from 7 to 100 m2. The statistical R2-value of the training data set was equal to 0.988 for the first case and 0.999 for the second. Unknown data were subsequently used to investigate the accuracy of prediction. Predictions within 10% for the first case and 9% for the second were obtained. These results indicate that the present method can successfully be used for the prediction of a building’s heating load. 


Title : Artificial neural networks for modeling the starting-up of a solar steam generator.
Authors :

Soteris Kalogirou, Costas Neocleous, Christos N. Schizas 

Abstract :

An experimental solar steam generator, consisting of a parabolic trough collector, a high pressure steam circuit, and a suitable flash vessel has been constructed and tested in order to establish the thermodynamic performance during heat-up. Preliminary tests demonstrated that the heat-up energy requirement has a marked effect on the system performance since solar energy collected during the heating-up period is lost at night due to the diurnal cycle. This depends mostly on the dimensions and the inventory of the flash vessel, and the prevailing environmental conditions. Experimental data were obtained and used to train an artificial neural network in order to implement a mapping between easily measurable features (environmental conditions, water content, vessel dimensions) and the system temperatures. Such mapping may be useful to system designers when seeking to find the optimum vessel dimensions. The trained network predicted very well the response of the system, as indicated by an obtained statistical R-squared value of 0.999 and a maximum deviation between predicted and actual values confined to less than 3.9 %. This degree of accuracy is acceptable in the design of such systems. The results are important, because the system was tested during its heat-up cycle, under transient conditions, which is quite difficult to model analytically.




Title : Artificial neural networks in energy applications: A review. 
Authors : Soteris Kalogirou, Costas Neocleous, Christos N. Schizas 
Abstract : Artificial neural networks are widely accepted as a technology offering an alternative way to tackle complex and ill specified problems. They can learn from examples, are fault tolerant in the sense that they are able to handle noisy and incomplete data, are able to deal with non-linear problems, and once trained can perform prediction and generalisation at high speed. They have been used in diverse applications in control, robotics, pattern recognition, forecasting, medicine, power systems, manufacturing, optimisation, signal processing, and social/psychological sciences. They are particularly useful in system modelling such as in implementing complex mappings and system identification. This paper presents various applications of neural networks in energy problems in a thematic rather than a chronological or any other order. Artificial neural networks have been used by the authors in the field of solar energy, for modelling the heat-up response of a solar steam generating plant, for the estimation of a parabolic trough collector intercept factor, for the estimation of a parabolic trough collector local concentration ratio and for the design of a solar steam generation system. They have also been used for the estimation of heating loads of buildings. In all those models a multiple hidden layer architecture has been used. Errors reported in these models are well within acceptable limits, which clearly suggest that artificial neural networks can be used for modelling in other fields of energy production and use. The work of other researchers in the field of energy is also reported. This includes the use of artificial neural networks in heating ventilating and air-conditioning systems, solar radiation, modelling and control of power generation systems, and load forecasting. 


Title : Artificial neural networks for the estimation of the performance of a parabolic trough collector steam generation system. 
Authors :

Soteris Kalogirou, Costas Neocleous, Christos N. Schizas 

Abstract : The parabolic trough collector (PTC) is the most preferred type of solar collecting system employed for steam generation. This is due to the fact that the collector can work with high efficiencies at high temperature. Such a system consists of a PTC, a flash vessel and an associated circulation and control system. The amount of steam produced depends on the solar radiation available, ambient air temperature, collector area, total system water capacity, and the diameter and height of the flash vessel. Data for a number of cases were used to train an artificial neural network in order to generate a mapping between the above easily measurable inputs and the desired output system performance. The collectors used had areas varying from 3.5 m2 to 2160 m2. Different neural architectures have been used in order to find the one with the best possible performance. The multilayer feedforward architecture using the standard backpropagation learning algorithm have been the most successful so far. Many other architectures are still under investigation. The results obtained for the training set are such that they yield a statistical R2 = 0.999. The network was used subsequently for predictions of the performance of cases other than the ones used for training, both within and outside the above range. Typical value of the accuracy obtained was 95% (at an R2 = 0.999). This is considered as acceptable for such estimations


Title : Artificial neural networks for the estimation of U-values.
Authors : Soteris Kalogirou, Costas Neocleous, Christos N. Schizas 
Abstract : An important parameter for the estimation of building thermal loads is the U-value. An accurate value will enable a more correct estimation of the load. Neural networks are widely accepted as a technology offering an alternative way to tackle complex and ill specified problems. They can learn from examples, are fault tolerant in the sense that they are able to handle noisy and incomplete data, are able to deal with non-linear problems, and once trained can perform prediction and generalization at high speed. Artificial neural networks have been used in diverse applications in control, robotics, pattern recognition, forecasting, medicine, power systems, manufacturing, optimisation, signal processing, and social/psychological sciences. They have shown to be particularly useful in system modelling such as implementing complex mappings and for system identification. The objective of this study is to investigate the suitability of artificial neural networks as tools for the estimation of U-values quickly and accurately. This will facilitate the work of design engineers in the field. The data presented as input to the network were the thicknesses of hollow and solid bricks, concrete, various types of insulating materials, air gap, and plaster. Materials which are not used in a construction are given a thickness equal to zero. The network output is the U-value. The data provided to the network for training are values of U obtained from well established references. The statistical R2-value for the training set was equal to 0.9999. The network then, will hopefully learn the basic mechanics of calculation, and thus be able to generalise to an unknown set of data. Unknown data, i.e., constructions with material thicknesses not known to the network, were subsequently used to investigate the accuracy of the prediction. Predictions with R2-value equal to 0.9817 were obtained. This is considered satisfactory for design purposes. Although estimated U-values have been used in this work, if experimental values are available these will enhance the capability of the method, as the prediction will not depend on handbook recorded data but on real values. It should be noted that in this case it would not be required to perform experiments with all combinations of materials but only a limited number for each different combination of materials.



Title : An artificial neural network for the short-term forecasting of electric power load
Authors : Thalis Papazoglou, Costas Neocleous, Christos N. Schizas 
Abstract : A feedforward multilayer neural network has been used for the estimation of the four-hour ahead power load in Crete. An attempt was made to use few variables for the input vector, while keeping the degree of accuracy to acceptable levels. Different neural network architectures, activation functions and learning rates were tried, aiming at finding the network that resulted in the best possible estimation. A feedforward two hidden layer architecture using backpropagation was chosen. The correlation coefficient obtained for both the training and the unknown testing data sets was 0.984. The prediction error for 90% of the cases was confined to less than 10 MW, which is considered adequate. The neural network results have also been compared to a multiple linear regression and a multiple non-linear regression analysis. In both cases the neural network approach is slightly better.



Title : Wind Speed Prediction Using Artificial Neural Networks
Authors :

Soteris Kalogirou, Costas Neocleous, S. Pashiardis, Christos N. Schizas  

Abstract : A multilayered artificial neural network has been used for predicting the mean monthly wind speed in regions of Cyprus where data are not available. Data for the period 1986-1996 have been used to train a neural network, whereas data for the year 1997 were used for validation. Both learning and prediction were performed with adequate accuracy. Two network architectures of the similar type have been tried. One with eleven neurons in the input layer and one with five. The second one proved to be more accurate in predicting the mean wind speed. The maximum percentage difference for the validation set was confined to less than 1.8% on an annual basis, which is considered by the domain expert as adequate.

 

 

Key People

Kalogirou Soteris A.
(Higher Technical Institute, Cyprus)

skalogir@spidernet.com.cy
Michaelides Silas Chr.
(Meteorological Service, Cyprus)

cssilas@ucy.ac.cy
Neocleous Costas C.
(Higher Technical Institute, Cyprus)

costas@ucy.ac.cy
Papazoglou Thalis.
Pashiardis S.
Schizas Christos N.
(
University of Cyprus)
schizas@ucy.ac.cy