When applied correctly, a neural or adaptive
system may considerably outperform other methods. Also, these adaptive systems have been configured delivering best results, in many other areas of important engineering applications such as signal enhancement, noise cancellation, prediction, and control. NA
George Sachinoglou AMRINA, has
Fig.5. Performance of ANN, NEWWAT2 & Newwat3 tested on the 7 previoulsy unseen Container Ships.
Te last test showed comparisons of the
Table 1. ANN tested on Lightship Weight results of previously unseen container ships, extraced from Significant Ships 1993-1994.
known and well established mathematical models, such as regression techniques. For given case-values of the 16 variables, they enter the (17x1) input vector x , (Fig. 2). Input vector follows a hyperbolic tangent activation process and the addition of a bias equal to one. Te product of the (4x17) W1 matrix and the hyperbolic tangent function of x follows and the (4 x 1) matrix is the result, undergoes hyperbolic tangent activation again, along with a bias addition on top of the four rows, i.e. on top of the four hidden units. The resultant (5x1) matrix is multiplied by the (1x5) transpose of W2. Te result of the above multiplication is a single value which passes through the hyperbolic tangent function and the estimate of lightship weight is collected in tonnes.
ANN Validity Tests on Previously Unseen Container Ship Data Firstly, the ANN was tested on generalisation performance using seven previously unseen to it container ships (Table 2), from Significant Ships 1993 - 1994, [3]. Four out of seven lightship results provided values which deviated within + _ 10% from their corresponding actual values. Tis test gave a satisfactory initial impression of the ANN model as a reasonably good lightship weight approximator. In case the ANN had been trained on 100
container ships the accuracy of its predictions would be much improved, and having 200 container ships even more accurate.
The Naval Architect November 2009
ANN computed lightship for the seven test-ships and the ones computed by NEWWAT2 and Newwat3. NEWWAT2 and Newwat3 are computer programs designed for the determination of design particulars for deadweight oriented vessels with dimensional ratio combinations of about L/B = 6.12 and T/D = 0.72
, B/D = 1.86 . Te results can be found on Table
7.6 at [9]. Te ANN performed within 10% error
with a 60% successful estimate, 43% for NEWWAT2 and 14% for Newwat3. The interesting and useful observation here is that NEWWAT2 and Newwat3 deviated with the same error in all seven test ships while the ANN did not, (Fig. 5). This indicates the useful comparison that can be made between statistical correlations – empirical formulae and the nonlinear approximator.
Conclusions Artificial Neural Networks have recently become the focus of much attention, largely because they can identify and learn correlated patterns between sets of input data and corresponding target values. Tey mimic the human learning process and can handle problems involving highly nonlinear and complex data. Aſter training ANNs are used to predict outcomes from new input data. It is found that the ANN method can
accurately predict the initial lightship weight estimate of container carrier designs as with empirical formulae derived by regression analysis. Nevertheless, it must be noted that the main objective of this paper was to investigate and demonstrate the applicability, weaknesses and strengths of using ANNs in developing design formulae.
graduated from TEI of Athens (2006), Department of Marine Engineering & Naval Architecture. Graduated (2009) from School of Marine Science & Technology of Newcastle University, BEng (Hons) Naval Architecture, MSc (Hons) Naval Architecture. Contact email: ambrose_1984@
hotmail.com
References [1] Wright, P. N. H., (2008), - “Lightship Mass Estimate and Deadweight Check’’, Ship Design MAR3002, School of Marine Science & Technology, Newcastle University [2] Watson, D. G. M., (1998) – “Practical Ship Design’’, Elsevier Science Ltd., ISBN 0-08-044054-1 [3] Significant Ships (1993 – 2007), Te Royal Institution of Naval Architects (RINA) [4] Buxton, I.L., (2009) – “Appointing ship design parameters for each ship type’’, Personal Discussion, School of Marine Science & Technology, Newcastle University [5] Barrass, B., (2009) – “Sources for Extracting Ship Design Data’’, Personal Discussion, School of Marine Science & Technology, Newcastle University [6] Pu, Y. & Mesbahi, E. (2006) – “Application of artificial neural networks to evaluation of ultimate strength of steel panels’’, Engineering Structures 28 (2006) pp. 1190–1196, School of Marine Science and Technology, Newcastle University [7] Lefebvre, C., Principe, J., (2000) – “Neural and Adaptive Systems: Fundamentals Trough Simulations’’, John Wiley & Sons, Inc., ISBN: 0-471-35167-9 [8] ASME PTC 19.1 (1998) - “Test Uncertainty’’, The American Society of Mechanical Engineers, pp. 1-3, pp. 9-12, pp. 23-25, pp. 37-39 B89.7.4.1 [9] Sachinoglou, G., (2009) – “An Artificial Neural Network Model for Lightship Mass Prediction of New Container Ship Designs”, MSc. Tesis, School of Marine Science & Technology, Newcastle University.
25
In-depth
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52 |
Page 53 |
Page 54 |
Page 55 |
Page 56 |
Page 57 |
Page 58 |
Page 59 |
Page 60 |
Page 61 |
Page 62 |
Page 63 |
Page 64 |
Page 65 |
Page 66 |
Page 67 |
Page 68