Javascript must be enabled for the correct page display

The influence of fluid model complexity on optimal sensor configuration for object localization in water

Naber, Niek (2019) The influence of fluid model complexity on optimal sensor configuration for object localization in water. Bachelor's Thesis, Artificial Intelligence.


Download (781kB) | Preview
[img] Text
Restricted to Registered users only

Download (119kB)


The lateral line is an organ which fish use to sense their surroundings. Lateral lines consist of several neuromasts which serve as sensors of fluid flows. These fluid flows are used to make estimates about objects in the fish's surroundings. Artificial lateral lines can also be used to estimate objects positions and orientations in water. Different simulation methods exist to simulate water, where some do include viscosity, and others do not. This study researches differences between these simulation methods, the way extreme learning machines need to be trained on fluid velocities of the simulations to predict an objects location, the prediction errors of those extreme learning machines and the optimal sensor positions assigned by a genetic algorithm. Because of the noisy effects of viscosity, extreme learning machines that use viscous simulations need more training samples to reach their optimal performance. Their optimal performance is worse than that of extreme learning machines that use non-viscous simulations, especially when grid-like sensor set ups are used. The extreme learning machines that use viscous simulations seem to benefit more when fluid velocities are sensed at a lower height below objects, and when sensors are placed more to the outside of a sensor plane, as compared to a grid like setup. When an extreme learning machine is trained on either one of the simulations, there seem to be major differences in prediction errors when we vary the positions that are used to measure the fluid velocities. When a genetic algorithm is used to find the sensor configuration with which the extreme learning machine has the lowest prediction error, sensor positions seem to be pushed to the sides of the sensor plane. This study had limited computational and storage capacity, therefore findings can be improved by further research which takes higher resolution simulations, and bigger sample sizes.

Item Type: Thesis (Bachelor's Thesis)
Supervisor nameSupervisor E mail
Netten, S.M.
Degree programme: Artificial Intelligence
Thesis type: Bachelor's Thesis
Language: English
Date Deposited: 12 Jul 2019
Last Modified: 16 Jul 2019 09:05

Actions (login required)

View Item View Item