Description:
Some of the major concerns regarding sewer overflows to receiving water bodies include serious environmental, aesthetic and public health problems. Water management authorities are increasingly receiving public complaints that have led engineers to focus on means of retaining the entrained sewer solids within the sewer system during overflow events. During wet weather conditions, sewer overflows to receiving water bodies raise serious concern to environmental and community health concerns. To address these problems, different types of screening devices are used. Moreover, floatable control is preferred by most of the proposed and existing environmental regulations. This requirement triggers the need to research the different types of screening devices and screenings handling systems to select the most appropriate for a particular installation especially at unmanned locations. In the present study the sewer overflow device consists of a rectangular tank and a sharp crested weir that are followed by series of vertical parallel combs to separate entrained sewer solids from the overflow. The device does not require electrical or mechanical power for the self-cleansing mechanism, enabling the device to work efficiently in unmanned locations. Extensive laboratory investigations are underway to assess the effectiveness of a novel self-cleansing sewer overflow screening device. A series of laboratory tests to determine trapping efficiencies for common sewer solids were conducted for different flow conditions, number of combs layers, spacing of combs and weir crest lengths. Sewer solids from different density materials make sewer flow to analyze in complex Non-Newtonian fluid system with huge computational cost and complicity using physical law based modeling. On the flipside artificial neural model has the capacity to accurately predict the outcome of complex, non-linear physical systems with relatively poorly understood physicochemical processes which makes them highly desirable in the present study. Artificial Neural Networks (ANN) have already been successfully used to simulate flood forecasting in urban drainage system, real time control in combined sewer system, real time water level predictions of sewerage systems covering gauged and un-gauged sites etc. In case of sewer solid capture efficiency: neural network modeling is able to recognize nonlinear input output relations with adapting approach for changing circumstances. In the present study, feed forward artificial neural networks using back propagation algorithms were used, as such networks have been used almost exclusively in environmental modeling. A series of forty seven (47) sets of experimental data were collected to train (calibrate) the ANN model. In addition to these, eight (8) sets of experimental data were collected to validate the trained ANN network to be used in wider prospective of urban drainage conditions. The major areas covered in the ANN modeling include selection of input and output variables, optimization of the model, consideration of different learning algorithms, designing ANN's training & cross training processes and model validation. In the studied case, complex physical characteristics of different sewer solids, together with multi-fluid sewer system with variable flow phenomena makes it difficult to model with physical considerations. In case of sewer solid capture efficiency; artificial neural network modeling is able to learn the complex input-output relations with adapting approach for changing circumstances. Model considered different learning algorithms, diverse hidden layer structure with varied training samples to optimize the network. It is found that the model can successfully predict the experimental results with average absolute percentage errors varying from 4 to 7 percent.
Description:
This paper presents a comparative study conducted on the performance of the commonly used object tracking and location prediction algorithms for mobile robot navigation in a dynamically cluttered and GPS-denied mining environment. The study was done to test the different algorithms for the same set criteria (such as accuracy and computational time) under the same conditions.The identified commonly used algorithms for object tracking and location prediction of moving objects used in this investigation are Kalman filter (KF), extended Kalman filter (EKF) and particle filter (PF). The study results of those algorithms are analyzed and discussed in this paper. A trade-off was apparent. However, in overall performance KF has shown its competitiveness.The result from the study has found that the KF based algorithm provides better performance in terms of accuracy in tracking dynamic objects under commonly used benchmarks. This finding can be used in development of an efficient robot navigation algorithm.
Description:
Proceedings of the IEEE International Conference on Industrial Technology
Description:
This paper presents a summary of research which was conducted in developing a new human-like navigation methodology based on the Agoraphilic algorithm. This new methodology is capable of maneuvering robots in both static and dynamically clutter unknown environments. The Agoraphilic algorithm is an "optimistic" navigation algorithm. The algorithm is based on free space attraction rather than repulsion of obstacles for navigation. Therefore, this algorithm directs robots to follow the free space leading to the goal instead of avoiding obstacles. This approach has eliminated many draw backs of the traditional APF algorithm. However, the major limitation of the previously developed Agoraphilic algorithm could only deal with static environment. The new proposed algorithm has successfully extended the capacity of Agoraphilic algorithm to deal with environment cluttered with dynamic obstacles. The new Agoraphilic algorithm uses a tracking and prediction methodology to estimate the path of unknown moving objects. The estimated locations of the moving objects are combined with static object locations in the robot's visible region to generate time-varying free space attractive forces. These time varying forces maneuver the robot to the goal in dynamically cluttered unknown environment without collusions. To demonstrate the algorithm's ability, several simulations were performed. Furthermore, the new algorithm was tested and bench-marked against other APF published work under similar environment and conditions. The comparative results are discussed and showed the effectiveness of the new Algoraphilic navigation algorithm.
Description:
A functional prototype of a friction-based object slippage detection gripper for robotic grasping and manipulation has been designed and built. Object grasping and manipulation experiments have been successfully performed to study the appropriateness of the methodology and the newly built slippage detection gripper. The main advantage of this slippage detection method is that slippage detection is an inherent capability of the sensing element, and not a derived capability like that of sensors based on vibration. This slippage detection and control strategy is simple by design and low in cost, but robust in function. It has the potential to be used in a variety of environments such as high temperatures, low temperatures and underwater. The robustness of the design makes it highly suitable for grasping and manipulating safely a large range of object weights and sizes.
Description:
This paper presents a newly developed parallel gripper prototype capable of sensing grasp force, tangential force and slippage. In this design the gripper itself is used as part of the sensing strategy rather than just being simply a structural support for other sensors. The sensing capability of this gripper is simple in design and reliable. The sensing strategy can be customised to specific applications such as the ability to handle large loads while maintaining its ability to detect slippage as reliably as when handling lighter loads.
Description:
This work identifies an important, previously unaddressed issue for regression based on neural networks – learning to accurately approximate problems where the output is not a function of the input (i.e. where the number of outputs required varies across input space). Such non-functional regression problems arise in a number of applications, and can not be adequately handled by existing neural network algorithms. To demonstrate the benefits possible from directly addressing non-functional regression, this paper proposes the first neural algorithm to do so – an extension of the Resource Allocating Network (RAN) which adds additional output neurons to the network structure during training. This new algorithm, called the Resource Allocating Network with Varying Output Cardinality (RANVOC), is demonstrated to be capable of learning to perform non-functional regression, on both artificially constructed data and also on the real-world task of specifying parameter settings for a plasma-spray process. Importantly RANVOC is shown to outperform not just the original RAN algorithm, but also the best possible error rates achievable by any functional form of regression.
Description:
This work identifies an important, previously unaddressed issue for regression based on neural networks – learning to accurately approximate problems where the output is not a function of the input (i.e. where the number of outputs required varies across input space). Such non-functional regression problems arise in a number of applications, and can not be adequately handled by existing neural network algorithms. To demonstrate the benefits possible from directly addressing non-functional regression, this paper proposes the first neural algorithm to do so – an extension of the Resource Allocating Network (RAN) which adds additional output neurons to the network structure during training. This new algorithm, called the Resource Allocating Network with Varying Output Cardinality (RANVOC), is demonstrated to be capable of learning to perform non-functional regression, on both artificially constructed data and also on the real-world task of specifying parameter settings for a plasma-spray process. Importantly RANVOC is shown to outperform not just the original RAN algorithm, but also the best possible error rates achievable by any functional form of regression.