A new and unique directional weighted error function was introduced into the backpropagation algorithm used in Artificial Neural Networks (ANNs) for applications where yes or no decisions are made on the output. A continuous error function based on a weighted curve is suggested for use in the backpropagation algorithm in an effort to increase the number of correct decisions. Results were compared to the standard and weighted error methods. A higher number of correct decisions were made with the new method. / Thesis (M.S.)--Wichita State University, College of Engineering, Dept. of Electrical and Computer Engineering. / "July 2006." / Includes bibliographic references (leaves 36-37).

A new methodology was developed for flow regime identification in pipes. The method utilizes the pattern recognition abilities of Artificial Neural Networks and the unprocessed time series of a system-monitoring-signal. The methodology was tested with synthetic data from a conceptual system, liquid level indicating Capacitance signals from a Horizontal flow system and with a pressure difference signal from a S-shape riser. The results showed that the signals that were generated for the conceptual system had all their patterns identified correctly with no errors what so ever. The patterns for the Horizontal flow system were also classified very well with a few errors recorded due to original misclassifications of the data. The misclassifications were mainly due to subjectivity and due to signals that belonged to transition regions, hence a single label for them was not adequate. Finally the results for the S-shape riser showed also good agreement with the visual observations and the few errors that were identified were again due to original misclassifications but also to the lack of long enough time series for some flow cases and the availability of less flow cases for some flow regimes than others. In general the methodology proved to be successful and there were a number of advantages identified for this neural network methodology in comparison to other ones and especially the feature extraction methods. These advantages were: Faster identification of changes to the condition of the system, inexpensive suitable for a variety of pipeline geometries and more powerful on the flow regime identification, even for transitional cases.

In this thesis, we study the least trimmed squares artificial neural networks (LTS-ANNs), which are generalization of the least trimmed squares (LTS) estimators frequently used in robust linear parametric regression problems to nonparametric artificial neural networks (ANNs) used for nonlinear regression problems. Two training algorithms are proposed in this thesis. The first algorithm is the incremental gradient descent algorithm. In order to speed up the convergence, the second training algorithm is proposed based on recursive least squares (RLS). Three illustrative examples are provided to test the performances of robustness against outliers for the classical ANNs and the LTS-ANNs. Simulation results show that upon proper selection of the trimming constant of the learning machines, LTS-ANNs are quite robust against outliers compared with the classical ANNs.

<p>Today one of the greatest aims within the area of bioinformatics is to gain a complete understanding of the functionality of genes and the systems behind gene regulation. Regulatory relationships among genes seem to be of a complex nature since transcriptional control is the result of complex networks interpreting a variety of inputs. It is therefore essential to develop analytical tools detecting complex genetic relationships.</p><p>This project examines the possibility of the data mining technique artificial neural network (ANN) detecting regulatory relationships between genes. As an initial step for finding regulatory relationships with the help of ANN the goal of this project is to train an ANN to predict the expression of an individual gene. The genes predicted are the nuclear receptor PPAR-g and the insulin receptor. Predictions of the two target genes respectively were made using different datasets of gene expression data as input for the ANN. The results of the predictions of PPAR-g indicate that it is not possible to predict the expression of PPAR-g under the circumstances for this experiment. The results of the predictions of the insulin receptor indicate that it is not possible to discard using ANN for predicting the gene expression of an individual gene.</p>

This report examines the trends in sick leave usage among nurses in a hospital and aims at creating a forecasting model to predict sick leave usage on a weekly basis using the concept of artificial neural networks (ANN). The data used for the research includes the absenteeism (sick leave) reports for 3 years at a hospital. The analysis shows that there are certain factors that lead to a rise or fall in the weekly sick leave usage. The ANN model tries to capture the effect of these factors and forecasts the sick leave usage for a 1 year horizon based on what it has learned from the behavior of the historical data from the previous 2 years. The various parameters of the model are determined and the model is constructed and tested for its forecasting ability. / text

This thesis discusses aspects of design, implementation and theory of expert systems, which have been constructed in a novel way using techniques derived from several existing areas of Artificial Intelligence research. In particular, it examines the philosophical and technical aspects of combining techniques derived from the traditional rule-based methods for knowledge representation, with others taken from connectionist (more commonly described as Artificial Neural Network) approaches, into one homogenous architecture. Several issues of viability have been addressed, in particular why an increase in system complexity should be warranted. The kind of gain that can be achieved by such hybrid systems in terms of their applicability to general problem solving and ability to continue working in the presence of noise, are discussed. The first aim of this work has been to assess the potential benefits of building systems from modular components, each of which is constructed using different internal architectures. The objective has been to progress the state of knowledge of the operational capabilities of a specific system. A hybrid architecture containing multiple neural nets and a rule-based system has been designed, implemented and analysed. In the course of, and as an aid to the development of the system, an extensive simulation work-bench has been constructed. The overall system, despite its increased internal complexity provides many benefits including ease of construction and improved noise tolerance, combined with explanation facilities. In terms of undesirable features inherited from the parent techniques the losses are low. The project has proved successful in its stated aims and has succeeded in contributing a working hybrid system model and experimental results derived from the comparison of this new approach with the two, primary, existing techniques.

Artificial neural networks (ANNs) have been used in many fault detection and identification (FDI) applications due to their pattern recognition abilities. In this study, two ANNs, a supervised network based on Backpropagation (BP) learning and an unsupervised network based on Adaptive Resonance Theory (ART-2A), were tested for FDI on an automated assembly machine and compared to a conventional rule-based method. Three greyscale sensors and two redundant limit switches were used as cost-effective sensors to monitor the machine's operating condition. To test each method, sensor data were collected while the machine operated under normal conditions, as well as 10 fault conditions. Features were selected from the raw sensor data to create data sets for training and testing. The performance of the methods was evaluated with respect to their ability to detect and identify known, unknown and multiple faults. Their modelling and computational requirements were also considered as performance measures. Results showed that all three methods were able to achieve perfect classification with the test data sets; however, the BP method could not classify unknown or multiple faults. In all cases, the performance depended on careful tuning of each method’s parameters. The BP method required an ideal number of neurons in the hidden layer and good initialization. The ART-2A method required tuning of its classification parameter. The rule-based method required tuning of its thresholds. Although it was found that the rule-based system required more effort to set up, it was judged to be more useful when unknown or multiple faults were present. The ART-2A network created new outputs for these conditions, but it could not give any more information as to what the new fault was. By contrast, the rule-based method was able to generate symptoms that clearly identified the unknown and multiple fault conditions. Thus, the rule-based method was judged to be the best overall method for this type of application. It is recommended that future work examine the application of computer vision-based techniques to FDI with the assembly machine. The results from this study, using cost-effective sensors, could then be used as a performance benchmark for image-based sensors. / Thesis (Master, Mechanical and Materials Engineering) -- Queen's University, 2014-05-16 17:21:13.676

This thesis investigates the viability of using dynamic-based ‘damage fingerprints’ in combination with artificial neural network (ANN) techniques and principal component analysis (PCA) to identify defects in civil engineering structures. Vibration-based damage detection techniques are global methods and are based on the principle that damage alters both the physical properties, such as mass, stiffness and damping, as well as the dynamic properties of a structure. It is therefore feasible to utilise measured dynamic quantities, such as time histories, frequency response functions (FRFs) and modal parameters, from structural vibration to detect damage. Damage identification based on vibrational characteristics is essentially a form of pattern recognition problem, which looks for the discrimination between two or more signal categories, e.g., before and after a structure is damaged, or differences in damage levels or locations. Artificial neural networks are capable of pattern recognition, classification, signal processing and system identification, and are therefore an ideal tool in complementing dynamic-based damage detection techniques. Likewise, PCA has pattern recognition abilities and is capable of data reduction and noise filtering. With these characteristics, both techniques can help overcome limitations associated with previously developed vibration-based methods and assist in delivering more accurate and robust damage identification results. In this study, two types of dynamic-based damage identification methods are proposed. The first is based on the damage index (DI) method (initially proposed by Stubbs et al.), while the second approach uses changes in FRF data as damage fingerprints. The advantage of using damage patterns from the DI method, which is based on changes in modal strain energies, is that only measured mode shapes are required in the damage identification, without having to know the complete stiffness and mass matrices of the structure. The use of directly measured FRF data, which provide an abundance of information, is further beneficial as the execution of experimental modal analysis is not required, thus greatly reducing human induced errors. Both proposed methods utilise PCA and neural network techniques for damage feature extraction, data reduction and noise filtering. A hierarchical network training scheme based on network ensembles is proposed to take advantage of individual characteristics of damage patterns obtained from different sources (different vibrational modes for the DI-based method and different sensor locations for the FRF-based method). In the ensemble, a number of individual networks are trained in parallel, which optimises the network training and delivers improved damage identification outcomes. Both methods are first tested on a simple beam structure to assess their feasibility and performance. Then, the FRF-based method is applied to a more complicated structure, a two-storey framed structure, for validation purposes. The two methods are verified by numerical simulations and laboratory testing for both structures. As defects, notch type damage of different severities and locations are investigated for the beam structure. For the two-storey framed structure, three different types of structural change are studied, i.e. boundary damage, added mass changes and section reduction damage. To simulate field-testing conditions, the issue of limited sensor availability is incorporated into the analysis. For the DI-based method, sensor network limitations are compensated for by refining coarse mode shape vectors using cubic spline interpolation techniques. To simulate noise disturbances experienced during experimental testing, for the numerical simulations, measurement data are polluted with different levels of white Gaussian noise. The damage identifications of both methods are found to be accurate and reliable for all types of damage. For the DI-based method, the results show that the proposed method is capable of overcoming limitations of the original DI method associated with node point singularities and sensitivities to limited number of sensors. For the FRF-based method, excellent results are obtained for damage identification of the beam structure as well as of the two-storey framed structure. A major contribution is the training of the neural networks in a network ensemble scheme, which operates as a filtering mechanism against individual networks with poor performance. The ensemble network, which fuses results of individual networks, gives results that are in general better than the outcomes of any of the individual networks. Further, the noise filtering capabilities of PCA and neural networks demonstrate great performance in the proposed methods, especially for the FRF-based identification scheme.

This thesis investigates the viability of using dynamic-based ‘damage fingerprints’ in combination with artificial neural network (ANN) techniques and principal component analysis (PCA) to identify defects in civil engineering structures. Vibration-based damage detection techniques are global methods and are based on the principle that damage alters both the physical properties, such as mass, stiffness and damping, as well as the dynamic properties of a structure. It is therefore feasible to utilise measured dynamic quantities, such as time histories, frequency response functions (FRFs) and modal parameters, from structural vibration to detect damage. Damage identification based on vibrational characteristics is essentially a form of pattern recognition problem, which looks for the discrimination between two or more signal categories, e.g., before and after a structure is damaged, or differences in damage levels or locations. Artificial neural networks are capable of pattern recognition, classification, signal processing and system identification, and are therefore an ideal tool in complementing dynamic-based damage detection techniques. Likewise, PCA has pattern recognition abilities and is capable of data reduction and noise filtering. With these characteristics, both techniques can help overcome limitations associated with previously developed vibration-based methods and assist in delivering more accurate and robust damage identification results. In this study, two types of dynamic-based damage identification methods are proposed. The first is based on the damage index (DI) method (initially proposed by Stubbs et al.), while the second approach uses changes in FRF data as damage fingerprints. The advantage of using damage patterns from the DI method, which is based on changes in modal strain energies, is that only measured mode shapes are required in the damage identification, without having to know the complete stiffness and mass matrices of the structure. The use of directly measured FRF data, which provide an abundance of information, is further beneficial as the execution of experimental modal analysis is not required, thus greatly reducing human induced errors. Both proposed methods utilise PCA and neural network techniques for damage feature extraction, data reduction and noise filtering. A hierarchical network training scheme based on network ensembles is proposed to take advantage of individual characteristics of damage patterns obtained from different sources (different vibrational modes for the DI-based method and different sensor locations for the FRF-based method). In the ensemble, a number of individual networks are trained in parallel, which optimises the network training and delivers improved damage identification outcomes. Both methods are first tested on a simple beam structure to assess their feasibility and performance. Then, the FRF-based method is applied to a more complicated structure, a two-storey framed structure, for validation purposes. The two methods are verified by numerical simulations and laboratory testing for both structures. As defects, notch type damage of different severities and locations are investigated for the beam structure. For the two-storey framed structure, three different types of structural change are studied, i.e. boundary damage, added mass changes and section reduction damage. To simulate field-testing conditions, the issue of limited sensor availability is incorporated into the analysis. For the DI-based method, sensor network limitations are compensated for by refining coarse mode shape vectors using cubic spline interpolation techniques. To simulate noise disturbances experienced during experimental testing, for the numerical simulations, measurement data are polluted with different levels of white Gaussian noise. The damage identifications of both methods are found to be accurate and reliable for all types of damage. For the DI-based method, the results show that the proposed method is capable of overcoming limitations of the original DI method associated with node point singularities and sensitivities to limited number of sensors. For the FRF-based method, excellent results are obtained for damage identification of the beam structure as well as of the two-storey framed structure. A major contribution is the training of the neural networks in a network ensemble scheme, which operates as a filtering mechanism against individual networks with poor performance. The ensemble network, which fuses results of individual networks, gives results that are in general better than the outcomes of any of the individual networks. Further, the noise filtering capabilities of PCA and neural networks demonstrate great performance in the proposed methods, especially for the FRF-based identification scheme.

A new methodology was developed for flow regime identification in pipes. The method utilizes the pattern recognition abilities of Artificial Neural Networks and the unprocessed time series of a system-monitoring-signal. The methodology was tested with synthetic data from a conceptual system, liquid level indicating capacitance signals from a Horizontal flow system and with a pressure difference signal from a S-shape riser. The results showed that the signals that were generated for the conceptual system had all their patterns identified correctly with no errors whatsoever. The patterns for the Horizontal flow system were also classified very well with a few errors recorded due to original misclassifications of the data. The misclassifications were mainly due to subjectivity and due to signals that belonged to transition regions, hence a single label for them was not adequate. Finally the results for the S-shape riser showed also good agreement with the visual observations and the few errors that were identified were again due to original misclassifications but also to the lack of long enough time series for some flow cases and the availability of less flow cases for some flow regimes than others. In general the methodology proved to be successful and there were a number of advantages identified for this neural network methodology in comparison to other ones and especially the feature extraction methods. These advantages were: Faster identfication of changes to the condition of the system, inexpensive suitable for a variety of pipeline geometries and more powerful on the flow regime identification, even for transitional cases.