Yes / In engineering design, there are many functional relationships which are difficult to express into a simple and exact mathematical formula. Instead they are documented within a form of line graphs (or plot charts or curve diagrams) in engineering handbooks or text books. Because the information in such a form cannot be used directly in the modern computer aided design (CAD) process, it is necessary to find a way to numerically represent the information. In this paper, a data processing system for numerical representation of line graphs in mechanical design is developed, which incorporates the process cycle from the initial data acquisition to the final output of required information. As well as containing the capability for curve fitting through Cubic spline and Neural network techniques, the system also adapts a novel methodology for use in this application: Grey Models. Grey theory have been used in various applications, normally involved with time-series data, and have the characteristic of being able to handle sparse data sets and data forecasting. Two case studies were then utilized to investigate the feasibility of Grey models for curve fitting. Furthermore, comparisons with the other two established techniques show that the accuracy was better than the Cubic spline function method, but slightly less accurate than the Neural network method. These results are highly encouraging and future work to fully investigate the capability of Grey theory, as well as exploiting its sparse data handling capabilities is recommended.

Early detection of rail defects is necessary for preventing derailments and costly damage to the train and railway infrastructure. A rail surface flaw can quickly propagate from a small fracture to a broken rail after only a few train cars have passed over it. Rail defect detection is typically performed by using an instrumented car or a separate railway monitoring vehicle. Rail surface irregularities can be measured using accelerometers mounted to the bogie side frames or wheel axles. Typical signal processing algorithms for detecting defects within a vertical acceleration signal use a simple thresholding routine that considers only the amplitude of the signal. As a result, rail surface defects that produce low amplitude acceleration signatures may not be detected, and special track components that produce high amplitude acceleration signatures may be flagged as defects. The focus of this research is to develop an intelligent signal processing algorithm capable of detecting and classifying various rail surface irregularities, including defects and special track components. Three algorithms are proposed and validated using data collected from an instrumented freight car. For the first two algorithms, one uses a windowed Fourier Transform while the other uses the Wavelet Transform for feature extraction. Both of these algorithms use an artificial neural network for feature classification. The third algorithm uses the Wavelet Transform to perform a regularity analysis on the signal. The algorithms are validated with the collected data and shown to out-perform the threshold-based algorithm for the same data set. Proper training of the defect detection algorithm requires a large data set consisting of operating conditions and physical parameters. To generate this training data, a dynamic wheel-rail interaction model was developed that relates defect geometry to the side frame vertical acceleration signature. The model was generated by using combined systems dynamic modeling, and the system was solved with a developed combined lumped and distributed parameter system numerical approximation. The broken rail model was validated with real data collected from an instrumented freight car. The model was then used to train and validate the defect detection methodologies for various train and rail physical parameters and operating conditions. / Ph. D.

This thesis presents an artificial neural network (ANN) approach to diagnose and detect faults in oil-filled power transformers based on dissolved gas-in-oil analysis. The goal of the research is to investigate the available transformer incipient fault diagnosis methods and then develop an ANN approach for this purpose. This ANN classifier should not only be able to detect the fault type, but also should be able to judge the cellulosic material breakdown. This classifier should also be able to accommodate more than one type of fault. This thesis describes a two-step ANN method that is used to detect faults with or without cellulose involved. Utilizing a feedforward artificial neural network, the classifier was trained with back-propagation, using training samples collected from different failed transformers. It is shown in the thesis that such a neural-net based approach can yield a high diagnosis accuracy. Several possible design alternatives and comparisons are also addressed in the thesis. The final system has been successfully tested, exhibiting a classification accuracy of 95% for major fault type and 90% for cellulose breakdown. / Master of Science

Traditional methods of control allocation optimization have shown difficulties in exploiting the full potential of controlling a large array of control surfaces. This research investigates the potential of employing artificial intelligence methods like neurocomputing to the control allocation optimization problem of Hybrid Wing Body (HWB) aircraft concepts for minimizing control power, hinge moments, and actuator forces, while keeping the system weights within acceptable limits. The main objective is to develop a proof-of-concept process suitable to demonstrate the potential of using neurocomputing for optimizing actuation power for aircraft featuring multiple independently actuated control surfaces and wing flexibility. An aeroelastic Open Rotor Engine Integration and Optimization (OREIO) model was used to generate a database of hinge moment and actuation power characteristics for an array of control surface deflections. Artificial neural network incorporating a genetic algorithm then performs control allocation optimization for an example aircraft. The results showed that for the half-span model, the optimization results (for the sum of the required hinge moment) are improved by more than 11%, whereas for the full-span model, the same approach improved the result by nearly 14% over the best MSC Nastran solution by using the neural network optimization process. The results were improved by 23% and 27% over the case where only the elevator is used for both half-span and full-span models, respectively. The methods developed and applied here can be used for a wide variety of aircraft configurations. / Master of Science

Fixed-object crashes have long time been considered as major roadway safety concerns. While previous relevant studies tended to address such crashes in the context of roadway departures, and heavily relied on police-reported accidents data, this study integrated the SHRP2 NDS and RID data for analyses, which fully depicted the prior to, during, and after crash scenarios. A total of 1,639 crash, near-crash events, and 1,050 baseline events were acquired. Three analysis methods: logistic regression, support vector machine (SVM) and artificial neural network (ANN) were employed for two responses: crash occurrence and severity level. Logistic regression analyses identified 16 and 10 significant variables with significance levels of 0.1, relevant to driver, roadway, environment, etc. for two responses respectively. The logistic regression analyses led to a series of findings regarding the effects of explanatory variables on fixed-object event occurrence and associated severity level. SVM classifiers and ANN models were also constructed to predict these two responses. Sensitivity analyses were performed for SVM classifiers to infer the contributing effects of input variables. All three methods obtained satisfactory prediction performance, that was around 88% for fixed-object event occurrence and 75% for event severity level, which indicated the effectiveness of NDS event data on depicting crash scenarios and roadway safety analyses. / Master of Science / Fixed-object crashes happen when a single vehicle strikes a roadway feature such as a curb or a median, or runs off the road and hits a roadside feature such as a tree or utility pole. They have long time been considered as major highway safety concerns due to their high frequency, fatality rate, and associated property cost. Previous studies relevant to fixed-object crashes tended to address such crashes in the contexture of roadway departures, and heavily relied on police-reported accident data. However, many fixed-object crashes involved objects in roadway such as traffic control devices, roadway debris, etc. The police-reported accident data were found to be weak in depicting scenarios prior to, during crashes. Also, many minor crashes were often kept unreported. The Second Strategic Highway Research Program (SHRP2) Naturalistic Driving Study (NDS) is the largest NDS project launched across the country till now, aimed to study driver behavior or, performance-related safety problems under real-world scenarios. The data acquisition systems (DASs) equipped on participated vehicles collect vehicle kinematics, roadway, traffic, environment, and driver behavior data continuously, which enable researchers to address such crash scenarios closely. This study integrated SHRP2 NDS and roadway information database (RID) data to conduct a comprehensive analysis of fixed-object crashes. A total of 1,639 crash, near-crash events relevant to fixed objects and animals, and 1,050 baseline events were used. Three analysis methods: logistic regression, support vector machine (SVM) and artificial neural network (ANN) were employed for two responses: crash occurrence and severity level. The logistic regression analyses identified 16 and 10 variables with significance levels of 0.1 for fixed-object event occurrence and severity level models respectively. The influence of explanatory variables was discussed in detail. SVM classifiers and ANN models were also constructed to predict the fixed-object crash occurrence and severity level. Sensitivity analyses were performed for SVM classifiers to infer the contributing effects of input variables. All three methods achieved satisfactory prediction accuracies of around 88% for crash occurrence prediction and 75% for crash severity level prediction, which suggested the effectiveness of NDS event data on depicting crash scenarios and roadway safety analyses.

The suspension system is one of the most sensitive systems of a vehicle as it affects the dynamic behavior of the vehicle with even minor changes. These systems are designed to carry out multiple tasks such as isolating the vehicle body from the road/tire vibrations as well as achieving desired ride and handling performance levels in both steady state and limit handling conditions. The damping coefficient of the damper plays a crucial role in determining the overall frequency response of the suspension system. Considerable research has been carried out on semi active damper systems as the damping coefficient can be varied without the system requiring significant external power giving them advantages over both passive and fully active suspension systems. Dampers behave as non-linear systems at higher frequencies and hence it has been difficult to develop accurate models for its full range of motion. This study aims to develop a velocity sensitive damper model using artificial neural networks and essentially provide a 'black-box' model which encapsulates the non-linear behavior of the damper. A feed-forward neural network was developed by testing a semi active damper on a shock dynamometer at CenTiRe for multiple frequencies and damping ratios. This data was used for supervised training of the network using MATLAB Neural Network Toolbox. The developed NN model was evaluated for its prediction accuracy. Further, the developed damper model was analyzed for feasibility of use for simulations and controls by integrating it in a Simulink based quarter car model and applying the well-known skyhook control strategy. Finally, effects on ride and handling dynamics were evaluated in Carsim by replacing the default damper model with the proposed model. It was established that this damper modeling technique can be used to help evaluate the behavior of the damper on both component as well as vehicle level without needing to develop a complex physics based model. This can be especially beneficial in the earlier stages of vehicle development. / Master of Science / The suspension system is one of the most sensitive systems of a vehicle as it affects the dynamic behavior of the vehicle with even minor changes. These systems are designed to carry out multiple tasks such as absorbing shocks from the road as well as improving the handling of the vehicle for a smoother and safer drive. The level of firmness of the shock absorber/damper plays a crucial role in determining the overall behavior of the suspension system. Considerable research has been carried out on semi active damper systems as the damper stiffness can be varied quickly and easily as compared to other passive and fully active damper systems. Dampers are complex systems to model especially for high speed operations and hence it has been difficult to develop accurate mathematical models for its full range of motion. This study aims to develop an accurate mathematical model for a semi active damper using artificial neural networks. A semi active damper was fabricated and tested on a shock dynamometer at CenTiRe for multiple speeds and stiffness values. Thistest data obtained was used for training of the mathematical model using the computer software MATLAB. The developed model was evaluated for its accuracy and further analyzed for feasibility of use in computer simulations. It was established that this damper modeling technique can be used to help evaluate the behavior of the damper with high accuracy while still running the simulations relatively quickly whereas in current simulations compromise has to be made on at least the accuracy of the model or the simulation speed. This can be especially beneficial in the earlier stages of vehicle development.

Yes / Reverse Osmosis (RO) can be considered as one of the most widely used technologies used to abate the existence of highly toxic compounds from wastewater. In this paper, a multilayer artificial neural network (MLANN) with Genetic Algorithm (GA) have been considered to build a comprehensive mathematical model, which can be used to predict the performance of an individual RO process in term of chlorophenol removal from wastewater. The MLANN model has been validated against 70 observational experimental datasets collected from the open literature. The MLANN model predictions have outperformed the predictions of several structures developed for the same chlorophenol removal using RO process based on performance in terms of coefficient of correlation, coefficient determination (R2) and average error (AVE). In this respect, two structures (4-2-2-1) and (4-8-8-1) were also used to study the effect of a number of neurons in the hidden layers based on the difference between the measured and ANN predicted values. The model responses clearly confirm the successfulness of estimating the chlorophenol rejection for network structure 4-8-8-1 based on a wide range of the control variables. This also represents a high consistency between the ANN model predictions and the experimental data.

Yes / The content and concentration of beta-carotene, tocopherol and free fatty acid is one of the important parameters that affect the quality of edible oil. In simulation based studies for refined palm oil process, three variables are usually used as input parameters which are feed flow rate (F), column temperature (T) and pressure (P). These parameters influence the output concentration of beta-carotene, tocopherol and free fatty acid. In this work, we develop 2 different ANN models; the first ANN model based on 3 inputs (F, T, P) and the second model based on 2 inputs (T and P). Artificial neural network (ANN) models are set up to describe the simulation. Feed forward back propagation neural networks are designed using different architecture in MATLAB toolbox. The effects of numbers for neurons and layers are examined. The correlation coefficient for this study is greater than 0.99; it is in good agreement during training and testing the models. Moreover, it is found that ANN can model the process accurately, and is able to predict the model outputs very close to those predicted by ASPEN HYSYS simulator for refined palm oil process. Optimization of the refined palm oil process is performed using ANN based model to maximize the concentration of beta-carotene and tocopherol at residue and free fatty acid at distillate.

No / The quest for best practices may lead to an increased risk of poor decision-making, especially when aiming to attain best practice levels reveals that efforts are beyond the organization’s present capabilities. This situation is commonly known as the “best practice trap”. Motivated by such observation, the purpose of the present paper is to develop a practical methodology to support better practice benchmarking, with an application to the banking sector. In this sense, we develop a two-stage hybrid model that employs Artificial Neural Network (ANN) via integration with Data Envelopment Analysis (DEA), which is used as a preprocessor, to investigate the ability of the DEA-ANN approach to classify the sampled branches of a Greek bank into predefined efficiency classes. ANN is integrated with a family of radial and non-radial DEA models. This combined approach effectively captures the information contained in the characteristics of the sampled branches, and subsequently demonstrates a satisfactory classification ability especially for the efficient branches. Our prediction results are presented using four performance measures (hit rates): percent success rate of classifying a bank branch’s performance exactly or within one class of its actual performance, as well as just one class above the actual class and just one class below the actual class. The proposed modeling approach integrates the DEA context with ANN and advances benchmarking practices to enhance the decision-making process for efficiency improvement.

No / The purpose of this study was to determine whether artificial neural network (ANN) programs implementing different backpropagation algorithms and default settings are capable of generating equivalent highly predictive models. Three ANN packages were used: INForm, CAD/Chem and MATLAB. Twenty variants of gradient descent, conjugate gradient, quasi-Newton and Bayesian regularisation algorithms were used to train networks containing a single hidden layer of 3¿12 nodes. All INForm and CAD/Chem models trained satisfactorily for tensile strength, disintegration time and percentage dissolution at 15, 30, 45 and 60 min. Similarly, acceptable training was obtained for MATLAB models using Bayesian regularisation. Training of MATLAB models with other algorithms was erratic. This effect was attributed to a tendency for the MATLAB implementation of the algorithms to attenuate training in local minima of the error surface. Predictive models for tablet capping and friability could not be generated. The most predictive models from each ANN package varied with respect to the optimum network architecture and training algorithm. No significant differences were found in the predictive ability of these models. It is concluded that comparable models are obtainable from different ANN programs provided that both the network architecture and training algorithm are optimised. A broad strategy for optimisation of the predictive ability of an ANN model is proposed.