Filtration is an important process in drinking water treatment to ensure the adequate removal of particle-bound pathogens (i.e. Giardia and Cryptosporidium). Filtration performance is typically monitored in terms of filtered water turbidity. However, particle counts may provide further insight into treatment efficiency, as they have a greater sensitivity for detecting small changes in filtration operation. To optimize the filtration process at the Elgin Area WTP in terms of post-filtration particle counts, artificial neural network (ANN) models were applied. Process models were successfully developed to predict settled water turbidity and particle counts. Additionally, two inverse process models were developed to predict the optimal coagulant dosage required to attain target particle counts. Upon testing each model, a high correlation was observed between the actual and predicted data sets. The ANNs were then integrated into an optimization application to allow for the transfer of real-time data between the models and the SCADA system.

The formation of disinfection by-products (DBPs) in drinking water has become an issue of greater concern in recent years. Bench-scale jar tests were conducted on a surface water to evaluate the impact of enhanced coagulation on the removal of organic DBP precursors and the formation of trihalomethanes (THMs) and haloacetic acids (HAAs). The results of this testing indicate that enhanced coagulation practices can improve treated water quality without increasing coagulant dosage. The data generated were also used to develop artificial neural networks (ANNs) to predict THM and HAA formation. Testing of these models showed high correlations between the actual and predicted data. In addition, an experimental plan was developed to use ANNs for treatment optimization at the Peterborough pilot plant.

Influenza is a public health problem that causes serious illness and deaths all over the world. Vaccination has been shown to be the most effective mean to prevent infection. The primary component of influenza vaccine is the weakened strains. Vaccination triggers the immune system to develop antibodies against those strains whose viral surface glycoprotein hemagglutinin (HA) is similar to that of vaccine strains. However, influenza vaccine must be updated annually since the antigenic structure of HA is constantly mutation. Hemagglutination inhibition (HI) assay is a laboratory procedure frequently applied to evaluate the antigenic relationships of the influenza viruses. It enables the World Health Organization (WHO) to recommend appropriate updates on strains that will most likely be protective against the circulating influenza strains. However, HI assay is labour intensive and time-consuming since it requires several controls for standardization. We use two machine-learning methods, i.e. Artificial Neural Network (ANN) and Logistic Regression, and a Mixed-Integer Optimization Model to predict antigenic variety. The ANN generalizes the input data to patterns inherent in the data, and then uses these patterns to make predictions. The logistic regression model identifies and selects the amino acid positions, which contribute most significantly to antigenic difference. The output of the logistic regression model will be used to predict the antigenic variants based on the predicted probability. The Mixed-Integer Optimization Model is formulated to find hyperplanes that enable binary classification. The performances of our models are evaluated by cross validation.

Current activated sludge models are deterministic in character and are constructed by basing on the fundamental biokinetics. However, calibrating these models are extremely time consuming and laborious. An easy-to-calibrate and user friendly computer model, one of the artificial intelligence techniques, Artificial Neural Networks (ANNs) were used in this study. These models can be used not only directly as a substitute for deterministic models but also can be plugged into the system as error predictors. Three systems were modeled by using ANN models. Initially, a hypothetical wastewater treatment plant constructed in Simulation of Single-Sludge Processes for Carbon Oxidation, Nitrification &amp / Denitrification (SSSP) program, which is an implementation of Activated Sludge Model No 1 (ASM1), was used as the source of input and output data. The other systems were actual treatment plants, Ankara Central Wastewater Treatment Plant, ACWTP and iskenderun Wastewater Treatment Plant (IskWTP). A sensitivity analysis was applied for the hypothetical plant for both of the model simulation results obtained by the SSSP program and the developed ANN model. Sensitivity tests carried out by comparing the responses of the two models indicated parallel sensitivities. In hypothetical WWTP modeling, the highest correlation coefficient obtained with ANN model versus SSSP was about 0.980. By using actual data from IskWTP the best fit obtained by the ANN model yielded R value of 0.795 can be considered very high with such a noisy data. Similarly, ACWTP the R value obtained was 0.688, where accuracy of fit is debatable.

In this study, an alternative seismic vulnerability assessment model is developed. For this purpose, one of the most popular artificial intelligence techniques, Artificial Neural Network (ANN), is used. Many ANN models are generated using 4 different network training functions, 1 to 50 hidden neurons and combination of structural parameters like number of stories, normalized redundancy scores, overhang ratios, soft story indices, normalized total column areas, normalized total wall areas are used to achieve the best assessment performance. Duzce database is used throughout the thesis for training ANN. A neural network simulator is developed in Microsoft Excel using the weights and parameters obtained from the best model created at Duzce damage database studies. Afyon, Erzincan, and Ceyhan databases are simulated using the developed simulator. A recently created database named Zeytinburnu is used for the projection purposes. The building sesimic vulnerability assessment of Zeytinburnu area is conducted on 3043 buildings using the proposed procedure.

In the first part of this study, two entropy methods under different distribution assumptions are examined on a network of stream gauging stations located in Kizilirmak Basin to rank the stations according to their level of importance. The stations are ranked by using two different entropy methods under different distributions. Thus, showing the effect of the distribution type on both entropy methods is aimed. In the second part of this study, autoregressive models with asymmetric innovations and an artificial neural network model are introduced. Autoregressive models (AR) which have been developed in hydrology are based on several assumptions. The normality assumption for the innovations of AR models is investigated in this study. The main reason of making this assumption in the autoregressive models established is the difficulties faced in finding the model parameters under the distributions other than the normal distributions. From this point of view, introduction of the modified maximum likelihood procedure developed by Tiku et. al. (1996) in estimation of the autoregressive model parameters having non-normally distributed residual series, in the area of hydrology has been aimed. It is also important to consider how the autoregressive model parameters having skewed distributions could be estimated. Besides these autoregressive models, the artificial neural network (ANN) model was also constructed for annual and monthly hydrologic time series due to its advantages such as no statistical distribution and no linearity assumptions. The models considered are applied to annual and monthly streamflow data obtained from five streamflow gauging stations in Kizilirmak Basin. It is shown that AR(1) model with Weibull innovations provides best solutions for annual series and AR(1) model with generalized logistic innovations provides best solution for monthly as compared with the results of artificial neural network models.

In recent years, artificial neural networks (ANNs) have been applied to estimate in many areas of hydrology and hydraulic engineering. In this thesis, multilayered feedforward backpropagation algorithm was used to establish for the prediction of unit discharge q (m3/s/m) in a rectangular free overfall. Researchers&rsquo / experimental data were used to train and validate the network with high reliability. First, an appropriate ANN model has been established by considering determination of hidden layer and node numbers related to training function and training epoch number. Then by applying sensitivity analysis, parameters involved in and their effectiveness relatively has been determined in the phenomenon. In the scope of the thesis, there are two case studies. In the first case study, ANN models reliability has been investigated according to the training data clustered and the results are given by comparing to regression analysis. In the second case, ANN models&rsquo / ability in establishing relations with different data clusters is investigated and effectiveness of ANN is scrutinized.

Damage to steel moment frames in the 1994 Northridge and 1995 Hyogken-Nanbu earthquakes subsequently motivated intensive research and testing efforts in the US, Japan, and elsewhere on moment frames. Despite extensive past research efforts, one important problem remains unresolved: the degree of panel zone participation that should be permitted in the inelastic seismic response of a steel moment frame. To date, a fundamental computational model has yet to be developed to assess the cyclic rupture performance of moment frames. Without such a model, the aforementioned problem can never be resolved. This dissertation develops an innovative way of predicting cyclic rupture in steel moment frames by employing artificial neural networks. First, finite element analyses of 30 notched round bar models are conducted, and the analytical results in the vicinity of the notch root are extracted to form the inputs for either a single neural network or a competitive neural array. After training the neural networks, the element with the highest potential to initiate a fatigue crack is identified, and the time elapsed up to the crack initiation is predicted and compared with its true synthetic answer. Following similar procedures, a competitive neural array comprising dynamic neural networks is established. Two types of steel-like materials are created so that material identification information can be added to the input vectors for neural networks. The time elapsed by the end of every stage in the fracture progression is evaluated based on the synthetic allocation of the total initiation life assigned to each model. Then, experimental results of eight beam-to-column moment joint specimens tested by four different programs are collected. The history of local field variables in the vicinity of the beam flange - column flange weld is extracted from hierarchical finite element models. Using the dynamic competitive neural array that has been established and trained, the time elapsed to initiate a low cycle fatigue crack is predicted and compared with lab observations. Finally, finite element analyses of newly designed specimens are performed, the strength of their panel zone is identified, and the fatigue performance of the specimens with a weak panel zone is predicted.

The ability to forecast machinery failure is vital to reducing maintenance costs, operation downtime and safety hazards. Recent advances in condition monitoring technologies have given rise to a number of prognostic models for forecasting machinery health based on condition data. Although these models have aided the advancement of the discipline, they have made only a limited contribution to developing an effective machinery health prognostic system. The literature review indicates that there is not yet a prognostic model that directly models and fully utilises suspended condition histories (which are very common in practice since organisations rarely allow their assets to run to failure); that effectively integrates population characteristics into prognostics for longer-range prediction in a probabilistic sense; which deduces the non-linear relationship between measured condition data and actual asset health; and which involves minimal assumptions and requirements. This work presents a novel approach to addressing the above-mentioned challenges. The proposed model consists of a feed-forward neural network, the training targets of which are asset survival probabilities estimated using a variation of the Kaplan-Meier estimator and a degradation-based failure probability density estimator. The adapted Kaplan-Meier estimator is able to model the actual survival status of individual failed units and estimate the survival probability of individual suspended units. The degradation-based failure probability density estimator, on the other hand, extracts population characteristics and computes conditional reliability from available condition histories instead of from reliability data. The estimated survival probability and the relevant condition histories are respectively presented as “training target” and “training input” to the neural network. The trained network is capable of estimating the future survival curve of a unit when a series of condition indices are inputted. Although the concept proposed may be applied to the prognosis of various machine components, rolling element bearings were chosen as the research object because rolling element bearing failure is one of the foremost causes of machinery breakdowns. Computer simulated and industry case study data were used to compare the prognostic performance of the proposed model and four control models, namely: two feed-forward neural networks with the same training function and structure as the proposed model, but neglected suspended histories; a time series prediction recurrent neural network; and a traditional Weibull distribution model. The results support the assertion that the proposed model performs better than the other four models and that it produces adaptive prediction outputs with useful representation of survival probabilities. This work presents a compelling concept for non-parametric data-driven prognosis, and for utilising available asset condition information more fully and accurately. It demonstrates that machinery health can indeed be forecasted. The proposed prognostic technique, together with ongoing advances in sensors and data-fusion techniques, and increasingly comprehensive databases of asset condition data, holds the promise for increased asset availability, maintenance cost effectiveness, operational safety and – ultimately – organisation competitiveness.

<p>n this dissertation a comparison of two evolutionary methods for evolving ANNs for robot control is made. The methods compared are SANE with enforced sub-population and delta-coding, and marker-based encoding. In an attempt to speed up evolution, marker-based encoding is extended with delta-coding. The task selected for comparison is the hunter-prey task. This task requires the robot controller to posess some form of memory as the prey can move out of sensor range. Incremental evolution is used to evolve the complex behaviour that is required to successfully handle this task. The comparison is based on computational power needed for evolution, and complexity, robustness, and generalisation of the resulting ANNs. The results show that marker-based encoding is the most efficient method tested and does not need delta-coding to increase the speed of evolution process. Additionally the results indicate that delta-coding does not increase the speed of evolution with marker-based encoding.</p>