M.S. Theses

Permanent URI for this collection

https://digitalarchive.library.bogazici.edu.tr/handle/123456789/11658

Browse

Now showing 1 - 11 of 11

Application of robust statistics on a crude distillation unit
(Thesis (M.S.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2017., 2017.) Nalbant Kurşun, Sinem.; Alakent, Burak.
Refineries are highly complex and integrated systems, separating and transforming crude oil into valuable products. One of the most important processes in refineries is the Crude Distillation Unit (CDU) process, in which raw crude oil is separated into various fractions to be further processed in other parts of the refinery. In the refinery, Heavy Diesel (HD) T95 value is very important quality indicator. In the current study, conventional and robust statistical methods were employed on the historical data of a CDU process in TUPRAS İzmit Refinery for monitoring and HD T95 prediction purposes. Process data consisted of online measurements of process variables and laboratory measurements of HD T95 values for a one-year period. In the first part of the study, trajectories of process variables were analyzed to identify relations between process variables and to distinguish normal from abnormal operating conditions in the distillation history. For this purpose, skipped- Principal Components Analysis (PCA) and Minimum Covariance Determinant (MCD)+PCA methods were applied to process data and MCD+PCA method was found as more efficient method in detecting disturbances in the operation conditions. In the second part of the study, Monte Carlo (MC) simulations were applied by creating clean and contaminated datasets to evaluate predictive performances of LS and various robust regression methods, and to assess the metrics (RMSE, MAE) for evaluating the quality of predictions under contamination. LTS10%+LS and LTS20%+LS were found as best predictive models, and RMSE was found to be reliable in assessing models when 70%- 90% of the highest absolute prediction errors were taken into account. In the last section, LS and robust regression methods were applied and compared to select the most convenient prediction method for HD T95 values. The best predictive performance was obtained by LTS30% model with 97.5% CL. By applying this method to historical dataset, 15% of training dataset was detected as outliers and when these outliers were excluded from dataset, the model can predict HD T95 value with a maximum 7 0C error.
Data reduction methods in just-in-time-learning
(Thesis (M.S.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2021., 2021.) Boy, Onur Can.; Alakent, Burak.
Industrial processes are monitored continuously to meet the standards dictated by the market and environmental regulations. There are many process variables like temperature, pressure, flow rate which can be easily measured online through mechanical sensors and lots of data can be stored thanks to the advancements in data storage technology. At the same time, there are variables that show the quality of a product, process safety or some other restricted chemical composition. These are called quality variables and they are measured less often due to requirement of a detailed laboratory analysis. Measuring quality variable ones in a shift is a weak link in process control and monitoring. Implementing a soft sensor is a very efficient way to predict quality variables through statistical learning methods. Traditional soft sensors are built in an offline manner and used for online prediction while it requires maintenance periodically as process shifts to another state. Just-in-time learning is an adaptive method in which a local model is built when a new sample is obtained, and the model is discarded after a prediction is made. JITL outperforms traditional methods in terms of efficiency and predictive ability. The prediction performance of a soft sensor is also affected by the quality of the training data stored in the data base. Data reduction methods are used to eliminate data that weaken prediction quality and to store meaningful data for increasing prediction performance and model efficiency. In this thesis, JITL models are trained with Lasso and least squares support vector regression and three different data reduction algorithms using four different data sets. It is shown that the effect of each data reduction method changes from data set to data set, and prediction accuracy of JITL using all data can be attained using a smaller training sets. Additionally, results show that prediction accuracy of nonlinear models trained by LSSVR outperforms that of Lasso.
Descriptive and predictive statistical modeling of a ring spinning process
(Thesis (M.S.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2016., 2016.) İşsever, Reyhan.; Alakent, Burak.
In the textile industry, spinning process is one of the most important stages to determine the yarn structure, on which the desired fabric properties highly depend, throughout the manufacturing chain. In the production of wool yarn, \ring spinning" is the most frequently used technology since wool is principally ring-spun. In the current study, it is aimed to determine how probability of end breakage, which is one of the most important quality variables, changes with respect to process variables using statistical methods. Process data are collected under normal operating conditions of YUNSA Worsted and Woolen Company in Turkey between 2012-2014. Nominal process variables consist of color and composition of the fed ber, ring spinning machine number, spinning, and twist direction, while continuous process variables are lot size, roving count, draft, twist level, ring traveler number, yarn count, spindle speed, and machine age. In the rst part of the study, Principal Component Analysis (PCA) is used to examine how historical operating conditions described by continuous process variables and binary nominal variables change for di erent runs, while Correspondence Analysis (CA) is employed to elucidate which machines are preferred for certain operating conditions. In the second part, failure probability is modeled with respect to process variables using logistic regression. The predictive powers of the regression models constructed for the rst, second and third types of machines, area under its ROC was found to be 0.66, 0.70 and 0.75 with optimal true positive and false positive rates as 0.64 and 0.40, 0.67 and 0.40, and 0.65 and 0.28, respectively.
Identification of PTP1B dynamics using frequency response analysis
(Thesis (M.S.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2012., 2012.) Eren, Dilek.; Alakent, Burak.
Protein tyrosine phosphatase 1B (PTP1B) plays a key role as negative regulator of insulin and leptin signaling. In human PTP1B, a significant conformation change is seen in the WPD loop from open (inactive) to closed (active) conformation, when a ligand is bound to the enzyme. The aim of this study is to see how functionally important distant portions, such as S-loop, are perturbed by disturbances in the active site, and suggest a method which could be utilized to find communication pathways between distant regions in a protein, using frequency response techniques. Therefore, five different periods of TMD simulation cycles (one closing-opening motion of WPD loop) were applied between WPD loop in open (WPDopen) and in closed (WPDclosed) conformations of PTP1B. TMD potential force is applied on the WPD loop and the R-loop Using Discrete Fourier Transform to filter out the perturbed frequency components from random fluctuations, it was observed that the effects of this continuous periodic excitation of the active site spread throughout the whole protein, manifesting itself with increase in mobility of distant regions. Increase in fluctuations are seen not only in the vicinity of WPD loop and R- loop, but also at some distant parts of PTP1B, not directly in contact with the active site or WPD loop, i.e. α1', α2', pTyr recognition loop, R-loop, L11, WPD loop, α3, P-loop, Q-loop and α6. Moreover, most of the displacements of the residues are practically in the same direction with those of the crystal structures. A surprising result is also obtained in the functionally important region, S-loop. Although there is no mobility increase in S-loop with respect to equilibrium MD simulations, opening/closing of WPD loop makes the first eigenvector of S-loop displacements at the base frequency of oscillations more aligned with that between the crystal structures. This result shows that allostericity is not always manifested as increase in residue mobility, but also in adjustment of the directionality of residue fluctuations. TMD simulations at higher frequencies of WPD loop motions show that amplitude of oscillations of many of the distant regions, and even the active site, decreases, as that would be expected from a linear system.
Identification of the relations between synthesis conditions and compressive strength in geopolymer systems via statistical learning methods
(Thesis (M.S.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2019., 2019.) Köse, Kadriye Deniz.; Alakent, Burak.; Soyer-Uzun, Sezer.
Geopolymers are inorganic amorphous materials, which can be used in construction industry, and their production is characterized by low CO2 emission. Compressive strength (CS) is a durability measure, which, in the case of geopolymers, depends on various manufacturing factors in a rather complex functional relation. The aim of this study is to develop a decision tree and extract reliable decision rules based on it. A database consisting of 879 data points and 19 predictors is constructed based on available publications. Forward Selection (FS), Backward Elimination (BE) and Rough Set theory are used as feature selection algorithms. The decision tree parameters (maximum number of splits, Smax and minimum parent size Pmin) and number of relevant predictors for FS and BE and threshold parameter (eps), which is introduced for reduct set formation, are optimized via 5-fold CV with 20 repetitions. The reliability of the obtained decision rules is expressed via Stability index (SI) whereas the accuracy of the models via measures, such as RMSE. Minimum test RMSE is decreased up to 10.0 MPa for Smax = 150 BE based tree, but the interpretation of such large trees is demanding. Hence, a smaller sized (Smax = 60) reduct based tree is used for rule extraction. The CV RMSE and SI, test RMSE and R2 of the resulting tree is 13.2 MPa, 0.79, 13.2 MPa and 0.64 respectively. Based on this tree, the most important predictor affecting geopolymer CS is found to be the type of aluminosilicate precursor material. Two material groups are identified corresponding to higher and lower CS. The data space is subsequently divided, and relations (linear regression, local models etc.) regarding smaller subspaces are developed. For higher CS group, water/Al, Na/Al and ambient curing time have a significant effect, whereas for lower CS compounds, purity and heat curing time are also found to be effective. Moreover, the effect of the predictors on CS can vary based on the precursor types, showing the presence of significant predictor interactions.
Investigation of PTP1B activation mechanism computational methods
(Thesis (M.S.)-Bogazici University. Institute for Graduate Studies in Science and Engineering, 2010., 2010.) Özkaral, Burcu.; Özkırımlı, Elif.; Alakent, Burak.
Protein tyrosine phosphatase 1B (PTP1B) is a negative regulator of insulin and leptin signaling, and is a major molecular target for the treatment of type II diabetes and obesity. WPD loop is a key element in the mechanism of PTP1B catalysis. In the apo form of the enzyme, WPD loop is usually in an “open” conformation, whereas it closes over the active site upon substrate binding. Here, targeted molecular dynamics (TMD) simulations are reported to examine the transition of the WPD loop from the open to closed states as well as the effect of this motion on the PTP1B conformational activation mechanism. Targeting potential was applied to the WPD loop only and to the whole protein in different sets of simulations. Residue-residue interactions and dihedral angles that contribute to the WPD loop conformational transition were identified. Two major conformational transitions between the open and closed states of WPD loop were observed using PCA and K-means clustering analysis of the Cα atoms in all TMD simulations, except TMD4 simulation. The first transition was the backbone dihedral angle rotation of Ser187 or Pro188 at the WPD loop C-terminus. The second transition was the simultaneous rotations of Trp179 and Arg221 sidechain dihedral angles, and the formation of a polar interaction between the WPD loop and the Arg221 sidechain. A third subtle conformational change, which was not observed in the clustering analysis, was the rotation of the backbone dihedral angle between Asp181 and Phe182 backbone dihedral angles, resulting in the formation of hydrogen bonds between Asp181-Gly183 and between Phe182-Gln262 at. It was also was found that WPD loop closing motion was hindered due to the alternative conformations of the Arg221 and the closed conformation of the R loop. Hydrophobic interactions between the WPD loop and the regions around it, such as the active site, helices α3, α6, α7, loop between β9-β10 (loop11) and S loop were also investigated and it was found that these regions may act as steric hinderance that could influence the WPD loop closure during the simulations. Elucidating the detailed mechanism of PTP1B conformational activation will guide future drug design efforts toward type II diabetes and obesity.
Modeling of polymer / ceramic composites via molecular dynamics simulations
(Thesis (M.S.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2019., 2019.) Göreke, Melike Dilara.; Soyer-Uzun, Sezer.; Alakent, Burak.
Polymer-calcium phosphate (CaP) composites arise as alternatives to biological bone substitutes and are widely used as biomaterials. Among various CaP ceramics, hydroxyapatite (HAp), β-tricalcium phosphate (β-TCP) and their mixtures known as biphasic calcium phosphates (BCP) are the most important bioceramics due to their superior features as bioactivity, biocompatibility, and stability in physiological environ ment. In the current thesis, polymer-CaP ceramics, including HAp, β-TCP, and BCP, are comparatively studied via MD simulations. In the ﬁrst part, binding mechanism of (poly)lactic acid (PLA)-HAp and (poly)ethylene (PE)-HAp systems is examined using MD simulations with diﬀerent number of monomers (10 ≤ N ≤ 400) on HAp surfaces at two diﬀerent thicknesses. HAp models with thicker bulk region consistently yielded positive global binding energy values. Change in binding energy and the occupied area by polymer (occA) show exponential recovery relationships as a function of N. Binding energy values in PLA-HAp systems converge to higher values compared to PE-HAp complexes while occA values stabilize at lower values in PLA-HAp complexes. Bulk re gion of HAp is found to be a major constituent of the total binding energy, followed by polymer-surface interactions for both systems. Concentration proﬁles revealed that O= units are mainly responsible for the PLA-HAp interaction, intensifying until N ≈ 200, in agreement with surface-polymer interaction and Ca-O coordination number proﬁles. In PE-HAp systems, interface concentration constantly increases with respect to N, parallel to surface-polymer interaction proﬁles. In the second part of the thesis, interac tion of PLA with biphasic calcium phosphate and its building blocks, HAp and β-TCP is studied in an introductory manner. The initial results are promising, yielding bind ing energy ranked as β-TCP>BCP>HAp, complemented by concentration proﬁles, in which O= units are again found to be responsible for the interfacial adhesion.
Optimization of a commercial hydro-isomerization process
(Thesis (M.S.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2015., 2015.) Küçük, Ülfet Ece.; Avcı, Ahmet Kerim.; Alakent, Burak.
Isomerization process is one of the most commonly used processes in the petroleum refinery. Product of isomerization unit is “isomerate” that has a great importance for gasoline pool, since its research octane number (RON) is easy to be adjusted and it obeys the legislations about sulfur and benzene content of the gasoline pool. UOP PenexTM process is mostly used application for isomerization. PenexTM is created for the catalytic isomerization of low octane light straight run (LSR) naphtha to obtain high octane and branched isomers. Benzene reduction reactions also occur in the unit, product can be obtained with high octane number and low benzene content. In this thesis, the PenexTM unit at Tüpraş Izmir Refinery are operated at steady state, and optimized for given conditions. Optimization of the process is performed in HYSYS Optimizer Tool. Decision variables are determined after applying degree of freedom analysis to whole flow-sheet and constraints are determined by taking into account of top product rate and quality. Optimization is carried out in mixed mode, which is combination of both SQP and BOX methods, with nine decision variables and two constraints. Objective function of the process is chosen as maximum isomerate barrels production (maximum liquid yield). During the current study, catalyst of the PenexTM unit has been changed and difference between previous and current situation is investigated. The results show that, current state of the unit with changing 0.5-5 °C range of temperature for all unit elements, changing 180 m3/d of deisohexanizer column recycle rate and changing 35 ton/h of stabilizer overhead vapor rate gives optimized process with 85.5 RON and 4673 isomerate barrel value, for first condition. With the change of trickle bed catalyst, value of change in deisohexanizer recycle rate decreases 100 m3/d and change in stabilizer overhead vapor rate decreases 15 ton/h. Change in temperatures range remains constant for second condition. Also, in second condition, 87 RON and 5232 isomerate barrel value is obtained for optimized process.
Prediction of heavy diesel T95 using just in time learning models
(Thesis (M.S.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2017., 2017.) Hasdemir, Sevi Zeynep.; Alakent, Burak.
One of the middle distillates of atmospheric distillation column is Heavy Diesel (HAD). T95 is the temperature, at which 95% volume of a sample is boiled, and it is the main controlled variable so accurate T95 predictions are required for a satisfactory performance of the model predictive control (MPC) algorithm, in which online T95 predictions are used to determine control actions. In the current thesis, just in time learning (JITL) methodology is used on historical process data to develop soft sensors for real time predictions of HAD T95. Local models are constructed using samples located in the neighborhood of a query point, and the constructed model is used for prediction of the response variable. Using 47 process variables in the historical data, three main groups of predictive models are constructed for HAD T95. In the first group of models, various subsets of variables, which are assumed to carry the highest information on variation of T95, are included into static and dynamic models. While there is no time lag between the selected input variables and the predicted quality variable in static models, previous day’s T95 values (response variable) are included in dynamic models, as in autoregressive exogenous (ARX) input modeling. In the second group of models, least-squares (LS), partial LS (PLS), and subset regression via stepwise regression methods are employed on a predictor set, which consists of seven “most important” process variables. JITL models are evaluated with respect to various reference data selection methods, reference set size, window size and neighborhood size. The best model of this group is found to have predictive root means square error (RMSE) and mean absolute error (MAE) statistics equal to 5.66 and 4.23 oC, respectively. In the last group of models, interaction and quadratic predictor terms are included in the JITL model, and neighboring samples are selected a different subset of predictors. Using this method, RMSE and MAE of prediction statistics are decreased to 4.77 and 3.82 oC, respectively. This, to our knowledge, is the first time predictor and neighbor selection predictor subsets are separated from each other in the literature, and this seems a promising method in constructing soft-sensors for industrial applications.
Reweighted robust dispersion estimation methods for univariate s-charts
(Thesis (M.S.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2017., 2017.) Mutlu, Ece Çiğdem.; Alakent, Burak.
Maintaining the quality of manufactured products at a desired level depends on the stability of process dispersion and location parameters and detection of perturbations in these parameters as promptly as possible. In the application of S-Charts, which are one of the most widely used techniques to monitor process variability in statistical process monitoring, sample standard deviation and sample mean are known to be the most e cient traditional estimators in determining process parameters, based on the assumption of independent and normally distributed datasets. In the cases of estimated process parameters from Phase I data clouded with outliers, e ciency of traditional estimators is signi cantly reduced, and performance of S-Charts are undesirably low. The aim of this thesis is to propose various robust estimators and reweighting procedures to increase the performance of S-Charts in Phase II monitoring. Three dispersion estimators: sample standard deviation, median absolute deviation and scale M-estimator, and three location estimators: sample mean, Harrell-Davis qth quantile estimator and location M-estimator, are employed to directly construct the Phase II control limits of S-Charts, and also reweighted via di erent methods. Phase I e ciency of the proposed estimators and Phase II performance of S-Charts constructed from these estimators are assessed both under normality and against di use-localized and symmetric-asymmetric contaminations at di erent contamination density and magnitudes using 50,000-100,000 Monte Carlo simulations. As a result, scale M-estimator combined with Harrell-Davis 0:5th quantile estimators yield parameter estimates with the highest e ciency, and reweighting at skipping level 2-4% using a common location estimate in individuals charts to screen outlier subgroups, and individual observations are found to improve the Phase II performance of the S-Charts.
Soft sensor design in chemical processes using statistical learning methods
(Thesis (M.S.) - Bogazici University. Institute for Graduate Studies in Science and Engineering, 2018., 2018.) Urhan, Aysun.; Alakent, Burak.
Today, production facilities employ rigorous approach to process control, mon itoring and fault detection strategies, due to increased competition in the industry, and more severe manufacturing restrictions and safety concerns. A large number of process variables are measured online during operation, while quality variables, which may not be available online at the same rate as process variables need to be predicted using soft sensors. The traditional approach to data driven soft sensor design is based on statistical learning methods, such as, global ARX modeling, PLS and PCR. In this thesis, it is aimed to address the issues of multicollinearity and redundancy in process data, and concept drift in data driven soft sensor design. Experiments are performed on two synthetic datasets obtained from steady state and dynamic simulations, and one dataset comprised of dynamic industrial data. Incorporating feature selection and ensemble modeling into PLS and ANN models is shown to handle multicollinearity and redundancy in process data, and stabilize learners. RVM, an embedded feature selection method, yields superior prediction performance than PLS, under both virtual and real concept drift. RVM is also observed to handle redundancy in predictor space more eﬀectively. Several RVM-based adaptive learning algorithms are developed to cope with concept drift. Adaptive window sizing in moving window models is shown to improve predictions, and the best overall performance is achieved in window size adjustment via explicit concept drift detection. Combining MW and JITL models in ensemble learning is suggested to further increase prediction accuracy, and it is ob served to yield the best predictive performance among all algorithms. The suggested adaptive learners are shown to outperform conventional methods from the literature on both synthetic and real data, while complying with time limits of online prediction.

Browse

Browsing M.S. Theses by Author "Alakent, Burak."

Results Per Page

Sort Options