Aydin K. Sunol

Chemical plant fault diagnosis through a hybrid symbolic-connectionist machine learning approach

A novel hybrid symbolic-connectionist approach to machine learning is introduced and applied to fault diagnosis of a hydrocarbon chlorination plant. The learning algorithm addresses the knowledge acquisition problem by developing and maintaining the knowledge base through instance based inductive learning. The performance of the learning system is discussed in terms of the knowledge extracted from example cases and its classification accuracy on the test cases. Results indicate that the introduced system is a promising alternative to neural networks for fault diagnosis and a complement to expert systems.

A multi-period mixed integer linear programming model for water and energy supply planning in Kuwait.

The demand for water often necessitates desalination, particularly in arid coastal environments. Desalination is often integrated with electrical cogeneration. The demands for water and electricity change over time and are subject to uncertainty. A country-wide large-scale energy and water cogeneration planning model for Kuwait is formulated as a multi-period mixed integer linear programming problem and solved to minimize the net present value over the time period of 2013–2050. Five different plant technology options were considered for desalination and cogeneration including Oil & Multi Stage Flash, Natural Gas & Multi-Effect Distillation, Natural Gas & Reverse Osmosis, Solar Energy & Multi-Effect Distillation, and Solar Energy & Reverse Osmosis. Both water and energy usage in Kuwait and data from existing plants were utilized in providing the parameters and forecasts necessary for solution of the mathematical programming model. The model provides technology choice and associated capacity decisions for existing plants, new plants at green sites, and existing plant capacity expansions as well as their timing to meet the demands.

Development of titanium-dioxide-based aerogel catalyst with tunable nanoporosity and photocatalytic activity.

Nanostructured highly porous TiO(2)/WO(3)/Fe(3+) aerogel composite photocatalysts are prepared, characterized and tested for model photocatalytic reactions. The catalyst structure is tailored to capture environmental pollutants and enable their decomposition in situ under both ultraviolet (UV) and visible light through oxidation to smaller benign molecules. A novel and green method is utilized to prepare the unique surfactant-templated aerogel composite photocatalyst that has a highly accessible porous nanostructure with high surface area and tailored pore size distribution. The sol-gel process is combined with supercritical extraction and drying. Supercritical drying with heat treatment results in titanium dioxide with anatase crystal form. Templates used further enable retention and tuning of the nanopore structure and surface properties. The synthesized catalysts were characterized using SEM, FIB, XRD and porosimetry prior to post-evaluation in model reactions. The bandgap of the catalyst particles was also determined using diffuse reflectance. The resulting aerogel TiO(2)/WO(3)/Fe(3+) has similar photocatalytic capability compared to highly optimized commercial Degussa P25 under UV exposure and offers much superior photocatalytic capability under visible light exposure. The model reaction utilized employed methylene blue (MB) photooxidation under visible and UV light.

Automatic generation of global phase equilibrium diagrams for binary systems from equations of state

Abstract An automated and reliable procedure for systematic generation of global phase equilibrium diagrams (GPED) for binary systems is proposed. The approach utilizes equation of state, incorporates solid phase and is successful in generation of type VI phase diagram. The procedure enables automatic generation of GPED which incorporates calculations of all important landmarks such as critical endpoints (CEP), quadruple point (QP, if any), critical azeotropic points (CAP), azeotropic endpoints (AEP), pure azeotropic points (PAP), critical line, liquid–liquid–vapor line (L1L2V, if any), solid–liquid–liquid line (SL1L2, if any), solid–liquid–vapor line (SLV) and azeotropic line. The proposed strategy is completely general in that it does not require any knowledge about the type of phase diagram and can be applied to any pressure explicit equation of state model. The overall methodology include predictor-corrector method for line tracing, homotopy continuation for critical point calculations and phase stability test, automatic differentiation to calculate derivatives and trust region dogleg approach to compute thermodynamic landmark such as critical endpoints and quadruple points.

Homotopy continuation based prediction of azeotropy in binary and multicomponent mixtures through equations of state

A method to compute all the azeotropes in homogeneous binary and multicomponent mixtures for entire two-phase pressure range is described. The method is based on solving the necessary condition of azeotropy for all the possible solutions through homotopy continuation approach. The method is mathematically guaranteed to predict all the possible azeotropes and is in close agreement with experimental data for any equation of state that can adequately represent the phase behavior. Both vapor and liquid phase non-idealities are incorporated using fugacity coefficients from the Peng–Robinson–Stryjek–Vera equation of state with the Wong–Sandler mixing rules. The method is also capable of predicting the value of bifurcation pressure where homogeneous azeotropes will appear or disappear. Polyazeotropy in binary mixtures is a singular case of vapor-liquid equilibrium and can also be computed by solving the necessary condition of azeotropy in binary mixtures for multiple solutions through homotopy continuation based approach. The approach can also systematically search the entire two-phase pressure range for the appearance, disappearance and persistence of azeotropy and polyazeotropy in binary and multicomponent mixtures.

Stress formation during heating in supercritical drying

Silica aerogel preparation is analyzed and two different supercritical drying techniques, autoclave drying of the solvent and drying following carbon dioxide exchange of the solvent, are compared in terms of stress formation during the heating step. The model used is a partial differential equation that relates stress to thermal expansion and flow of pore liquid for a radially bounded cylindrical geometry and the model is modified to see the effect of condensation reactions that result in the contraction of the body. The finite element collocation method is used to solve the model for different drying techniques, at several heating rates, and shear moduli. The carbon dioxide exchange method is found to be more advantageous as far as the stress formation in the heating step of supercritical drying is concerned.

SQFDiag: semi-quantitative model-based fault monitoring and diagnosis via episodic fuzzy rules

A method for chemical process fault diagnosis using semi-quantitative model generated behavior envelopes is described. The method generates a sequence of rules for each fault class, with any rule in a sequence valid within the bounds of its time interval. This can be viewed as a qualitative description of the trend of numerical sensor measurements. For each variable in each fault class two sequences of episodic fuzzy rules are automatically generated, one for the lower and one for the upper numerical behavior envelope. The diagnostic system monitors a process via the measured sensors. The measurements are matched against the fuzzy rules for the current time in the rule base. In the case of an overlapping region defined by behavior envelopes, the distance introduced and time based fault belief scaling allows ranking of fault candidates. A novel abnormal situation will not pass the introduced system undetected due to a novel class detection mechanism. In two case studies, the system detected the correct fault even in cases of nearly total overlapped fault regions bounded by behavior envelopes.

Chemical Process Fault Diagnosis Using Kernel Retrofitted Fuzzy Genetic Algorithm Based Learner (FGAL) with a Hidden Markov Model

A hybrid generative-discriminative diagnostic system based on a symbolic learner (FGAL) retrofitted with Gaussian kernel densities for generating instantaneous class probabilities which are further used by a hidden Markov model to estimate the most likely state (fault) given the past evidence is introduced for real time process fault diagnosis. The system allows symbolic knowledge extraction, it is modular and robust. The diagnostic performance of the developed system is shown on a nonisothermal cascade controlled continuously stirred tank reactor (CSTR).

A hierarchical approach to simultaneous design of products and processes

Abstract Product and process design involve algorithmic and heuristic processing of symbolic and numeric data. Therefore, for such a design task, a hybrid approach that interweaves numerical and heuristic paradigms is warranted. The increasing rigor in modeling along with the necessary knowledge feedback results in a generalized system architecture that forms the basis of this paper. The approach is implemented using KAPPA on a Sun SPARC 5 station. The superstructure developed using the heuristic method is optimized with respect to the choice of technology, operating conditions, the technology sequencing, and the stream flows using Mixed Integer (Non) Linear Programming (MI(N)LP). Product design involves relating product mix and processing conditions to various product characteristics. The multi-objective approach called for this type of problem is addressed through relative weighing of the objectives in the objective function. The lumped parameters used are derived from detailed distributed models using a two-tier approach. The first example used is porous matrix-polymer composite design through impregnation and surface treatment. A second example on catalyst design is also used. The rigorous models utilize a genetic algorithm for search in the discrete variable space. Learning from the rigorous models is used to update the process flowsheets as well as the knowledge bases.

Chemical plant fault diagnosis through a hybrid symbolic- connectionist approach and comparison with neural networks

Abstract A novel hybrid symbolic approach to machine learning is illustrated for fault diagnosis of a hydrocarbon chlorination plant. The learning algorithm addressed the knowledge acquisition problem by developing and maintaining the knowledge base through inductive learning. The performance of the learning system is discussed in terms of the knowledge extracted from example cases and its clasiffication accuracy on the test cases. Results indicate that the introduced system is a promising alternative to neural networks for fault diagnosis and a complement to expert systems.