ModGen: A Model Generator for Instrumentation Analysis. Industrial Application using New Observability Techniques

VAZQUEZ, GUSTAVO E.; PONZONI, IGNACIO; SÁNCHEZ, MABEL C.; BRIGNOLE, NÉLIDA B.

Miami, Florida, Estados Unidos

Conferencia; AIChE Annual Meeting 1998; 1998

AIChE (American Institute for Chemical Engineering)

Introduction: A window-based graphical user interface (GUI) for the generation of steady-state mathematical models of process plants has been developed. It is called ModGen and it was specifically designed for instrumentation studies, with tools for easy interaction with algorithms that perform the classification of unmeasured variables (observability analysis) on the basis of structural analyses. This interface made it possible to carry out instrumentation analysis of plant-wide problems with rigorous non-linear formulations that led to realistic results. Due to the magnitude and complexity of the corresponding models, the data input and rearrangements must be computer-driven, otherwise resulting in a long hard task prone to errors. The structural techniques for observability studies constitute a topic of current research (loris and Kaliventzeff, 1987; Sanchez et aI., 1992; Ponzoni et aI., 1997; Ponzoni et aI., 1998). These methods carry out the classification by permuting the model s occurrence matrix to special forms. Ponzoni et ai. s approach in particular is focused on the design of more efficient and robust algorithms which can handle strongly non-linear equations satisfactorily. In this paper, we present the latest improvements incorporated into the Global Strategy with First Least-Connected Node (GS-FLCN) presented by Ponzoni et ai. (1997). Branching factors were designed in order to speed up the execution time. ModGen was linked to the corresponding implementation to allow treatment of industrial case studies. Main Features and Implementation: ModGen was developed using Microsoft Visual Basic under Windows 95/NT platforms. The user can enter, modify and delete units easily; choose the desired level of complexity for the equations (total/component mass and energy balances, thermodynamic relations); introduce additional equations in symbolic terms and define the measured variables. The environment is user-friendly, allowing interaction with graphical displays of the items of equipment, instruments and stream charts in a presentation style which is familiar for process engineers. The interface is intelligent in the sense that it checks the consistency of the data and it automatically defines output features, such as state or composition, whenever possible. In this way, the amount of time required to enter the data is significantly reduced and the information is always coherent. Once the model has been defined, it is possible to generate the corresponding set of equations and visualize it symbolically. As to model analysis, ModGen generates the occurrence matrix and gives information to the observability algorithms about unallowable subsets arising from parallel streams or implicit functions. The former are associated to numerical singularities, while the latter hinder the redundancy analysis. The classical software engineering paradigms were not appropriate for ModGen s development because they do not take into account the distinctive features associated with most engineering applications, i.e. high specialization and handling huge amounts of heterogeneous information. An additional requirement was modularity, since it was desirable to conceive ModGen as a multi-purpose model generator to ensure flexibility for further applications. Therefore, a new paradigm was proposed. It comprises the following four steps, which are repeated until convergence to the final product: planning, software engineering, developer evaluation and user evaluation. Planning involves the definition of objectives, alternatives and constraints, while the software engineering stage consists in building prototypes. In contrast with the evolutive model (Pressman, 1990), this is done not only at the first few iterations but at all levels of the project evolution and the concept of risk analysis is replaced by the idea of evaluation of the prototype condition by the developer in order to ensure software quality. An Industrial Application: ModGen is an indispensable tool to solve plant-wide problems concerning instrumentation design (a typical industrial plant is usually associated to more than 1000 variables and equations). In this work, we use ModGen to test a new improvement in GS-FLCN. The method was originally generated to overcome the drawbacks regarding robustness or applicability range exhibited by already existing techniques when applied to observability studies. The strategy explores the graph G(M ), where M = TM, M is the model s occurrence matrix and T is its transpose, using depth-first searches guided by the first least-connected node heuristics. Though significantly more robust, GS-FLCN lacked efficiency for flashes. One alternative to speed up the algorithm was its combination with a direct method called Cycle Detection in HyperGraphs (Ponzoni et aI., 1998), which employs a wide set of heuristics in order to go along different search paths. The method is successful, but the quality of the results depends strongly on the set choice. In this sense, no set is applicable to all problems and this constitutes a disadvantage. In this paper, we introduce another alternative. The new improvement in GS-FLCN consists in cutting back the amount of paths explored along G(M) by means of branching factors. These paths shape a tree-like search space, whose breadth is chosen by the engineer through the branching factors. In this way, the method becomes more flexible and hence, more efficient. The performance of this proposal was tested on a medium-size process plant modeled with ModGen. The results were satisfactory, revealing significant improvements in execution time, without deteriorating its robustness. Conclusions: A model generator specially adapted for instrumentation purposes was developed. It is modular, user-friendly and allows easy handling of plant-wide problems. The software can interact with observability packages. Its performance was assessed with several classification algorithms applied to industrial process plants. In particular, we describe a new improvement in GS-FLCN s structural technique based on branching factors. ModGen is versatile because it would be possible to implement with little effort interfaces for other purposes, such as steady-state and dynamic simulation or optimization. References: Joris P., KalitventzeffB. "Process Measurement Analysis and Validation", CEF 87, 41-46, Rome, 1987. Ponzoni I., Sanchez M.C., Brignole N.B. "A New Partitioning Algorithm for Classification of Variables in Process Plant Monitoring", AIChE 1997 Annual Meeting, USA, 1997. Ponzoni I., Sanchez M.C., Brignole N.B. "A New Hybrid Approach for Instrumentation Design of Chemical Plants", IV WCCM, Buenos Aires, 1998. Pressman R.S. "Software Engineering", Mc Graw Hill, 1990. Sanchez M.C., Bandoni A.J., Romagnoli J.A. "PLADAT: A Package for Process Variable Classification and Plant Data Reconciliation", Compo Chem. Engng,S499-S506,1992.