COUPLING CONSTRAINT BOUNDARY MAPPING IN THE PROCESS DESIGN PARAMETER SPACE WITH COMMERCIAL PROCESS SIMULATOR TO ESTIMATE PROCESS DESIGN RELIABILITY

Abstract

Chemical process designs include safety factors to compensate for inherent parameter uncertainty in the design process. These safety factors require additional capital and operating expenses. Since these factors are based on rules of thumb, they may be ineffective and wasteful. The certainty that a process will meet process constraints during normal operation despite the underlying uncertainty in process design parameters is the process design reliability. The estimation of the reliability of a proposed design and an evaluation of the safety factor effectiveness in increasing reliability would identify which equipment is unnecessarily oversized, which is critically undersized and which uncertainties are the principal contributors to low reliability. The designer could then adjust the safety factors to optimize reliability, capital investment and operating expenses. Traditionally, reliability has been evaluated by conventional Monte Carlo integration. This methodology is computationally too expensive since it requires a large number of model simulations. Recalculation for sensitivity analysis is prohibitive. An alternative is required. Furthermore, an evaluation tool that assists designers to maximize reliability while minimizing cost would be substantially useful in a commercial process simulator. A computationally efficient methodology for estimating reliability, Monte Carlo integration of the c-constraints mapped onto the p-parameter space, was developed, improved and coupled with a widely available commercial process simulator, CHEMCAD. In this methodology, the design's success region is mapped onto the parameter space. Sets of parameter values falling on the constraint boundary of the success region are found through process simulation coupled with a search algorithm. This search is independent of the parameter uncertainty. These boundary points are then connected via hyper-planes through interpolation. These connected hyper-planes represent a hyper-volume. Parameter sets falling within this volume successfully meet all constraints. Monte Carlo integration of the parameter uncertainty within this volume leads to an estimate of the process design reliability. This integration does not require process simulations. The procedure adds new boundary points in the regions of greatest uncertainty to improve the reliability estimate. The methodology requires two to three orders of magnitude fewer simulations than conventional Monte Carlo. The method is coupled with CHEMCAD through an EXCEL-driven central program making the method widely available to the process design community. Safety factor impact on reliability can be readily evaluated. Viability and efficiency are demonstrated using distillation case studies based on industrial challenges.