"Making Little Things Count," ME Online Web Exclusive, Sept. 10, 2003

for 9/10/03

Making Little Things Count

Because it had the right skills and software, an engineering firm was able to conclude its design analysis and verification of a refinery expansion project on time, despite stringent deadlines.

This article was prepared by staff writers in collaboration with outside contributors.

When engineers at BES Engineering, a unit of The Bergaila Companies, tackled their latest design analysis and verification of a refinery expansion project, time was of the essence. Looking at California's electricity problems and sharply rising gasoline and oil prices, owners wanted construction to get under way as soon as possible.

Nevertheless, the refinery owners and contractors were unwilling to rush into construction. They did not want any mistakes, especially ones that might not show up for years. So they asked BES in Houston to review the design for a variety of potential stress and thermal problems. To the refinery's designers and owners, it was the little things that counted.

As the analysis got under way in the first quarter of 2001, some of those little things turned out to have a big impact on the project. For example, the BES engineers had to:

• Transform the refinery drawings and models from the "local" system of coordinates typically used in piping design systems to conventional CAD-type coordinates. Doing this more or less automatically let BES get started as much as two weeks sooner on the modeling, finite-element mesh generation, and analysis.

• Use their geometry for multiple analyses on one portion or another of a very large model, one with 120,000 nodes and 700,000 degrees of freedom.

• Use a variety of solvers for their analyses. The model was too big to run on all but the speediest and most powerful workstations using frontal-type conventional finite-element analysis solvers. Instead, they used a preconditioned conjugate gradient solver.

Because they had the skills and software to do these things, BES was able to meet tight deadlines for the new petroleum refinery unit being built in North Salt Lake City, Utah.

The project was the Flying J Inc. Millisecond Catalytic Cracking (MSCC) unit, a 10,000-barrel-per-day facility. This process unit required a regenerator vessel that was designed and built by Mark Steel Corp., a Salt Lake City custom steel fabricator. Mark Steel subcontracted the engineering analysis and drafting of the regenerator pressure vessels and related components to BES Engineering.

BES Engineering used the Structural module of Ansys Mechanical software, a leading design analysis and verification package from Ansys Inc. of Canonsburg, Pa.

The analysis-linear elastic-focused on the cracker's regenerator, a slender, refractory lined vertical, cylindrical pressure vessel with a hemispherical head at the top and a cone with four outlet nozzles pointing sharply downward at the bottom. The cone section analyzed was about 11.5 feet in diameter at the bottom and 17 feet at the top. The overall vessel length is about 94 feet. The largest outlet is 72 inches in diameter, while others range from 42 inches to 2 inches in diameter. The specified pressure vessel material was carbon steel ranging from 0.75 inch to 1.75 inches thick.

One of the meshing challenges was that the nozzles pointed downward at an included angle of about 160 degrees. Because of the sharp downward angle, the regenerator openings that accommodated the nozzles were elongated ovals, their vertical dimensions much longer than the horizontal.

"The large included angle made the openings very large relative to the pipe diameter," said Dana E. Petroni. He and other BES design analysts built the models and ran the analyses. "Most analysis packages have trouble producing good results with a nozzle fitted to an oval hole if the long dimension is more than twice the short one," Petroni said.

Adding to the analysis challenges is the refinery's site in a moderately active seismic area defined as Zone 3 by the American Society of Civil Engineers; California is Zone 4.

"Without all these things, this would have been a fairly simple problem," Petroni observed. "As it turned out, we spent two and a half man-months on the analyses with Ansys." The job included making design calculations and writing a detailed, illustrated report.

"We had to build six or seven smaller models, including one with special mitered nozzles for the main inlets and outlets," Petroni noted. The analyses indicated that two design changes were needed. They were:

• The junction of the regenerator's transition cone/catalyst cooler nozzle, should there be catalyst cooler slump combined with a seismic tremor. The analyses predicted excessive stress and excessive radial displacement; this could be remedied by increasing the thickness of the junction's steel.

• The carbon steel knuckle joining the regenerator shell/cone to the combustor riser cone would see too much heat from the steady-state operating temperature of 1,000°F. Suggested remedies included a material change, a fatigue evaluation, and redesigning the junction's attachment details.

The main model-comprising the regenerator cone, the four attached nozzles, and the openings-was built with 40,000 elements. When completed, the model had 120,000 nodes and approximately 700,000 degrees of freedom.

"Because of its complexity, that one model let us take care of four analyses-and each of those had five load cases," Petroni said. "If we'd had to build additional models, it would have taken as much as two weeks longer."

The analyses were for thermal stress, mechanical stress, local distortion of shell-nozzle junctions, and/or temperature profile. The four analyses on the big model were done on the transition cone junctions for three standpipes (catalyst regenerator, catalyst recirculator, and the hot stripper) plus the catalyst cooler.

"Using the normal frontal solver, nearly 40 gigabytes of storage space would have been required," Petroni said. He used an Ansys preconditioned conjugate gradient, or PCG, solver.

PCG solvers are iterative; that is, the results of each time-step analysis in the solution become the initial conditions for the next step.

Typical time to solve these analyses was about 12 hours per load step, Petroni said. His group used custom-built PCs running dual 350-megahertz Intel Corp. Pentium II processors. The machines have 256 megabits of RAM.

An upgraded, custom-built PC running on dual 500 Mhz Pentium IIIs (with 390 MB of RAM, a 64 MB video card, and a 17 gigabyte disc drive) is used solely to run models, so BES engineers can continue working on other projects. On these machines, runs were done in three hours per load step.

Three other sections of the regenerator shell were analyzed, using models ranging from about 70,000 nodes down to 48,000. These models had 420,000 and 288,000 degrees of freedom, respectively. These were for the regenerator junctions with the catalyst/air inlet to the elliptical head, flue gas outlet and plenum at the hemispherical head, regenerator combustor riser cone, recirculation catalyst standpipe, and two manway access openings.

The structural stress analysis loads included dead weight, design pressure, operating pressure, catalyst slump (partial blockage of the piping), piping load, seismic load, and thermal gradient. The piping loads include effects of normal operation (temperature, pressure, and weight), catalyst slump, expansion joint pressure thrust during design and blast conditions, and seismic acceleration loading (potential earthquakes).

These individual loads were combined to produce and determine governing load cases. Not all loads and load cases were considered in every analysis. All of the regenerator is lined with 5 inches of refractory material. "This was not modeled because its stiffness was not a factor in the structure," Petroni noted. However, it was factored into the load cases as a deadweight factor for displacement or gravity loads and for its relatively high sensitivity to thermal expansion.

The two thermal analyses used steady-state heat transfer data, taking into account convection on the outside surface of the pressure vessel and a process temperature applied to the internal wetted surface. Petroni used forced convection to determine the thermal gradient that governs the thermal stress analysis.

Free or natural convection was used to determine the thermal profile that, in turn, limits the maximum operating temperature of the pressure vessel's carbon steel. The structure was designed to the American Society of Mechanical Engineers' Pressure Vessel Code.