TURBINE DESIGN OPTIMIZED IN JUST FOUR DAYS USING ALGOR FEA ON A PC

Engineer Tony Enos analyzing design of tangential 60,000 RPM turbine.

Finite element analysis (FEA) on a personal computer has enabled a producer of turbine testing equipment to optimize a rotor design, allowing a significant increase in power while maintaining an infinite fatigue life. Over the course of four days, Barbour Stockwell Company made three iterations of the rotor model, without constructing a single prototype, arriving at a design in which the additional load caused by a larger bucket was redistributed along the length of the shaft, dramatically reducing bore stress at the disk level.

The firm, which did not have a minicomputer, considered time-sharing to perform FEA. But Barbour Stockwell staff engineer Tony Enos relates, "We decided it would be too much out of our control, so we began evaluating the FEA software on the PC level. We have many PCs with commercial software, as well as a fair amount of software written in-house."

Since December of 1986, Barbour Stockwell has used the Algor Supersap Full Stress and Dynamic Modeling and Analysis package on an IBM AT with the NECK multi-sync color monitor. This Algor FEA software package provides tools to create and analyze models with powerful modeling and meshing capabilities, as well as impressive graphics for pre- and post-processing.

"We evaluated a number of personal computer-based FEA programs and decided that Supersap was the most proficient program for our applications, while the price was very reasonable. Supersap is also very easy to use, unlike some other technical programs, which require you to be a specialist in order to effectively run them. All the engineers at Barbour Stockwell use Algor.

"Without Algor, we wouldn't have been able to get the perspective on the part's stress areas," says Enos. "Algor allows you to see very quickly how the stresses are distributed and the relationship between the elements making up the model. On the screen, I could take the deformed model and overlay it on the original model and actually see effects of the different shapes that I was inputting.

"When the model is analyzed under load," continues Enos, "the principal stress lines may be in any direction and not necessarily in line with the coordinate axes. Without calculating each and every element by hand, there isn't any way to determine all of these stresses. Fortunately, Algor does this for you." Therefore, the Cambridge, Massachusetts firm has begun to employ the PC-based program to support the design of other equipment.

Barbour Stockwell, located next door to MIT and a participant in the institute's industrial liaison program, produces customized equipment related to their test turbines for the gas turbine industry (jet engines in particular), the automotive industry, aerospace industry, fluid machinery industry, and government laboratories.

Barbour Stockwell's customers commonly use turbines to test a part's integrity or endurance. The part is often either cycled up and down to cause fatigue or is oversped to reveal any fatal manufacturing discontinuities. With the stage of a jet engine, for example, which may start 3000 times before overhauls, they would use a test turbine to drive it up and down 3000 times, accelerating the stage up to a predetermined speed and then decelerating it back to an unloaded condition, simulating the cycle life of the part. After the test, they would take the stage out and inspect it with X-ray or other inspection techniques to see if the stage had developed any fatigue cracks.

Additionally, the firm builds entire systems, such as various custom-designed "spin pit" facilities, which include a vacuum chamber in which the part is tested, the test turbine, a complete control package including data acquisition, along with all the peripheral equipment, such as the vacuum pump, an air compressor and lubrication units. Barbour Stockwell also designs high-speed photography systems to film the bursting of parts which enables it to actually see how the part will fail and reveal the failure points.

In this case, Barbour Stockwell was contracted to design a tangential 60,000 RPM turbine with increased power output over their standard turbine, while keeping the stresses below the endurance limit.

"We went through quite a bit of analysis," relates Enos, "to get the rotor profile that would give us infinite fatigue life, along with the required turbine power output. Since power is produced by directing high speed air into row buckets machined in the disk's periphery, the two things which affect turbine power are the momentum of the air and the radius of the disk. There is a relationship which must be met between the air velocity and disk tip speed. This relationship dictates the disk diameter for a maximum power output at a particular speed. Therefore, in order to obtain a higher power output in the same operating range, the air momentum had to be increased. Since the air velocity couldn't be increased, the mass air flow was. In order to accommodate the additional air flow, the rotor buckets had to be larger, thereby, increasing the amount of mass on the outer portion of the rotor disk."

The rotor was designed to have approximately a four-inch disk diameter and would be made of a high strength material which is used extensively at Barbour Stockwell. Enos says the firm has developed a process for machining rotors from this high strength material and has a great deal of experience and success with it.

To create a model of the rotor, Enos used Algor's own full-featured graphics drawing program, SuperDraw II, which provides tools to create models onscreen with a combination of lines, arcs, circles, and text. The firm uses AutoCAD for CAD/CAE and could have used it to create models for FEA, since Algor's Superlink program can interpret data from several popular CAD programs.

"But we used SuperDraw II," says Enos, "because it's more efficient than AutoCAD in creating drawings specifically for a FEA model. We had in excess of 1000 elements, so to draw every one of those would have taken many hours.

"To make the model, I essentially drew a profile of the rotor with the disk. I put a generous radius at the corner of the upper and lower face of the web and the protruding shaft. The outer disk diameter had to be approximately twice as thick as the disk web area and there was also a generous connecting radius there."

Enos used Algor's MSHGEN, an automatic mesh-generator, to mesh the 2-D axisymmetric model. Enos entered keynodes into the mesh to define the outline of the model, then defined the keynode regions by dividing the model into polygons. Next, Enos inputted the locations of intermediate nodes, then specified the density, edge curvature and pressure for each region. During the process, Enos was able to view the mesh onscreen using TDraw, allowing him to visually check the mesh as he built it. Algor's mesh program computed sectional properties, including moment of inertia, automatically checking adjoining regions for compatible curvatures and meshing divisions.

"In disk stress," explains Enos, "the most critical area is the bore, so I put a row of 0.001 inch thick elements all the way down the shaft. These very thin elements would allow me to view the bore very closely, to see what the stresses are on both faces and the center of the element. I tried to align the elements in such a way that they would be in line with the expected stress lines of the rotor."

With mesh completed, Enos used Algor's AEdit to establish the boundary conditions on the model. He attached springs only to one element, which was in the lower shaft, to try to remove it from the disk area of the shaft as much as he could to keep it from affecting the overall model. "I put an infinite spring," Enos states, "a spring rate of one billion pounds per inch, on it so during the analysis this point became fixed."

Enos used the graphics display system TDraw to examine the results after running the analysis, displaying the resultant deformed model on top of the original, in different colors. "With the deformed model overlaid on the original, it becomes very apparent which part of the bore is the most highly stressed because it is the most highly deformed," says Enos. "You can narrow in on that very quickly. I could clearly see that the stresses were much too high, so I used TDraw to find the deflected nodes, zooming into a certain section and having TDraw number the nodes to give me a quick reference for the tabulated data."

Enos also checked the stress according to the printouts, explaining that Algor printouts record the data in tabular form, referencing a node, an element, or an element face against the resultant stress or principal stress.

Based on the results of the first analysis, Enos built a second model with SuperDraw II. "I thinned the center web of the disk where it seemed the stress was relatively low," he describes. "I increased the radius of the web to the outer area of the disk. I also noticed from the displaced model that the stress wasn't quite following the radius down to the shaft the way I wanted it to, so I added a little material to the web so the disk shaft radius was met by the web with a slight angle."

Using the same boundary condition, Enos analyzed the modified model. "This model yielded much better stress levels," he says. "When I ran the displacement display in TDraw, all distortion seemed symmetric. Everything appeared to be all right except at the shaft bore at the disk level. It was still taking too much load. So I decided I had to distribute more of the stress over the length of the shaft."

Enos modified the model again, increasing the angle of the web even further, giving him a variable thickness disk no longer having parallel sides. Completing the modifications with SuperDraw II, Enos ran the analysis for a third time, again using the same boundary condition. Now the stress followed the radius down shaft the way he wanted it to.

"The resultant model was a dramatic improvement. It was exactly what we wanted. My modifications had led to much of the bore stress at the disk level being redistributed along the length of the shaft. This dramatically reduced the stress at the disk level. The model now had acceptable stress levels," notes Enos, "and we were well within the fatigue life stress limits, providing a rotor with infinite fatigue life. I took only four days to arrive at this optimal design."

With the success of Algor's FEA in helping to design the optimal rotor, Enos says the firm plans to employ Algor's PC based software in future design upgrades and any new designs. According to Enos, Algor's Supersap Full Stress and Dynamic Modeling and Analysis package provides exact analysis and opens the door to testing applications not possible by other means. Supersap also integrates FEA with graphics, enabling easy modeling and meshing of structures. It provides perspective on part's relative distortion under stress, and has most of the features of mini-computer level finite element software, but at a fraction of the price.

Copyright © 1987 Algor, Inc. All rights reserved.