A while back, we received an emergency call from a pulp mill client in Canada. He had just suffered a catastrophic failure of a gigantic centrifugal compressor which was used to provide suction to his evaporators.

Operating at nearly 6,000 RPM and driven by a 5,000 HP motor, the five-year old compressor had failed without warning. Fortunately, no one was in the immediate vicinity of the compressor when it failed, so no one was hurt. However, flying debris had severed some hydraulic lines in the compressor building and triggered a brief but intense fire. The fire had only been out for a couple of hours before our client was mobilizing to assess the damage; could we catch the next plane and be on site as soon as possible?

A Jumbled Mess

Once we arrived on site we found quite a mess. The workers were just starting to disassemble the inlet piping, which would provide us with access to the impeller. However, it was already apparent that the failure must have been spectacular. The main shaft was fractured, which had allowed the rapidly spinning impeller to drop and wedge in the housing. In a fraction of a second, the immense rotational energy was being dissipated as heat and fracturing metal.

The cast iron housing, which was over 13 feet in diameter and weighed 35,000 pounds, had many large through-wall cracks, some of them up to 10 feet long. Peering through the cracks in the housing, we could see that the rotor was severely torn and bent from its impact with the housing. Several blade fragments had been thrown through the housing, the bearing cap was fractured, pieces of the bearing were strewn about the room, and of course, everything was covered with soot, ash, and melted aluminum conduit from the subsequent fire.

How to Sort It All Out?

So, how to piece together the fragments and figure out what happened first? Perhaps a bearing had seized; or perhaps the shaft had fractured; or perhaps the bearing cap had failed. Whatever it was, there had been a lot of collateral damage which we would just have to sort out by a systematic examination of all the evidence.

A Clear Cause!

After several hours work, the workers finally completed removing the inlet piping, giving us our first clear look at the rotor. Although bent and distorted, one thing stood out: one of the rotor blades was gone, and its fracture surface on the hub clearly showed there had been a massive fatigue crack present when the failure occurred.

Quickly then, the pieces of the puzzle began to fall into place. The rotor blade had apparently developed a fatigue crack which had propagated until the remaining section could no longer support the service loading, at which time it fractured completely. Although the fractured blade had no doubt done some damage on its way out, the real damage was done by the huge imbalance to the rotor caused by the blade's departure.

At 6,000 RPM, the rotor quickly shook itself to death. Whether the shaft or the bearing housing failed next was a moot point; once the blade failed, the unit's fate was sealed.

The Next Question

Of course, the next obvious question was why had the blade failed? However, this would have to wait for a laboratory analysis of the rotor. Of more immediate concern to the client was how to get the system back on line with a minimum of downtime.

A spare rotor was available in the client's warehouse, but what about the housing? Because the housing was cast iron and had such extensive cracking, we recommended against an attempt at weld repair. Several phone calls by the client quickly confirmed the worst: No spare housing existed (in the world!) and it would take up to 10 months to build one.

Housing Repair

It was at this point that an innovative solution to the housing problem was suggested by a firm from Baton Rouge, Louisiana. Their specialty is the mechanical repair of failed or cracked machinery castings. Starting with a finite element analysis of the stresses in the housing, they engineer an on-site repair procedure that can generally be completed in a matter of weeks. This procedure consists of a combination of mechanical stitching and external, reinforcing gussets.

First, the cracked casting is forced back into position with hydraulic jacks and appropriate bracing. The crack is then "stitched" back together by drilling out a series of dumbbell-shaped patterns perpendicular to the crack and pressing alloy steel dumbbells into the casting in place of the drilled out material. These dumbbells provide a mechanical lock to hold the mating crack halves together. The cracked area of the casting is then further reinforced by a specially designed set of external gussets which are welded and/or bolted into place.
 

Back to the Laboratory

The repair is intended to be permanent, and with the approval of their insurance carrier, the mill proceeded with the arrangements for this. Meanwhile, we returned to our laboratory to proceed with a failure analysis of the failed rotor.

The first step was a thorough nondestructive inspection to see if there were any other cracks; sure enough, a second blade had a major crack nearly as long as the one that had caused the service failure, and a third blade had a small, 1/4-inch long crack.

A fractographic analysis of the cracks revealed they were of recent origin and had been actively growing. Based on the estimated growth rate of the cracks and the cyclic loading frequency of the rotor, we estimated that the second crack had been only a mater of a few hours from catastrophic failure itself when the first crack let go.

Elsewhere, we found no significant problems. The hardness and chemistry of the blades and hub met the specifications. And although there were some interesting microstructural segregations within the blade due to its chemistry, these had not been a factor in the failure. Also, even though we found a slag imperfection and a small hot crack near the fatigue crack nucleus, a close examination of these revealed no evidence of service extension, indicating that neither had been a cause of the fatigue crack. So, why had the blade suddenly developed fatigue cracks and failed after five years of uneventful service? One possibility was that the service loading had suddenly changed for the worse; however, the client didn't believe this had happened.

Resonant Vibrations

The only other clue was a small amount of erosion at the blade tips; this turned out to be the culprit. According to the compressor manufacturer, the rotor had a critical speed at which it was not supposed to be operated because doing so would result in destructive resonant vibrations.

The resonant frequency is a function of blade mass, and apparently, the erosive loss of material at the blade tips had gradually reduced the blade mass, causing the rotor's resonant frequency and critical speed to gradually shift until it coincided with the rotor's normal operating speed. Thus, what had been acceptable service for 5 years, eventually became self destructive. With the rotor thrown into resonant vibrations, it had quickly developed the fatigue cracks which led to its demise.

As we mentioned in our last newsletter, MEI-C, in cooperation with the City of Portland and Saturday Academy, sponsored Nicole Salvus, a junior at Benson High School, in the Apprenticeships for Science and Engineering program. Nicole's project at MEI-C last summer was a study of the permeation of solvents through disposable gloves. As discussed in Part I, the permeation rates for pure solvents varied a great deal, depending on the solvent and the type of glove.

The second phase of Nicole's project was to investigate how the gloves responded to mixtures of solvents. The question we hoped to answer was, "are the permeation rates for mixtures directly proportional to the concentration of the components?"

We started by mixing acetone and water. Testing different proportions for both latex rubber and nitrile gloves, we found that the permeation rate was not directly proportional to the acetone concentration. Permeation rates were much lower for the mixtures than would have been predicted based on the permeation rate for pure acetone.

The graph shows the strongly decreased permeability, or inhibitory effect, of acetone-water mixtures on nitrile gloves, compared to pure acetone. A 50-50 acetone-water mixture only diffused through the glove at one-eighth the rate predicted by a straight line proportion. So, one could expect that with latex gloves a similar one-eight rate would apply, right? Of course not! With latex gloves, the rate was generally about one-fourth the straight line prediction; and oddly, at a concentration of 80 percent, acetone initially diffused through the latex gloves at a rate similar to its rate through nitrile, but then, after about 45 minutes, increased to a rate about the same as pure acetone.

Next, we switched to ethyl acetate and water and reran the same experiments. Here, mixtures diffused through the latex gloves as predicted, but with nitrile gloves the diffusion rates varied from way below to above the straight line prediction.

Another mixture we tried was toluene plus ethanol. Again, latex gloves had permeabilities that were close to the values predicted by straight line averaging, while nitrile gloves had inhibited permeabilities for the mixtures.

One very complex mixture of great interest is gasoline. Gasoline is a mixture of many different compounds, one of which, methyl tertbutyl ether (MTBE), is of concern to many people because of its toxicity. In this instance, we measured the vapors coming through the glove with gas chromatography. This enabled us to determine and compare the permeation rates of each separate ingredient in the gasoline.

The results showed that nitrile gloves gave better protection than latex for all the ingredients, including the MTBE. The gasoline came through the latex gloves immediately and at a fairly constant rate, while the nitrile gloves provided very good short term protection, about ten minutes, before becoming permeable.

In summary, the answer to our initial question about whether the permeability was directly proportional to the solvent concentration, was no; the situation was much more complicated than that. In some cases, the effect of the mixture was synergistic, in others, it was inhibitory.

For some solvent combinations, gloves can be good barriers initially and then deteriorate rapidly, or be poor initially and then get better, or be good at one solvent concentration and poor at another.

All in all, this was an excellent opportunity for an aspiring young scientist like Nicole to see first hand how real world science usually raises more questions than it answers, and it was a fun reminder for us also!
 

Wilma E. Czyzewski died December 20, 1997 at age 83. Mrs. Czyzewski and MEI founder, Harry Czyzewski, had been married since 1943. The family suggests remembrances be given to the American Heart Association.