Failure Analysis

The San Bruno Pipeline Disaster of September 9^th, 2010

On September 9^th, 2010, a 30-inch, PG&E natural-gas transmission line exploded in a quiet residential neighborhood in San Bruno, California. The disaster destroyed 38 homes and eight people were killed as a result of the blast. The National Transportation Safety Board (NTSB) launched an investigation the next day; and on March 1,2 and 3, the NTSB will report their findings (to date) in a briefing to be held in Washington, D.C.

Since that fateful day, there has been much discussion in the media about the pipeline, how it was fabricated, installed and inspected; the existence or non-existence of relevant paperwork, over-pressurization of the line. What is needed is factual information and this should become available through the March hearings. What I hope to do with this overview is to help the public to understand the perspective of a metallurgical engineer and the role he/she plays in the accident investigation. I hope to present a series of engineering schematics that may help one to understand the relevant questions in the failure analysis of any failed, thin-walled pipeline including one like the San Bruno natural-gas transmission line.

Before we get started, let’s look at one of the key pipe-sections from the disaster. This large section of pipe came free and was thrown clear, creating a huge pit at the ‘initiation’ point of the explosion. It will be a ‘main feature’ in the upcoming hearings.

Please note the following: the two ends of this one-section of pipe look distinctly different. On the left, it looks like someone took a knife and severed the pipe in half; on the right, the pipe is split open like a gutted fish, a seam running parallel to the rails (or pipes?) lying on the ground (along the length of the pipe). Will analysis of this section of pipe help us to understand what occurred in San Bruno on September 9, 2010?

Let’s begin. The pipeline is what an engineer would call a ‘thin-walled, pressure vessel’. It is ‘thin-walled’ because the thickness of the wall is small compared to the diameter of the pipeline. This is a cylindrical pressure vessel and the stresses in such a structure, pressurized to 400 psi in our instance, are well known. Engineers know how much pressure it would take to fail the pipeline; and they apply a certain ‘safety factor’ against this to set an ‘upper-limit pressure’ or ‘allowable pressure’ that the pipeline can sustain with no problem. Of course, this assumes that the nature of the pipeline and the material used in its construction remains unaltered. Of course, we all realize that this is not possible. Corrosion, for example, will ‘rust’ steel, and thus the material, at least at the surface, inner or outer, is not the same. So, one must inspect the pipeline with some regular frequency to assure that conditions are known… not that they haven’t changed, but that the new pipeline condition will not adversely impact the integrity of the pipeline.

Now many of these issues are in question, as regards the utility and contractors that maintained the service. Let’s focus solely on the nature of stresses that push and pull on the steel membrane of this thin-walled pressure vessel when the system is pressurized.

Figure 1 shows generalized stresses: shear stress and normal stress. To understand shear stress, place an imaginary, but heavy, cardboard box on the floor and try and slide it across the floor. Do so by placing your hands on the top of the box (awkward and not too efficient) and push. The box will take the shape you see in Fig. 1, won’t it? You are applying a shearing load to the box.

Alternatively , lets try to pull a rope apart in tension. We apply a ‘normal’ force (ie, one normal to the cross-section of the rope) and the rope gets longer and narrower. We are applying a ‘normal stress’. These are the two main types of stress active throughout the containment membrane of our pressurized cylinder and they can be schematically illustrated.

Now here shows one, tiny-tiny cube of steel that makes up the ‘wall’ of our pipeline; and also shown are the normal stresses that act on it. I have one very special cube, oriented in such a way that the shear stresses are ‘zero’ for this little guy; and only the normal stresses show. I do this so that you see that there are normal stresses acting in the three directions: x, y and z. These are very important when consider ‘why’ something failed. You will see this in the following storyline: the failure of a toy balloon.

A toy balloon is a pressure vessel! Too, it is thin-walled, now isn’t it? The rubber membrane thickness is a small fraction of the radius of the balloon. Do we know the stresses in a pressurized balloon? The stress is a function of the amount of pressure (say 15 pounds per square-inch), the radius of the balloon, say 12-inches; and the thickness of the membrane (say .040 inches). Using the equation in Fig. 3, the stress in the membrane containing the 15 psi pressure is 2,250 psi. We need a strong material to fabricate our balloon and contain this pressure!

Note that a ‘small element’ of the membrane (shown in Fig. 3) has normal stresses acting perpendicular to each other. There is a third stress; the stress pushing on the surface of the balloon. On the outside of the balloon, there is zero gage pressure acting; but, on the inside of the balloon there is 15 psig acting. The inside surface of the balloon is carrying a slightly larger burden to contain the pressure than is the outside surface of the membrane. Interesting, ain’t it!

So now, let’s take a toy-balloon and begin to blow it up. Let’s keep blowing until we make the balloon pop (ie, fail). See Fig. 4

The ‘relaxed’ state of the balloon before pressurization is state-1. As we progressively pressurize the balloon (ie, blow it up), we go through states 2, 3, 4, 5 and then 6. At state-6, the balloon bursts! Pop!

At this point, I want to introduce another ‘engineering diagram’ called the Mohr’s Circle. It is a representation of the stress state of one of these small units or cells (see Fig. 2). The normal stresses are plotted on the horizontal axis; and the shear stresses on the vertical axis. Now there is an ‘orientation’ story here; but it is outside the scope of what we need to understand basic stresses in pressurized pipes, so I’ll ignore it. You should concentrate on what is happening right on the two, perpendicular axes!

In Fig. 5, we see the Mohr’s Circle of Stress for the balloon example of Fig. 4. The maximum shear stress at any pressurization is seen as the vertical ‘height’ of the balloon on the coordinates shown. The maximum, tensile normal stress is shown on the x-axis (ie, the normal stress axis) to the right; and you will note there is a small, negative or compressive normal stress to the left. Where does the compressive stress come from? It represents conditions on the inside surface of the balloon. The degree of pressurization (in the example, the 15 psi) within the balloon accounts for this negative stress value. A progression of circles are represented matching sequential pressurization, from state 2 to 3, to 4, to 5 and to 6, where the balloon finally bursts.

We can introduce the concept of a ‘failure envelope’ in this diagram, Fig. 5, the point at which the balloon bursts. All would agree that there is some pressurization at which the strength of the membrane material is insufficient to bear the load/stress to contain the pressurized gas. The two vertical lines in the diagram show this strength limit: the vertical envelope (or actually, the point on the normal stress axis) marks the limit for normal stress, and the horizontal envelope (the point of maximum shear stress) marks the limit of shear stress to which the membrane material will remain intact.

If you were going to throw a party for your 5-year old and balloons were going to be distributed, it may be wise to set a particular balloon diameter that should not be exceeded so that the children will have fun instead of being terrorized! You might call the limit the ‘allowable balloon diameter’. A good guess at what this diameter might be would be, for example, 2/3 of the critical diameter, the critical diameter being equivalent to the pressurization level that would pop the balloon. You have thus set a ‘design limit’!

Pity the poor parent that chooses a balloon with a small manufacturing flaw! Before they reach the allowable balloon diameter, the balloon bursts! What has happened? The flaw was not taken into account! We’ll get into this a little later.

Fig. 6A shows a long, cylindrical pipe. It is a pressurized thin-walled (because low value of the ratio of the thickness of the steel to the radius of the cylinder) pressure vessel. Just like the balloon, there are two principal stresses to note. One is the Hoop Stress and the other, the Longitudinal Stress. The hoop stress tries to split the pipeline along its length. The longitudinal stress tries to blow the two ends off (assuming this long cylinder is ‘capped’ at both ends. As you will soon see, for a given pressurization level, the stress trying to blow the ends off is only one-half of the stress trying to split the cylinder along its length! This is profoundly different than the spherical balloon where the two normal stresses balanced.

So, let’s pressurize a thin-walled cylinder and see what happens. It is a metal cylinder being tested in what is termed, for obvious reasons, a ‘burst test’. See Fig. 6B, below

Fig. 6B- Burst test of a metal capped metal cylinder (Internet source)

Now take a look at this; and then refer back to the photo showing the section of pipeline that was thrown from the pit at the San Bruno Pipeline Failure site. What do you see that is similar; and what that is different?

Before we leave this comparison, let me offer one more photo from the Internet (labeled Fig. 3 and captioned) showing failure of a pipe near a 45-degree nipple, failure that occurred in a circumferential weld-joint. Again, compare this photo with the fracture surfaces seen in the second photo on this web page, the one showing the critical pipeline section from San Bruno. What do you note?

You need more information, so let’s return to our story-through-schematics approach. See Fig. 7, below.

In Fig. 7, we are blowing-up a wine-barrel with pressurized water. Water is not compressible like an inert gas, so when this barrel bursts, stuff won’t go flying in all directions! The purpose of the figure is to reinforce the concept of ‘hoop stress’, like in barrel-hoops that hold the staves in place. The hoop stress is twice the value of the next-biggest normal stress, the longitudinal stress. Take a look at Fig. 8.

In Fig. 8, we see the Mohr’s Circle for the stress states in a pressurized, thin-walled, pressure vessel or pipeline. The hoop stress is stress-state 1 in the schematic. The longitudinal stress is stress-state 2; and the internal pressurization of the pipe-line results in stress-state 3, the compressive normal stress, representing conditions on the inside surface of the pipe. One can see the maximum shear stress (the radius of the Mohr’s Circle); and the area that is ‘pink’ marks the stress state in the steel membrane for any imaginable orientation of the small-element (or cube) being considered. You will only need to keep your eye on the prize…. the maximum tensile normal stress and the maximum shear stress.

As we did in the case of the balloon, let’s consider sequential pressurization of our pipeline: see Fig. 9.

It is kind of hard to follow, but let’s try. All three normal stresses are zero before we begin to pressurize, so ss-1 = ss-2 = ss-3 = zero and we are at the origin where the two coordinate axes cross. Now let’s follow stress-state 2, the Longitudinal Normal Stress as a function of pressurization. As the pressure increases, the two inner circles grow as does the overall, large circle, keeping the same pattern. The left-edge of the big-sphere pushes to the left in response to the (compressive) pressurization. The issue now is ‘how far can we go?’.

In Fig 10, we again see the stress state for an element along the length of the pipeline and here, show the magnitude of the stresses. Note that the Hoop Stress (1) is twice the Longitudinal stress (2). Both are a function of the pressure, pipe radius and wall thickness, just as was the case with the spherical balloon.

For the sake of example, and using make-believe numbers, let us say that the pressurization is 400 psi, the radius of the pipe is 15-inches and the wall thickness is .350 inches. The hoop stress in the steel membrane of the cylinder would then be 17,000 psi, a very reasonable stress for carbon-steel to handle. Our failure-envelope(s) would be well beyond reach! If the membrane were thinner, the stresses would go up for the same degree of pressurization. The longitudinal stress, in the above example would be about 8,500 psi, a factor of two less than the hoop stress.

In Fig. 11, we return to the Mohr’s Circle and introduce the concept of the failure envelopes. We indicate one for shear and one for a critical normal stress. At the pressurization shown, the pressurized pipeline would be fine. But do the failure envelopes remain fixed? Do they represent ‘off’ conditions? Let’s consider this in Fig. 12.

Suppose the pipe contains a crack, for one reason or other. Let’s make it a longitudinal crack, just the kind hoop stress would love to tear apart. What does this do to our failure envelope? It pushes it to the left, as shown. Now, considering this ‘revised’ failure envelope, the pipeline will fail at a lower hoop stress than the homogeneous membrane. As the crack grows bigger (and it can, with time, as will later be shown), the failure envelope will move even further to the left. At some time, it may even be reduced enough to intercept the ‘allowable’ or ‘safe’ normal stress limit; and what was safely contained at say 400 psi can no longer bear the load. Catastrophic failure will result.

Does it have to be a crack to influence the failure envelope? No. Someone may have used the wrong steel to make the pipe; or the pipe may be improperly welded. Many things can alter the position of the failure envelope. Thorough initial inspection and inspection over time must be employed to determine the real or operational failure envelope.

Fig. 13 is interesting. Here we specifically consider material conditions in the longitudinal direction. Suppose we have a transverse crack (or perhaps a circumferential weld, joining segments of pipe together). Now a failure envelope to consider ‘resistance to longitudinally applied, normal stresses, must be considered. The situation can arise where the longitudinal stress limits pressurization-containment and failure will be transverse to the long-axis of the pipeline. For example, let’s superglue two sections of pipe together and then pressure-test the assembly. What will happen? Before the hoop stress can do much of anything, the longitudinal component of stress will overcome the glue and the two halves will go flying! See Fig. 14 (marked 17!), below.

Fig. 14/17 is meant to show the variation in failure mode when a circumferential weld of low quality is present. It is possible for the pipeline to fail at a moderately low pressurization.

In the media, it has been reported that there are both longitudinal and circumferential welds in the failed pipeline. The circumferential weld(s) have to be there as the pipeline is created by welding pipe-sections together, sections that can be mounted and transported on a large truck. Properly welded, the weld is as strong or stronger than the base material of the pipe. Liquid pressurization tests would be done post-weld to proof test the pipeline and to assure that there are no through-leaks. You don’t want your natural gas pipeline to leak!

How are pipe-segments manufactured? They can be extruded from a thicker-wall cylinder and contain only junction-welds to connect them end-to-end; or they could be rolled from sheet steel and welded along the longitudinal seam to seal them. There is some indication that the San Bruno line contained both longitudinal and circumferential welds. The best practices in the 50’s probably called for seamless (or extruded pipe); but it is possible that some sections may have been seam-welded. It sounds like all the welds in this one, critical section of pipeline from San Bruno were improper (ie, very poor welding practices); but we await the factual report. I want to reiterate that welds, properly done, can match or exceed the base-metal. One way or another, I think ‘welding’ is getting bad press from this disaster and that is unfortunate.

Having said that, I share with you an e-mail I sent a few friends when the first photo of that critical section of pipe appeared in the media:

From: Patrick Pizzo <ppizzo@email.sjsu.edu>

Date: September 12, 2010 10:26:16 PM PDT

To: DEREK

Subject: Your video II

So in the pit, there are the two ends that define the 'extent' of the pipeline failure. One end (the first in the video) has the axial failure, just as if the pipe were cut in two. This is where the weld would be, joining two sections of pipe. The helicopter keeps circling the crater, the houses destroyed are shown; and then we are back to the pit, the other end of the pit. At this end, the break is not axial... this buried end has a 'pointed' fracture (fracture taking off in a couple of directions as the pipe splits. The center section is missing and is the section in the street some 160 feet from the crater.

The axial end of the pipe-piece that came out of the crater is brittle fracture, probably within the HAZ (heat-affected-zone) of the weld. The weld does look brittle. Something about the weld was not right... or hydrogen got into the weld with service (this electrochemistry I talked about before) and the weld became embrittled.

Recall that the people first heard the freight train, and then the explosion.

I think that the weld broke first and precipitated events. Then gas rushed out of the pipe at 300 psi like a gas-jet. This was the roaring sound that people reported. Then, as the dirt was saturated with the gas escaping, a spark set the 'leak' off, and the rip down the length of pipe occurred and the section was thrown out of the ditch! I may be wrong but I think the weld broke first and then the flaying of the pipe occurred. It is possible though that the sequence was reversed; ie, the pipe split open; and the force of the split pulled on the weld, breaking it, and the piece came flying out. Need more data!@! gp

I share this because problems associated with the circumferential weld were apparent, to a metallurgist, from day one. It was not until later that we learned that the split in the pipe at the other end of this critical section was apparently along a longitudinal weld. Since then too, we learn of the possible presence of cracks in the weld(s); and the pressurization variations have been noted. What does all of this mean?

Fig. 15 shows a schematic representation of a crack in the wall of the steel membrane (ie, in the wall of the pipeline). Although there are many modes that can cause initiation and growth of cracks into metal, fatigue is the most likely culprit. Fatigue occurs when one repetitively loads and unloads metal. This relates to the exercise of breaking a steel paperclip by bending the clip back and forth until it breaks. The first few repeat bends harden the material by movement of line-defects until the material can sustain no further bending. Then cracks form and grow into the hardened mess. One crack becomes king; and soon the repeat bending motion fails the steel paperclip. The same thin happens in a pipeline. From where does the repeat loading come?

When a system is ‘down’ for repair, it is unloaded. This is one type of repeat cycle. Then, one must realize that the natural gas is not maintained at an exact pressure. It is cycled between a high and a low pressure during pumping and this too is cyclic in nature. The steel material sees many, many cycles of this type of pumping motion. Then, the utility alters the amount of gas flowing through the pipeline to meet customer demand; and more flow means one must pump at a greater pressure. So the service-pressurization level is not constant. The pipeline steel is subjected to cyclic pressurization loading and could succumb to fatigue at some point.

I do notice that the NTSB hearings from the metallurgical team will present some Scanning Electron Microscopic SEM data. This could be most anything; but SEM is a powerful tool to detect cyclic-driven crack advance. It would not surprise me to see that fatigue, at some level, was occurring in and around weld defects in the San Bruno pipeline disaster. We shall see.

Well I hope this exercise has given you, the general public, a better idea of the stresses that act upon pressurized natural gas pipelines; and of some of the considerations that will undoubtedly play a role in the NTSB analysis and interim report. Remember, this is just conjecture and in no way represents an analysis of the catastrophic events of September 9, 2010. We need to see real data!

Respectively,

Patrick P. Pizzo

Professor Emeritus, SJSU-Chemical and Materials Engineering