Lab vs. Reality - Part 2

In our first "Lab vs. Reality" article, we discussed that lab tests have to reflect the real world, or they are meaningless. Many of you asked questions about the details – what specific changes to the test protocol did we make, and why?

Lab vs. Reality - Part 2


Lab v Reality 2 Fig 1

Figure 1 Cervélo bike with strain gauge instrumentation for recording reality 

We published an article in the “Thinking & Processes” section of our website called “Lab vs. Reality(click here to read part 1). In it, we discussed that lab tests have to reflect the real world, or they are meaningless. We said of the two industry standard tests: the bottom bracket stiffness test matches reality quite well, but the torsional stiffness test didn’t. In response, we developed the Cervélo torsional test to better match reality, and all Cervélo frames are now developed following that test protocol. Many of you asked questions about the details – how do we know, precisely, that it’s true and not just because we said so? And what specific changes to the test protocol did we make?

Here in part two of “Lab vs. Reality”, we’ll take a closer look at reality, and at two lab test protocols: “Standard” and “Cervélo”. We’ll see strain data gathered from a bike ridden in reality, and also from the same bike tested both ways in the lab. Do they match? We’ll get into the engineering details to compare exactly how well the two lab protocols match reality or not, and what the result means to the engineer and most importantly, to the rider.

Reality, recorded

Lab v Reality 2 Fig 2

Figure 2: Reality, recorded. This is a time trace from riding the strain gauge bike in the real world.

In the figure above, you can see that the strain from seven strain gauges varies with time. The seven strain gauges alternate between tension (positive strain, shown above the horizontal “zero strain” axis) and compression (negative strain, shown below the horizontal “zero strain” axis).

This repeating positive and negative strain pattern reflects the more or less symmetrical steering inputs in the slalom part of the test course: repeatedly swerving left, right, left, etc. Naturally, in between left and right excursions of the frame, the strain goes through zero. To simulate the strain we’re interested in, we need to look closely at the peaks; zero is uninteresting for this purpose.

The magnitudes of these strains reflect this rider on this bike; if the rider were heavier or swerved harder, all the strains would increase in proportion; if the bike were more flexible or stiffer, again all the strains would increase or decrease in proportion as well. The proportion among the strains is what’s important (more on this later). The point isn’t to design to a sweet spot, nor to minimize (nor maximize) these strains; it’s simply to record the response of this real bike to this real ride: a record of reality.

Reality, frozen

Because the recorded data from the real ride is dynamic, or constantly changing, and the lab test is quasi-static, or virtually unchanging, in order to make a comparison we need to extract from the dynamic data the typical maximum and minimum strain values at the apex of each swerve on the road. Basically, what are the average values at the peaks (or valleys) of the dynamic data? Let’s take a look at the eight individual moments when the blue data series, the top tube’s forward strain gauge, hits its most negative value, namely, the eight low points in its curve.

Lab v Reality 2 Fig 3

Figure 3: Eight stars indicate the eight points of most negative strain in the top tube’s forward strain gauge.

These eight points represent the eight moments when the top tube saw the most compressive stress, indicated by the most negative values in each of the eight swerves. They vary a bit from each other, because the human rider in real life varies a bit, but on average they are all fairly close to about -375 microstrain. We could have chosen the most positive values, but they’re more or less symmetrical, so it doesn’t matter.

In the same way, we examine each of the seven strain gauges at the same eight moments in time, and see that the average strain in each gauge is as follows:

Lab v Reality 2 Fig 4

Figure 4: Average strain in each of seven gauges at the moment of peak swerve.

In the figure above, the bar chart (left) quantifies the strain value in each of the gauges on the bike, extracting from the dynamic data (on the right) the typical maximum and minimum strain values at the apex of each swerve on the road. This bar chart lists the data from REALITY that the LAB test must match. To make the graphic easier to follow, each bar in the bar chart is coloured to correspond to the each data series in the dynamic data: blue for the top tube’s forward strain gauge, pink for the top tube’s middle strain gauge, etc.

Lab data: Standard protocol

After recording reality, we put the same frame onto the test bench in the lab. (If we had used a different frame, the strain distribution would also be different, since a change in frame design redistributes the strain among the various strain gauge sites as well.)

 Lab v Reality 2 Fig 5

Figure 5: The “standard” lab test for measuring torsional stiffness.  On the left, a schematic showing the boundary conditions of the test. On the right, a photo showing the same frame used in the road test being tested in our Project California lab. A torsional load is applied using the bar through the head tube.

We disassembled the bike, and put the frame (still with the same strain gauges) through the standard torsion test protocol in our Project California lab. This “standard” torsion test is used worldwide by most bike companies and many bike magazines, and at first, by us, too. But does it reflect reality?

Lab v Reality 2 Fig 6

Figure 6: Typical record of strain from the lab. (Standard torsion test protocol)

The figure above is a read out from the lab test. This data shows how the frame reacts to the standard torsion test protocol. As above, each triangle is coloured to correspond to the dynamic data: blue for the top tube’s forward gauge, pink for the top tube’s middle gauge, etc.

The lab test applies a quasi-static load to the frame, and the strain gauges measure the frame’s resulting strain at each spot. The hope is that the lab test is a good one, and it simulates reality and thus can reliably predict performance when riding. So how do you know if the test is a good reflection of real riding or not?

There are two possible cases:

  1. If the load in the lab is applied in the same way the rider loads the frame on the road, then the strains in various tubes will be in the same proportion in the lab as they are on the road. In this case, the lab matches reality and the test is good.
  2. If the load in the lab is applied in a different way than the rider loads the frame in reality, then the strains in the various tubes will NOT be in the same proportion to each other. In this case the lab test doesn’t match reality and the test is meaningless.

Let’s compare the standard torsion lab protocol to the strain pattern recorded in reality and see how closely the proportions match.

Lab v Reality 2 Fig 7

Figure 7: Standard lab test versus reality. This is a combination of two figures above. On the left, the bar chart showing the peak strain values, derived from the middle part, the dynamic record of reality. Added on the right is the lab data from the standard torsional test protocol. Note that the magnitude of strain in reality is about triple that of the lab test (-125 x 3 = -375). Tripling the lab test’s load would triple all the strains, but the proportion is what’s most important.

In the figure above, each shaded horizontal bar represents one strain gauge and it should align with each of the three graphs: most of the peaks of the dynamic data (centre), the average peak strain value we extracted from that data (left), and the lab data (right). The right end of each bar terminates in a colour coded circle. As before, each circle is coloured to correspond to the other data: blue for the top tube’s forward strain gauge, pink for the top tube’s middle strain gauge, etc.

It’s important to remember it’s the relative proportion of strain we’re interested in, not the magnitude. In reality, the strain changes with the rider’s force on the bike: more force, more strain; less force, less strain. In the lab, we apply the load by hanging weights: more weight, more strain; less weight, less strain. Since the frame remains below yield during riding, strain increases linearly with load (Hooke’s law), which means we can change the applied load to reach any magnitude of strain. Tripling the load, to match the magnitudes we recorded during this ride, would mean triple the work done by our lab tech. So reducing the load is a practical accommodation to keep our lab tech healthy, and we can still compare proportions to judge a match or not.

How does the “standard” torsion test lab protocol compare to reality?

If the standard lab test protocol strains the frame in the same proportions a rider does in reality, then the triangles should match the circles: each coloured triangle, which represents one strain gauge’s reaction to the standard lab test protocol, should appear beneath the circle of the corresponding colour, which represents reality. This is indeed the case with the blue circle and triangle: In the forward part of the top tube at least, the strain induced in the standard lab test matches reality. There are a few other regions of the frame that are pretty close: the down tube upper strain gauge (orange) and the head tube (red). But four regions show significantly different strain: the middle of the top tube (pink), the aft part of the top tube (light green), the lower down tube (dark green) and the seat tube (dark red) aren’t as close as they could be. There is room for improvement.

The Cervélo approach

Being engineers, we made the obvious corrections to the standard lab test set up: we changed the boundary conditions of the lab test to closer match those in reality. The new Cervélo lab test produced strains virtually identical to reality. 

Lab v Reality 2 Fig 8

Figure 8: Schematics of two torsion tests: Standard lab test (left) and Cervélo lab test (right). As before, in both cases a torsional load is applied using the bar through the head tube.

Generally, we changed the way the frame is restrained to better match the way the rider and road constrain the bike. Specifically, we:

  1. Pivot the frame at the rear wheel’s road contact point (instead of fixing the rear drop outs)
  2. Pivot the head tube at the lower edge, near the fork bearing (instead of the middle)
  3. Fix the top of the seat post to simulate the inertia of the rider’s body weight (ignored in the standard test)

Combined, these three changes more closely simulate the constraints the bike sees when ridden on the road, so produce a strain distribution that is virtually identical.

Include the rider

Of the three changes, the first two are basically moving the position of the standard test’s two existing constraints to better match reality. But it’s interesting to note that the one new constraint we added, the seat post, is completely missing from the standard protocol. Supporting the seat post represents the inertial anchor of the rider’s body weight on the saddle, which stabilizes the top of the seat tube during lateral changes in direction, as when cornering. 

Lab v Reality 2 Alex Howes

 Including the rider and their effect on the bike is fundamental to Cervélo’s system engineering approach, and is a repeating theme in all our engineering work. For example, the wind tunnel testing and aero engineering we do includes the rider, in the form of the world’s first accurate wind tunnel mannequin, Foam Dave. Likewise the dynamic testing and engineering structural modelling we do also include the effect of the rider. This system engineering approach goes all the way back to the beginning of Cervélo’s history, when we designed Cervélo TT/tri geometry around a steeper riding position to reduce aero drag of the rider. As demonstrated in this article, including the rider in stiffness testing is also important to match reality.

Lab data: Cervélo protocol

We’ve seen the limitations of the existing standard test protocol, and how the Cervélo test protocol improves the boundary conditions. Have a look at the result:

Lab v Reality 2 Fig 9

Figure 9: Record of strain from the lab following the Cervélo torsion test protocol.

The figure above is similar to the previous one, only this is produced using the new Cervélo lab test protocol. As before, each triangle is coloured to correspond to the dynamic data recorded above in reality: blue for the top tube’s forward strain gauge, pink for the top tube’s middle strain gauge, etc. 

 Lab v Reality 2 Fig 10

Figure 10: As you can see, the triangles match to the circles much better.

As before, the strain data from the Cervélo lab test is compared to the static and dynamic data from reality. Now compare the position of each triangle to the circle of the same colour. There’s a much better match now; the Cervélo lab test matches reality better than the standard lab test does.

What’s the difference?

How does the strain distribution compare between the Cervélo torsion test and the standard  torsion test?

 Lab v Reality 2 Fig 11

Figure 11: Strain data from the Cervélo protocol (left) and standard protocol (right). Horizontal bars trace each strain value on the left to the corresponding column on the right, where the differences are indicated by vertical arrows.

Compared to the way the standard test strained the frame, we found three areas of the frame that responded with significant differences in reality:

  1. The middle of the top tube sees less than half the stress imposed by the standard protocol (-27 vs. -67 microstrain)
  2. The back of the top tube sees about double (137 vs. 65 microstrain)
  3. The bottom of the seat tube sees about double (-117 vs. -57 microstrain)

(The down tube is also stressed less in the standard test, but that part of the frame is highly stressed in the power transfer load case and is engineered to be stiff for that, no matter what this load case reveals.)

How does this affect the engineers?

These differences are important when engineering bikes: lab tests are the basis for structural analysis tools like FEA. They’re also useful for measuring differences in performance between different frames, and changes in performance due to engineering changes in prototypes. In practice, frames rank differently – the stiffer frame in one test isn’t necessarily the stiffer frame in the other. Both FEA and lab tests are tools engineers use to optimize the bike. But if the bike is optimized to the wrong lab test, then it’s not optimized for reality.

How does this affect the rider?

We were testing the wrong way! To a design engineer, twice (or half) the previous stresses implies halving (or doubling) the wall thickness, or equivalent changes in fibre type or lay up. Now we know the Cervélo lab test protocol closely simulates reality, so our FEA tools and bench testing of Cervélos and other brands truly reflect what the athlete puts into the bike. As a result, when we optimize the bike, we’re confident the changes we’ve made are reflected in real riding on the road.

As mentioned above, the bottom bracket stiffness test already matched reality, so power transfer can be tuned. Now that the torsional stiffness test also matches, cornering and turning performance can also be tuned. You’ll feel this when initiating a turn, as a more solid, responsive feeling while you set up the bike to lean and change direction. Your Cervélo is lighter, stiffer and handles better because we need to put the fibers only where they are really doing the work.


Cervélo’s system engineering approach leads us to investigate every aspect of bike and rider performance, such as taking the time to investigate whether standard lab tests match reality or not. As a result, when we optimize a Cervélo, we know we’ve engineered a bike that will make a difference on the road. Which may be why Cervélo bikes keep winning races, reviews and awards.

Which Cervélos benefit?

All Cervélo models that have been engineered following the development of the Cervélo Torsion Test as part of our Project California mission benefit from this new knowledge, starting with the R5ca in 2010.

Comments (1)
  • With a Cervelo account, you can comment on articles and contribute to the conversation in our community.
  • sausskross

    That is what I call interesting insights to engineering. Even if my pre california soloist carbon is only effected by older lab results. The rocket still thrills me on the road oSo >>

    Report this comment.
Looking for Help?
Find a Retailer
Find a Retailer

Our professional staff can help determine the right bike and fit for you.

Find a Retailer
Visit Support
Get Support

Have a question for us? Browse our FAQ's or ask an expert.

Visit Support