Summary: Uniform heating does not change the angle of a bracket. It changes the size of the bracket, and size cancels out of an angle, so a bracket that grows evenly in the sun returns the same reading. What changes an angle is a temperature difference through the thickness, or a difference in expansion coefficient between two materials in the same load path. The first bends a cantilever by alpha times delta T times L divided by t, so the controlling quantity is the ratio of cantilever length to thickness rather than the absolute size. On steel that ratio converts one degree Celsius of gradient into 0.0069 deg at a ratio of 10 and 0.0137 deg at a ratio of 20, and both exceed the 0.005 deg accuracy of the sensor being carried. Restraint decides whether the result is a false reading or a hidden stress.
1. The Suspicion Is Right but the Mechanism Is Wrong
When a monitoring engineer starts to doubt the mounting, the first theory is usually that the bracket expands in summer and contracts in winter, so the sensor must be reading the bracket instead of the structure. The suspicion that the hardware matters is correct and worth acting on. The mechanism behind it is not, and correcting the mechanism is what turns a vague worry into a design rule.
Consider a plain steel angle bracket, one arm bolted to a wall and the other carrying a sensor, sitting outdoors through a twenty degree Celsius swing. Uniform heating makes both arms longer. The angle between the two arms does not change, because every dimension of the part, including any length you could use to define an angle, grows in the same proportion. An angle is a ratio of two lengths, so the expansion cancels, and the same argument holds for a flat plate, a pair of spacers, or a symmetric pedestal. A bracket that expands uniformly is geometrically similar to what it was, only larger, and it holds exactly the same attitude.
That is also why a bracket that is simply bolted down and left alone tends to be reliable in service while the same bracket under the same sun catches people out in analysis. The uniform part of the temperature change is harmless. The part of the temperature change that is not uniform is where the error lives, and that part is much smaller than the number on a weather report, which is why it is so often overlooked.
2. What Actually Changes an Angle
Only two mechanisms can rotate a bracket that nobody has loaded. The first is a temperature difference across the thickness of a member, which is what happens when one face of a bracket is in sun and the other is in its own shadow, or when a bracket is bolted to a large concrete mass at a stable temperature while its exposed arm tracks the air. The warmer face expands more than the cooler face, so the member cannot stay straight. It curves.
The second is a difference in expansion coefficient between two materials that share a load path. A sensor with an aluminium housing bolted to a steel bracket, or a steel bracket welded to a stainless base plate, has two materials that want to change length by different amounts when the assembly changes temperature. If the geometry allows them to slide or bend, the assembly curves. If it does not, the same mismatch becomes stress.
The two mechanisms are the same mechanism. Take the expansion difference between the outer surfaces of a member and divide it by the distance between those surfaces, and you have a curvature. Multiply curvature by the length over which it is free to act, and you have a rotation. Everything else in this article follows from those three quantities and the fact that one of them sits in a denominator.
3. The Formula, and Why the Ratio Is the Whole Story
For a member of thickness t and cantilever length L, free at one end and heated so that one face is warmer than the other by delta T, the rotation at the free end is:
theta = alpha x delta T x L / t(theta in radians, alpha the coefficient of linear thermal expansion, delta T the through thickness temperature difference, L the length over which bending is free to accumulate, t the member thickness)
This result is standard structural mechanics rather than a vendor claim. It appears in the classic mechanics of materials literature, where a cantilever with one face at T1 and the other at T2 develops a slope at its free end of alpha times the temperature difference times L divided by the depth, and it appears in the same form in MIT’s mechanics and materials course material, where the thermal curvature is written as two alpha delta T prime divided by h, and in optomechanical design references, where the rotation of a flexure under a thermal gradient is written as alpha L delta T divided by h. These are three independent presentations of the same relationship, which is worth noting because the equation is the entire basis for the design rules that follow.
Rearrange it once and the practical form appears:
theta / delta T = alpha x (L / t)(the rotation produced per degree Celsius of through thickness gradient)
The absolute length of the bracket has disappeared. So has the absolute thickness. The only geometry that matters is the ratio of the two, which is the reason a small bracket can be worse than a large one and a heavy bracket is not automatically a stable one. A part that is 300 mm long and 3 mm thick has an L over t ratio of 100. A part that is 60 mm long and 20 mm thick has a ratio of 3. The first is a spring, the second is a block, and the fact that the second is a fifth of the size is irrelevant.
One more consequence deserves to be stated before the numbers. A gradient across a thick member produces the same rotation as the same gradient across a thin member of the same L over t ratio, so no section escapes this by being massive. The only escapes are to make the ratio small, to make the gradient small, or to make the coefficient small. Those three options are the rest of this article.
4. What the Ratio Means on a Real Bracket
The arithmetic becomes uncomfortable quickly, and it is worth doing in the open because the result is the reason this subject is not a footnote in an installation guide. Take ordinary carbon steel at 12 parts per million per degree Celsius, and compare the resulting rotation with the 0.005 deg typical accuracy published for the geotechnical and structural inclinometers in this series.
| Cantilever length / thickness (L / t) | Rotation per degree Celsius of gradient | Multiples of a 0.005 deg sensor accuracy | Through thickness gradient that consumes the whole accuracy budget |
| 5 | 0.0034 deg | 0.7 | 1.45 degrees Celsius |
| 10 | 0.0069 deg | 1.4 | 0.73 degrees Celsius |
| 20 | 0.0137 deg | 2.8 | 0.36 degrees Celsius |
| 33 | 0.0227 deg | 4.5 | 0.22 degrees Celsius |
| 50 | 0.0344 deg | 6.9 | 0.15 degrees Celsius |
Read the last column as the answer to a design question. A steel bracket with an L over t ratio of 20, which is unremarkable for a field fabricated support, has consumed the entire accuracy of a 0.005 deg sensor once the two faces of the bracket differ in temperature by about a third of a degree Celsius. Reach a ratio of 50, which is what a long thin arm produces, and a sixth of a degree is enough.
Those gradients are not exotic. The Eurocode thermal actions document, EN 1991-1-5, treats a vertical temperature difference across a bridge deck as a design load rather than a curiosity, and publishes an upper bound linear component for a steel deck of 18 degrees Celsius with the top warmer than the bottom, and 13 degrees Celsius in the reverse case, with a nonlinear profile that reaches a difference of up to 24 degrees Celsius a tenth of a metre below the top surface. For cylindrical structures such as silos and storage tanks the same document recommends a circumference step of 15 degrees Celsius, with one quarter of the circumference warmer than the rest, which is exactly the shape of loading that a bracket on the shaded and sunlit sides of a tank experiences. Standards do not quantify a temperature difference this carefully if it is negligible.
The geometric consequence is worth stating plainly, because the instinct is to reach for a thicker plate. Using a fixed 0.005 deg budget, and the same carbon steel, a cantilever 100 mm long would need a thickness of 13.8 mm to stay inside it with a one degree Celsius gradient, 27.5 mm for two degrees, and 68.8 mm for five degrees. A cantilever 200 mm long at five degrees Celsius would need 137.5 mm of steel. The last two numbers are not design options. What they show is that above a certain cantilever length, no practical thickness fixes the problem, because the thickness required grows in direct proportion to the length and quickly passes the point where the part can be built and fixed to a structure. The fix has to come from shortening the arm or from controlling the gradient, not from buying more steel.
5. Restraint Decides Whether You Get a Bad Reading or a Hidden Stress
Everything above assumed a cantilever, free to bend. The opposite case is instructive in a way that is easy to miss, and it is where the practical warnings about brackets should start.
A member that is fully restrained at both ends cannot rotate at either end, so a thermal gradient through it cannot express itself as an angle. The classical treatment is explicit about the surprising result: a beam built in at both ends and subjected to a temperature gradient will remain horizontal, because the moments developed at the supports exactly cancel the thermal curvature, and what the member carries instead is a residual stress. The bending is gone and the stress has arrived.
The magnitude of that stress is easy to estimate. For a member fully restrained against expansion, the stress is the product of Young’s modulus, the expansion coefficient and the temperature change. For carbon steel at 20 degrees Celsius of change this is 48 megapascals, and for austenitic stainless steel, whose expansion coefficient is higher and whose modulus is only slightly lower, it is 67 megapascals. Aluminium comes out lower, at 32 megapascals, because its much lower modulus roughly offsets its higher coefficient, and a low expansion alloy such as Invar comes out at 3 megapascals.
Forty eight megapascals is below the yield strength of structural steel, which is why the fixed bracket does not visibly fail and why nobody notices. It is not below the stress at which bolted joints lose preload and at which interfaces creep, and those are slow effects. A bracket that is rigidly fixed at both ends and cycling through a daily thermal gradient may hold its angle for a year and then begin a slow monotonic drift as the joint relaxes and the contact between the mating surfaces changes. The reading looks like a structure that has started moving.
Put the two cases side by side, because the pairing is the most useful single idea in this article.
| Restraint | What the thermal gradient produces | How it appears in the data |
| Free at one end, as in a cantilever arm or a pedestal with an unsupported top | Real rotation of the sensor, proportional to L over t | A daily cycle that follows the sun, both axes moving together, returning to the same value overnight |
| Restrained at both ends, as in a plate bolted on two opposite edges or a bracket welded along both sides | No rotation, but a residual stress of the order of E alpha delta T | A stable reading for months, followed by a slow monotonic drift as joints relax and interfaces creep |
The second row is the more dangerous of the two, because a daily cycle is obviously environmental and a slow drift is not. A slow drift looks like the thing the monitoring programme was installed to detect. Before treating a new monotonic trend as structural movement, it is worth checking whether the affected sensor is on a bracket that is restrained at more than one point.
6. Material Choice, and What It Actually Buys
Since the rotation is proportional to the expansion coefficient, the material is a design variable. The coefficients below are handbook values near room temperature for common engineering materials, and they are worth reading next to each other rather than one at a time, because two of the entries contradict what most people assume.
| Material | Coefficient of linear expansion, per degree Celsius | Rotation at L over t of 20 with a 5 degree Celsius gradient | L over t x delta T that consumes a 0.005 deg budget |
| Invar, 36 percent nickel steel | 1.2 ppm per degree Celsius | 0.0069 deg | 72.7 degrees Celsius |
| Titanium, grade 2 | 8.6 ppm per degree Celsius | 0.0493 deg | 10.1 degrees Celsius |
| Cast iron, gray | 10.5 ppm per degree Celsius | 0.0602 deg | 8.3 degrees Celsius |
| Carbon steel | 12.0 ppm per degree Celsius | 0.0688 deg | 7.3 degrees Celsius |
| Stainless steel, 304 | 17.3 ppm per degree Celsius | 0.0991 deg | 5.0 degrees Celsius |
| Aluminium, alloys | 23.1 ppm per degree Celsius | 0.1324 deg | 3.8 degrees Celsius |
| Zinc, as in a galvanised coating or a die cast part | 30.0 ppm per degree Celsius | 0.1719 deg | 2.9 degrees Celsius |
The last column is the practical form of the rule. Multiply the L over t ratio of the bracket by the through thickness gradient you expect, in degrees Celsius, and compare the product with the number in that column for the material you have chosen. If the product is larger, the bracket is the limiting element in your measurement, not the sensor. The check takes about ten seconds and needs no software.
Two entries deserve comment. The first is stainless steel. It is the default choice for outdoor monitoring hardware, for good reasons of corrosion resistance and appearance, and it is forty four percent worse than plain carbon steel on thermal rotation and thirty nine percent worse on the residual stress of a restrained joint, because type 304 expands at 17.3 rather than 12.0 and its modulus is nearly as high. A galvanised or painted carbon steel bracket of identical geometry is the thermally better part. That is not an argument against stainless steel in salt air, where corrosion protection has to win, but it is an argument against reaching for stainless steel by reflex where it is not needed.
The second is aluminium, which is the most common bracket material in the field because it is light, easy to machine and takes an anodised finish. Its coefficient is nearly double that of carbon steel, so an identical aluminium bracket rotates about ninety percent more than its steel equivalent for the same gradient. Aluminium also matters for a second reason: many sensor housings are aluminium, so the sensor to bracket interface is a dissimilar material pair with a coefficient difference of eleven parts per million per degree, which is the second mechanism from section two. Matching the bracket material to the housing material removes that term, and where matching is not practical, the interface should be designed so that the mismatch is taken as stress in a short, stiff joint rather than as curvature in a long, slim one.
A low expansion alloy such as Invar is the textbook answer and it delivers an enormous margin, nineteen times better than aluminium and ten times better than steel on rotation. It is also expensive, difficult to machine and difficult to weld, so its realistic use in monitoring is limited to small precision spacers and reference components rather than to structural brackets. It earns its place in the table as a reminder of how far the coefficient can be pushed if a specific application justifies it.
7. Six Design Rules That Follow
The rules below are the working consequences of the arithmetic, in the order in which they are worth applying. The first two are geometry and they are free, which is why they come first.
Shorten the cantilever before anything else. L is the only geometric quantity that scales the error linearly and without limit, and it is also the quantity that field fabrication increases by accident, because a longer arm is easier to reach with a spanner and easier to align by eye. Put the sensor as close to the fixing plane as the installation allows.
Increase the section thickness, not the plan area. Because the rotation depends on L over t, adding material to the sides of a bracket while leaving its thickness and its cantilever length unchanged changes nothing thermal. The step that helps is making the load path deeper in the direction of the thermal gradient, which usually means a box section, a machined block or a thick spacer rather than a bent sheet.
Keep the shape symmetrical. ISO 5348, the standard for the mechanical mounting of transducers, states the requirement in general terms: the mounting should introduce minimum distorting motions of its own, and simple symmetrical mountings are best. The thermal reading of that sentence is that a symmetrical part has a symmetrical expansion, so the uniform part of the temperature change produces no rotation at all, and the asymmetry that remains is only the gradient. An L shaped bracket fabricated from unequal arms, which is the most common shape in the field, is the design that gives up this property for convenience.
Shade to remove the asymmetry, not to reduce the temperature. A sunshade works because it makes the temperature difference across the member smaller, not because it makes the member cooler. Cooling a bracket uniformly does nothing, since the uniform part of the change cancels out of the angle. So the effective shade is the one that stops one face being heated relative to the other, which usually means a shade that is wider than the bracket and open on the shaded side rather than an enclosure that traps air and equalises nothing. A shade is itself a cantilever and adds wind area, which has to be counted against its benefit.
Choose the material deliberately, and match it to the housing. Carbon steel outperforms stainless steel thermally, and both outperform aluminium by a wide margin. Where the sensor housing and the bracket are the same material, the differential expansion term at the interface disappears. Where they cannot be the same, keep the dissimilar joint short and stiff and keep the long, slim part of the assembly in a single material.
Avoid adjustable hardware unless it is needed, and lock it when it is. Every degree of freedom in a mount is a potential source of drift. A slotted hole, a spherical joint or a threaded adjuster with a locknut is a mechanism, and mechanisms have clearance, and clearance moves under vibration. Where adjustment is genuinely required to bring the sensor into range or to align an axis, treat the adjustment as a temporary setting and add a positive mechanical lock after it is made, rather than relying on friction alone for the life of the installation.
8. What a Bracket Error Looks Like in the Data
Most of the time the bracket cannot be redesigned, because it is already in place. What can still be done is to recognise the signature in the readings, and the signature of a thermal bracket error is quite specific, which is fortunate.
It repeats daily and it is reversible. A bracket responding to solar gain produces a cycle with a period of one day, rising through the morning, peaking in the afternoon and returning to the same value overnight. The defining test is reversibility: the value it returns to is the same value, day after day, without a residual. Structural damage does not typically behave this way, because damage accumulates.
Both axes move together. A single bracket has one gradient direction and one L over t ratio, so a dual axis sensor on that bracket tends to show correlated movement on both channels. A structural mechanism such as foundation rotation or a spreading slope usually loads one axis substantially more than the other. Correlation between the two axes at the one day period is a strong indicator that the driver is the mount rather than the structure.
It depends on orientation and on season. Two identical brackets on the same structure, one facing east and one facing west, will peak at different times of day. Two brackets on the north and south faces of a tank will differ in amplitude across the year. A structural trend does not know which way a bracket faces.
A temperature channel separates it. Where a temperature measurement is available at the sensor, the reversible component can be regressed against it and subtracted, which is the same logic the dam monitoring literature applies to reservoir level and season. The residue is what remains after the reversible effects have been accounted for, and on a bracket limited installation the residue is usually small and much easier to interpret than the raw reading. This is a processing solution rather than a hardware solution, and it should be described as such: it removes the reversible part of a bracket error, not the part that has already caused a joint to relax.
The practical use of this section is triage. A reading that fails all four tests deserves investigation as a possible structural change, and a reading that passes all four is more likely to be hardware and environment, which is a different investigation with a different remedy.
9. Where This Stops, and What It Does Not Replace
Four limits are worth stating, because bracket design is a subject that invites overreach in both directions, from dismissing it entirely to treating it as the explanation for every unexplained reading.
First, this is general structural mechanics, not a property of any particular sensor. The formula contains no manufacturer and no product, and every metal bracket on every structure obeys it. A more accurate sensor does not reduce a bracket error, it only makes it a larger fraction of the total, which is the counterintuitive consequence of buying precision and mounting the result on a compliant arm.
Second, a bracket is a compromise rather than a choice. The mounting instructions supplied with a digital inclinometer are explicit that the mounting surface of the sensor should sit completely against the mounting surface of the object being measured, with no angle between them, and that the sensor axis should be parallel to the axis of the surface being measured. A bracket inserted between the two violates the spirit of that instruction, and it should be understood as a necessary compromise, adopted when the structure surface is inaccessible, not flat enough, too hot or too corroded to fix to directly, or where the sensor needs protection. Where direct mounting is possible, it is thermally and mechanically superior, and it is worth saying so rather than designing a bracket out of habit.
Third, bracket design sits in a different layer from installation quality. The rules about flatness, cleanliness, correct axis alignment and adequate tightening torque are about executing a joint correctly, and a well designed bracket installed badly still produces bad data. The two subjects are complementary, and neither substitutes for the other.
Fourth, no general article can settle a specific installation. The gradient a bracket will actually see depends on its orientation, its finish, its shading, the local climate and the thermal mass of the structure it is fixed to, and the acceptable bracket error depends on the smallest movement the monitoring programme needs to resolve. Those two numbers should be estimated for the specific case and recorded with the installation, so that a later analysis knows what the hardware is capable of, and so that any monitoring specification is confirmed against the datasheet supplied with the unit and against the requirements of the structure’s own engineering authority.
Frequently Asked Questions
- Can I fix a bracket thermal error by fitting a thicker mounting plate? Only if the cantilever length stays the same and the added thickness is in the direction of the thermal gradient, because the rotation depends on the ratio of length to thickness rather than on thickness alone. Doubling the thickness halves the rotation, which is a real gain, but doubling the cantilever length as well cancels it exactly. In practice a thick plate bolted flat against a structure is a good solution because its free length is short, and a thick plate on a long standoff arm is not, because the arm sets the L over t ratio and the thickness of the plate does nothing for it.
- Is a stainless steel bracket worse than a mild steel one? Yes, on thermal grounds, and the difference is not marginal. Type 304 stainless steel expands at 17.3 parts per million per degree Celsius against 12.0 for carbon steel, so an identical bracket produces about forty four percent more rotation for the same through thickness gradient, and about thirty nine percent more residual stress when both ends are restrained, because its modulus is nearly as high as carbon steel’s. Stainless steel is still the right choice in salt air or in chemical exposure, where the corrosion resistance is worth paying for in thermal performance. The mistake is choosing it by default on a structure where a galvanised or painted carbon steel bracket would last just as long.
- Does shading a bracket reduce the error? It reduces the part of the error that comes from the gradient, and it does nothing for the part that comes from uniform temperature change, because uniform change does not produce rotation in the first place. So a shade is effective when the problem is one face of the bracket being heated relative to the other, which is the common outdoor case, and ineffective when the problem is a mismatch between the sensor housing and the bracket, which no shade can touch. A shade also adds wind area and its own cantilever, so it is worth fixing the geometry and the material before adding one.
4. Two sensors on the same structure disagree. Is that the brackets? It is a candidate, and it is worth testing before either sensor is replaced. Put both sensors side by side on a single thick flat plate in the shade, let them settle, and compare. If they agree there, then the in service difference is not the electronics, and the next candidates in order are the two brackets and the flexibility of the structure between the two mounting points. The thermal test is to log both channels against temperature through a full sunny day: if the disagreement is proportional to temperature and returns to zero at night, the brackets are implicated, and if the disagreement is constant and does not reverse, the more likely cause is a genuine difference in what the two mounting points are doing.
TAG: