Summary: A manhole cover is one of the hardest common targets for a tilt sensor, and the reason is not accuracy. The event that matters most, a cover lifted and displaced, produces little or no rotation, so a sensor fixed to the lid can miss it. The angles a cover does produce are small: a five millimetre lift on one edge of a six hundred millimetre cover is 0.48 deg. The traffic signal it has to be separated from is far larger and is real motion, not sensor error, because covers rock on three point bearing under wheel loads. A working design therefore pairs a presence contact with an angle channel, sets the alarm angle well above the noise floor, and treats the radio path and the chamber environment as the decisive installation problems.

 

1. The Failure You Must Catch Is Displacement, Not Tilt

Ask what goes wrong with a manhole cover and the answer is rarely that it has bent too far. It is that the cover is no longer where it should be. Sometimes that is theft, which is often deliberate and sometimes organised. Sometimes it is a cover that has been shifted by traffic, or that a heavy vehicle has lifted and cocked on its frame. In every one of those cases the chamber is left open or partly open in a traffic lane, a footway or a yard, and an open chamber is a fall hazard to anyone walking over it and an impact hazard to anything with wheels.

That single sentence describes a different class of problem from everything else in structural monitoring. A wall, a pier or a dam is a fixed object that accumulates strain and deforms. A manhole cover is a removable lid sitting in a frame, and the failure mode is a loss of position rather than an accumulation of deformation. The sensing question is therefore whether the cover is seated, and at what attitude, rather than how far it has bent.

There is a second difference that is easy to overlook, and it changes how the data has to be read. A cover is allowed to move. Standardisation for manhole tops requires the heavier load classes to be either non rocking or quiet in use, which is an instruction to the manufacturer that a cover in those classes should not move significantly under traffic. The existence of that requirement is itself the evidence: movement under traffic is the normal condition of this component, and the engineering effort goes into suppressing it.

So a monitoring specification for a manhole cover has to start by naming the event, because the same device placed on the same lid will detect some of those events and be blind to others. That is the subject of the next section, and it is not a limitation of one product. It is the arithmetic of the situation.

 

2. Five Motions a Cover Makes, and What a Tilt Sensor Sees

Manhole cover and frame assemblies fail in a small number of recognisable ways, and each one has a distinct physical signature. The descriptions below follow the maintenance literature on cover and frame behaviour, which treats rocking, rattling, shifting and frame movement as separate conditions with separate causes.

Condition	Mechanism	What a sensor fixed to the lid reads
Rocking	The cover bears on only two or three points instead of continuously around its seat, so it tips as it is loaded	A small repeated rotation, a real angle that alternates with each wheel pass
Rattling	A small vertical clearance lets the cover strike the frame under traffic vibration	Mainly an impact event with very little stable angle change
Shifting or walking	The cover moves laterally because of clearance, traffic forces, or inadequate restraint	Often no rotation at all until the cover cocks against the frame edge
Uniform lift	The whole cover rises evenly, as in a straight vertical lift	No angle change at all
Frame movement	The frame itself settles, tilts, rotates or breaks loose from its support	The lid may ride along with the frame and continue to read a flat, stable angle

Read that table as a specification and the difficulty becomes obvious. The two conditions that create the greatest danger, a uniform lift and a settled frame, are exactly the two that an angle measurement cannot see. A cover lifted vertically out of its frame has not rotated, and a cover whose frame has dropped uniformly has rotated only as much as the frame did, which is very little and very slow.

This is not a defect in the sensor. It is a mismatch between the quantity being measured and the quantity that has changed. An angle sensor answers a question about attitude. Removal and uniform settlement are questions about presence and about elevation, and no amount of resolution on an angle channel converts one into the other.

There is a further asymmetry hidden in the same table. Rocking and rattling are loud, frequent and highly visible in the data, and they are almost never the thing the operator cares about. Shifting and frame movement are quiet, slow and low amplitude, and they are exactly what the operator needs to catch. Any monitoring design that simply alarms on the largest signal will fire constantly on the harmless condition and stay silent on the dangerous one.

 

3. The Arithmetic: How Many Degrees These Events Actually Produce

Before choosing a sensor it is worth converting the physical events into angles, because the numbers are much smaller than most specifications assume. Two constants do all of the work. A tilt ratio of one percent is 0.01 radian, which is 0.573 deg; and one degree of tilt is 17.46 millimetres of rise per metre of span.

Take a cover with a six hundred millimetre opening, disturbed by traffic so that one edge rises five millimetres and the other edge stays on its seat. The rise over the span is 5 divided by 600, which is 0.83 percent, and the small angle constant turns that into 0.48 deg. A more serious disturbance, one edge lifted twenty millimetres, is 3.3 percent, or 1.91 deg. A cover that has merely developed one millimetre of play across a four hundred millimetre bearing span is rocking through 0.14 deg.

Now set against those figures the tolerance that the product standard already allows the cover itself. The deflection acceptance criterion for manhole tops limits permanent set after the proof load to one two hundredth of the span for clear spans up to five hundred millimetres, and one three hundredth of the span above that. For a six hundred millimetre cover the permitted permanent deformation is therefore two millimetres, and a cover that has deformed by that much is still a compliant cover rather than a failed one.

That last figure has a consequence worth stating plainly. The surface that the sensor is bolted to is permitted to change shape by millimetres over its service life, and it still passes. A sensor fixed to a lid is therefore not measuring against an absolutely stable reference, because the thing it is mounted on is allowed to move. Sub degree drift in the reading is not necessarily drift in the sensor, and it is not necessarily a defect in the cover either.

Everything the operator cares about on a manhole cover therefore lives below about two degrees, and most of it lives below half a degree. Hold that range in mind while reading the next section, because the number that follows is considerably larger.

 

4. The Noise Floor: Traffic Moves the Cover for Real

The temptation is to treat the signal from passing vehicles as electrical noise that better filtering will remove. On a manhole cover that reading is wrong, and it matters. There are two separate mechanisms and both of them are physical.

The first is genuine rotation. A cover that bears on three points rather than continuously around its seat tips as load crosses it, and it tips back as the load leaves. The angle is the rise divided by the bearing span, so the one millimetre of play described above is 0.14 deg of real rotation, and worn seating or a frame that has settled at one point produces more than a millimetre. This is a real angle of a real object, and it appears in the data at the same magnitude as the events of interest. Heavier load classes are built to reduce it, using anti rocking seat designs and in some cases multi part covers whose sections are deliberately left slightly loose so that they settle into the frame instead of tipping it, but that describes a new cover in good seating condition. A cover in service with a worn seat, a settled frame or debris under one edge will rock.

The second mechanism is in the sensor rather than the object. A tilt sensor derives angle from the direction of gravity, so any horizontal acceleration adds to the measurement and is indistinguishable from a real tilt. The apparent angle is the acceleration divided by g, and at these magnitudes it is close enough to linear.

Horizontal acceleration	Apparent tilt it produces	Multiple of the 0.005 deg accuracy figure
0.001 g	0.057 deg	11 times
0.005 g	0.29 deg	57 times
0.01 g	0.57 deg	115 times
0.05 g	2.86 deg	572 times
0.1 g	5.71 deg	1142 times

Read the first row against the published specifications of the wireless tiltmeter family used for this class of work. Its resolution is 0.001 deg and its accuracy is 0.005 to 0.01 deg. A horizontal acceleration of one thousandth of g, which is a very small disturbance indeed, already produces 0.057 deg of apparent tilt. That is eleven times the accuracy figure, and against the alarm accuracy of 0.01 to 0.1 deg it is more than five times the tight end of the band. In other words the smallest disturbance worth mentioning overwhelms the instrument’s precision specification.

The practical conclusion is not that the sensor is unfit for the job. It is that instantaneous angle readings on a trafficked cover cannot be compared with a sub degree threshold at all, and any design that tries will fail in one of two directions. Set the threshold low and the system alarms on every passing van. Set it high enough to be quiet and the system can no longer see the small angles that a displaced lid produces. The way out is not a better sensor, it is to stop treating a single sample as a measurement.

 

5. Presence and Attitude: Two Quantities, Two Channels

The company’s own published account of its manhole deployment in the Yizhuang district of Beijing describes exactly this conclusion, and it is worth quoting because it was reached in the field rather than in a specification meeting. The covers there were instrumented with a travel switch or a tilt sensor. When a cover is stolen the system sends an alarm immediately, and the manager opens a phone and sees the street position, the signal strength and the technical status of that specific cover on an electronic map. The account records the alarm angle as fifteen degrees, adjustable by installation type, and notes that the same device is used for cable theft protection.

That phrase, a travel switch or a tilt sensor, is not careless writing. The two devices measure two different quantities and they fail in opposite directions. A travel switch is a presence contact. It answers whether the lid is seated in its frame, which is a binary question that a uniform lift changes and a cocked lid may not. A tilt sensor answers at what attitude the lid is resting, which a cocked lid changes and a uniform lift does not. Each one covers the other’s blind spot from the table in section two.

This has a direct procurement consequence, and it is the single most useful sentence in this article. If the brief is theft and open chamber detection, specify the presence input, because an angle channel cannot promise it. If the brief is mechanical condition under traffic and slow settlement of the frame, specify the angle channel, because a presence contact has nothing to say about it. If a specification names only one of the two, it should state which failure it will not catch, and that statement should be written down rather than assumed.

There is also an installation consequence. Which part of the assembly the sensor is fixed to decides which events it can see at all. Fixed to the lid, it follows the lid, which is what you want for lid displacement and lid theft but means it rides along with a settling frame. Fixed to the frame, it follows the ground, which is what you want for settlement but means it may not notice that the lid has been exchanged. One device on one mounting face cannot do both, and that is an assembly fact rather than a product limitation.

 

6. Choosing the Alarm Angle, and Paying the Price Honestly

The published figures for the wireless tiltmeter family used for municipal work are worth setting out, because they explain why field practice lands so far above the physical signal. The measuring range is plus or minus 30 deg, resolution is 0.001 deg, accuracy is 0.005 to 0.01 deg, the alarm accuracy is 0.01 to 0.1 deg, and the default alarm angle is 3 deg. The datasheet also defines the alarm as requiring the measured tilt to remain beyond the alarm angle for a defined period, which is the mechanism that rejects a single wheel pass instead of reacting to every oscillation.

Now compare that default with the alarm angle recorded in the Beijing deployment, which was fifteen degrees. Three degrees may look generous against a resolution of 0.001 deg, and it is, by a factor of three thousand. It is also six hundred times the accuracy figure. But three degrees is already far above the small angles that a disturbed cover produces, and fifteen degrees is chosen because it sits clearly above the traffic floor and admits almost no ambiguity. A fifteen degree change of attitude is not a sensor artefact. It is a lid that has moved.

The honest cost of that choice should be written into the design rather than discovered later. At an alarm angle of fifteen degrees, a cover cocked by five millimetres on one edge, which is 0.48 deg, will never reach the threshold, and neither will a frame that has settled by a few millimetres. The angle channel in that configuration is a displacement and theft detector operating through attitude. It is not a settlement detector, and it should not be asked to be one.

The way to recover the settlement information is to stop reading the alarm and start reading the rest position. On an undisturbed cover the resting angle should be flat and stable, and a slow drift of that resting angle over months is precisely the signature of a frame settling or a chamber moving below grade. That is a trend question rather than an event question, it asks nothing of the alarm threshold, and it costs nothing extra once the data is on the platform. It also protects the operator from the failure mode that matters most in practice, which is a system that is trusted because it is quiet.

 

7. The Radio Path, Not the Bracket, Is the Hard Installation Problem

Mounting rules for industrial tilt sensors are well established: a rigid, flat, clean surface, and the sensor axis parallel to the axis of the surface being measured, with no angle between them. On a manhole cover those rules are easy to satisfy. The installation problem that actually decides whether the project works is somewhere else, and it is radio.

Cast iron attenuates NB-IoT and LoRa signals by ten to thirty decibels depending on the thickness and the alloy, and when the whole antenna sits under a cast iron cover a reliable connection can be difficult to establish at all. Field reports are consistent on the mechanism: an internal antenna pressed against a cast iron cover loses more than twenty decibels of signal strength and may fail to register on the network entirely. The published remedy that appears in installation guidance is to get the radiator outside the metal envelope, either through a top mounted antenna dome fitted into the cover or through an external antenna kit with the cable run to a connector on the cover. The wireless tiltmeter family used for this work publishes an external antenna option, about eleven centimetres long and ten millimetres in diameter, which is what makes that routing possible.

The size of the effect is documented. A flood monitoring project using three hundred smart manhole covers raised its first attempt signal reception rate from sixty two percent to ninety seven percent by replacing an under cover antenna with a dome antenna on the top face. The same account attributes a reduction of roughly thirty five percent in rework cost to that single change, equivalent to about one hundred and eighty labour hours on one project. Antenna placement was not a detail of that deployment. It was the difference between a working system and a partial one.

There is a tempting shortcut and it carries a real risk. Bolting a bracket through a certified cover changes the cover. If a heavy duty cover is to be drilled, the guidance that accompanies these installations is to keep bolt holes at least fifty millimetres from the edge of the cover, to avoid the outer quarter of the radius where the bending moment is highest, to use load spreading washers under the bolt heads, and to keep the number of holes to a minimum. The warning that comes with it is the important part: opening holes in a load class cover can compromise the structural certification that the cover was bought for. Where the cover is composite or resin the drilling question disappears and so does most of the radio problem, but the load class is then usually lower, so the material choice trades one constraint for another.

One commissioning rule deserves to be on a checklist. Test the link with the cover closed, and do it more than once. An open cover presents a healthy signal to the installer and the site then loses the device the moment the cover is lowered into place, which is a failure mode that has been reported often enough to be treated as normal. Where a signal strength figure is used as an acceptance criterion, it should be recorded in the covered condition, because that is the condition the system will live in.

 

8. The Chamber: Flooding, Gas, and What the Rating Covers

Below the cover is a chamber, and a chamber is not a dry cabinet. Three environmental factors decide how long the electronics last, and only one of them is settled by the headline rating.

The first factor is temperature, and here the published figures are unusually comfortable. The wireless tiltmeter family publishes an IP67 rating and an operating range of minus 40 to plus 85 degrees Celsius, with a zero point temperature drift of plus or minus 0.002 deg per degree Celsius. Buried soil damps the annual temperature swing compared with air, so even a generous estimate of twenty five degrees Celsius of annual movement is 0.05 deg of drift. Against a three degree alarm angle that is under two percent, and it is a footnote. This is worth noting because it is the opposite of the conclusion for precision applications, where the same drift term can dominate the whole error budget. Here, precision is not what limits the system, so the term that limits a precision system does not matter.

What the rating does not settle is flooding. IP67 describes protection against temporary immersion under defined conditions, which is a different requirement from a chamber that fills with storm water and stays full. That question has to be answered from the site rather than from a datasheet: either the device is mounted above the level the chamber has historically reached, or the enclosure is specified for continuous immersion, which is a higher and separately tested claim. Deciding this from the standard rather than from the site is how a monitoring scheme acquires a seasonal failure.

The second factor is gas. The standards being written for intelligent manhole covers treat hydrogen sulphide, methane, carbon monoxide and oxygen as expected constituents of the atmosphere inside these chambers and specify monitoring ranges and alarm values for them. That is a direct statement about the environment by the people who write the rules for it. A vented chamber carrying sewage is a corrosive atmosphere for anything with exposed metal, and a device that survives the first year without attention is a different product from one that survives the fifth.

The third factor is moisture, and it does not arrive as rain. It arrives along the cable. Water vapour migrates through cable entries and condenses inside the enclosure, and field practice in this sector is to treat sealing the housing as necessary but not sufficient, adding a conformal coating to the circuit board so that condensation inside the box does not reach the circuitry. Devices that look sealed and still fail after a few months are usually failing at that interface.

Finally, life. The published working time for this family is about three years at a twenty four hour heartbeat from an 8500 milliampere hour battery, and the standards for intelligent covers commonly expect five years of maintenance free operation. The gap closes in the operating pattern rather than in the cell, by lengthening the heartbeat and reporting events only, which is exactly what this application wants: a cover does not need hourly attention, and it does need to be heard from when something happens. That is a design decision about the reporting interval, and it should be made deliberately rather than inherited from a default.

 

9. Load Class, Standard, and the Record That Outlives the Cover

One more piece of the specification sits upstream of the sensor and is usually written by someone else. The load class of the cover tells you which vehicles will cross it, and the vehicle mix sets the noise floor the sensor has to work above. Choosing a class is therefore also choosing a monitoring difficulty, whether or not anyone notices at the time.

Class	Test load	Where it is installed
A15	15 kN, about 1.5 tonnes	Pedestrian and cycle areas only
B125	125 kN, about 12.5 tonnes	Footways, pedestrian precincts, car parks for light vehicles
C250	250 kN, about 25 tonnes	Kerbside channels and slow moving traffic
D400	400 kN, about 40 tonnes	Carriageways and hard shoulders, the default for adopted roads
E600	600 kN, about 60 tonnes	Docks, warehouses and heavy industrial yards
F900	900 kN, about 90 tonnes	Airport pavements and extreme wheel loads

The class letter is the proof load in kilonewtons, and the proof load runs about one and a quarter times the rated working load, so a D400 cover is proven at four hundred kilonewtons against a rated capacity of roughly three hundred and twenty. It is a static acceptance test against a defined location, not a statement about dynamic behaviour, and it says nothing at all about how much the cover will move. What it does tell you, reliably, is the traffic. A cover specified to D400 lives in a carriageway, and the alarms on that cover have to be designed for a carriageway.

Two standards matter for cross border projects. In Europe, the EN 124 series classifies manhole tops and gully tops and defines those six classes, and it is the reference that municipal specifications around the world quote even where they are not legally bound by it. In China, GB/T 23858 covers manhole tops using the same class letters and the same test loads, so a specification can move between the two systems without renegotiating the requirement. Separately, GB/T 41401-2022, issued in April 2022 and effective from November 2022, is the national standard for intelligent manhole covers themselves, which is the document that turns this from a component choice into a defined product category.

The most instructive documents are the industry standards being written for intelligent covers in specific networks. A power industry standard for covers in cable ducts sets alarm requirements for the tilt function: an alarm error not exceeding plus or minus 2 deg, an alarm response within thirty seconds, and zero missed events. Look carefully at what those numbers are. They are alarm requirements, not measurement requirements. A permitted alarm error of two degrees is four hundred times the accuracy of the sensor family discussed here. The people who wrote that standard had understood the same thing this article has been describing: on a manhole cover, the precision of the instrument is not the constraint. What has to be got right is the choice of quantity, the setting of the angle, the delay before an alarm, and the radio path back to the platform.

Item to specify	What to write
Failure to detect	Name displacement and removal, not tilt in general
Sensing channels	A presence contact for removal, plus an angle channel for attitude and condition
Alarm angle	Set above the site traffic floor, and record the reason for the value chosen
Dwell requirement	Require the condition to persist before alarming, not a single instantaneous sample
Radio path	Antenna outside the metal envelope or a non metallic cover, verified with the cover closed
Enclosure rating	Matched to the chamber flood history rather than to the minimum in a standard
Mounting face	State whether the device follows the lid or the frame, and which events that decision gives up
Lifecycle	A written procedure for re zeroing the device after cover replacement or road resurfacing

The last row is the one that gets forgotten, and it is the one that eventually decides whether the record is worth anything. A manhole cover is a consumable. It is replaced when it is damaged, when the carriageway above it is resurfaced, and when it is stolen. Every one of those events resets the physical baseline of the sensor, and a baseline that is not reset in the platform leaves a permanent offset that looks exactly like a tilted cover for the rest of the asset’s life. The published account of the Beijing deployment handled this by giving every cover its own identity and recording the owner, the contact and the telephone number against each one, so that the record and the hardware travel together. That is the right unit of procurement: not a sensor count, but a per cover record with a device attached.

One operational detail closes the loop. Authorised openings have to be suppressed. If a maintenance crew lifts a cover and the platform raises an alarm that nobody in the workflow can mark as expected, the operators will do what people always do with an alarm that is usually wrong, which is to stop reading it. A maintenance mode is not a convenience feature. It is what keeps the remaining alarms meaningful.

 

Frequently Asked Questions

1. Can a tilt sensor detect a stolen manhole cover? Partly, and the limit has to be understood before it is relied on. The sensor will register the change of attitude when the cover is lifted at an angle, tipped against a wall, or set down on its edge, and any of those produces a large angle compared with the flat reference. But a cover lifted straight up and laid flat again, or lifted and carried away, may produce almost no angle at all, because a uniform lift is a translation rather than a rotation. That is why the field deployment described in this article used a travel switch alongside the tilt sensor, and why the two channels should be specified together where theft is the concern.
2. What alarm angle should a manhole cover tilt sensor be set to? Higher than the traffic floor at that specific location, and the value should be recorded with its justification. The wireless tiltmeter family used for municipal work has a default alarm angle of 3 deg, adjustable across its range, and the published account of the Beijing deployment records fifteen degrees for that class of installation, adjustable by installation type. Both numbers are far above the physical angles that a disturbed cover produces, which is deliberate. A cover cocked by five millimetres on one edge of a six hundred millimetre opening changes attitude by 0.48 deg, so any angle threshold that is quiet in traffic will not reach it, and detection of that event has to come from the presence channel.
3. How accurate does a manhole cover tilt sensor need to be? Far less accurate than the datasheet’s best figure, which is the most counterintuitive point in this application. The industry standard for intelligent covers in cable ducts permits an alarm error of plus or minus 2 deg, which is four hundred times the 0.005 deg accuracy of the sensor family discussed here, and the alarm response is specified in seconds rather than in degrees. Precision is not the constraint on a manhole cover. The constraints are whether the sensor is measuring the quantity that actually changes, how high the alarm angle sits above the traffic signal, whether the alarm requires the condition to persist, and whether the radio link survives a closed cast iron lid.
4. Why does a wireless sensor under a cast iron cover lose signal? Because the cover is a metal plate larger than the wavelength, and cast iron attenuates NB-IoT and LoRa signals by ten to thirty decibels depending on thickness and alloy. An internal antenna pressed against a cast iron cover can lose more than twenty decibels and may fail to register on the network at all. The remedy used in the field is to move the radiator outside the metal envelope, either through a top mounted antenna dome fitted into the cover or through an external antenna kit, and the documented effect is large: one three hundred unit flood monitoring project raised first attempt signal reception from sixty two percent to ninety seven percent with that single change. Verify the link with the cover closed and in place, because an open cover flatters the measurement.

---