A passenger hears an MRT before they ever see the mathematics.
There is the low hum as the train begins to move.
A rising mechanical tone during acceleration.
The familiar rolling sound as steel wheels run over steel rails.
A sharper sound on some curves.
A rumble that enters through the floor.
A vibration that is felt rather than heard.
None of these is simply “noise”.
They are different physical signals generated by different mechanisms, travelling through different paths, reaching different receivers and occupying different frequency ranges.
One sound may begin with microscopic wheel–rail roughness. Another may begin with a motor or fan. Another begins when lateral wheel–rail forces excite a curve. One vibration travels through the train body to the seat. Another travels through the track slab, tunnel and ground before becoming a faint vibration inside a nearby building.
MRT noise and vibration are not one problem. They are a source–path–receiver system written in waves.
The mathematics includes logarithms, oscillations, frequency spectra, differential equations, Fourier transforms, probability, signal processing, resonance, damping, wheel–rail mechanics and optimisation.
Singapore gives the mathematics a direct public example. The Land Transport Authority has stated that underground MRT interior noise is generally managed against an 85 dB LAeq threshold and has identified some curved sections of the Thomson-East Coast Line where higher noise is associated with track curvature. Public mitigation has included reducing train speed in those areas and grinding the rail surface to smooth the wheel–rail contact.
SMRT similarly describes preventive wheel profiling, rail grinding, daily track inspection and Rail Vision monitoring as part of controlling track condition and noise.
This article continues the eduKateSG MRT mathematics cloud. The permanent whole-system hub is How MRT Works | It’s Mathematics. The closest specialist foundations are How MRT Wheel–Rail Contact Works Using Mathematics and How MRT Track Inspection Works Using Mathematics.
The RFE — Why Does MRT Noise and Vibration Mathematics Exist?
The weakest objective would be:
make the railway silent
A moving mechanical system cannot be perfectly silent.
Wheels must transmit forces. Motors must work. Fans move air. Rails and train structures deform elastically. Brakes operate. Doors move. Track supports carry repeated dynamic loads.
Zero vibration would be equally unrealistic.
The useful objective is:
control noise and vibration so that the railway remains safe comfortable reliable maintainable and acceptable to passengers and surrounding communities without destroying the transport function
That means the problem is not:
minimise sound at any cost
It is:
identify the meaningful source understand how it travels measure what the receiver experiences reduce the dominant mechanism and verify that the intervention worked
The RFE is not silence. It is controlled mechanical energy: enough to move the train, little enough unwanted vibration that the journey remains civilised.
Prompt 1 — What Is Sound Mathematically?
Sound is a pressure disturbance travelling through a medium such as air.
Let ordinary atmospheric pressure be patm.
A sound wave creates a much smaller fluctuating pressure p′(t):
p(t) = patm + p′(t)
For a simple pure tone:
p′(t) = P sin(2πft + φ)
where:
- P is pressure amplitude,
- f is frequency in hertz,
- φ is phase.
Frequency tells us how many oscillations occur per second.
Period is:
T = 1/f
Wavelength λ is related to wave speed c:
c = fλ
For airborne sound at ordinary room conditions, c is roughly a few hundred metres per second, varying with temperature and atmospheric conditions.
Why decibels are logarithmic
The range of sound pressure humans can encounter is enormous.
A logarithmic scale compresses that range.
Sound pressure level is:
Lp = 20 log10(prms/p0)
where p0 is the reference sound pressure, conventionally 20 micropascals in air.
The logarithm produces several non-intuitive consequences.
If pressure doubles:
ΔLp = 20 log10(2)
≈ 6.02 dB
But if acoustic power doubles:
ΔLW = 10 log10(2)
≈ 3.01 dB
This is why two equal noise sources do not add by ordinary arithmetic.
Adding two noise levels
Suppose two independent sources each produce 70 dB at a receiver.
Total level is:
Ltotal = 10 log10(10^(70/10) + 10^(70/10)) ≈ 73.0 dB
Not 140 dB.
If one source is 80 dB and another is 60 dB:
Ltotal = 10 log10(10^8 + 10^6) ≈ 80.04 dB
The quieter source contributes almost nothing to the total level.
This gives us an important diagnostic rule:
If one noise mechanism dominates by a large margin, reducing a smaller source may produce almost no measurable improvement.
Equivalent continuous sound level
Train noise changes with time.
Acceleration, cruising, curves and braking are different.
Equivalent continuous level Leq compresses a varying sound history into the constant level containing the same acoustic energy over interval T.
Leq = 10 log10[(1/T) ∫0T 10^(L(t)/10) dt]
LAeq applies A-frequency weighting before or during the equivalent-level calculation to reflect the frequency sensitivity of human hearing.
LTA’s public reference to 85 dB LAeq for underground MRT interior noise therefore describes an energy-equivalent, frequency-weighted level rather than one instantaneous sound-pressure spike.
Prompt 2 — What Is Vibration Mathematically?
Sound and vibration are related but not identical.
Vibration is oscillatory motion of a solid structure or body.
The simplest model is a mass attached to a spring and damper.
m ẍ + c ẋ + kx = F(t)
where:
- m is mass,
- c is damping,
- k is stiffness,
- x is displacement,
- F(t) is an external forcing function.
With no damping and no external force, natural angular frequency is:
ωn = √(k/m)
Natural frequency is:
fn = (1/2π)√(k/m)
Damping ratio is:
ζ = c/(2√km)
These equations apply conceptually to many railway structures:
- wheels,
- rails,
- track slabs,
- bogies,
- car bodies,
- panels,
- tunnel structures,
- ground and nearby building elements.
Resonance
If the forcing frequency approaches a natural frequency, vibration response can become much larger.
For harmonic force:
F(t) = F0 sin(ωt)
the steady-state displacement-amplitude ratio is:
X/(F0/k) = 1 / √[(1−r²)² + (2ζr)²]
where:
r = ω/ωn
When r approaches 1 and damping is small, response increases strongly.
This is resonance.
Railway noise research studies coupled wheel–track resonances because resonant vibration can increase both radiated sound and irregular wear.
Transmissibility
A vibration-isolation system tries to reduce how much motion passes from source to receiver.
For a simple base-excited single-degree-of-freedom system, displacement transmissibility has the form:
T(r) = √[(1+(2ζr)²) / ((1−r²)²+(2ζr)²)]
At frequencies sufficiently above resonance, isolation improves.
This is one mathematical reason resilient track supports, floating slabs and suspension elements can reduce vibration transmission in appropriate designs.
Noise asks how much acoustic energy reaches the ear. Vibration asks how much mechanical motion survives the path through solids.
Prompt 3 — Where Does MRT Noise Actually Come From?
A useful diagnosis begins by separating sources.
| Source family | Physical origin | Typical mathematical lens |
|---|---|---|
| Rolling noise | Wheel and rail roughness excite both structures | Roughness spectrum, contact dynamics, radiation efficiency |
| Curve squeal | Lateral creepage, stick–slip and structural vibration | Nonlinear friction, stability, modal response |
| Impact noise | Local irregularities such as wheel flats or joints | Impulse response and broadband vibration |
| Traction noise | Motors, gearboxes, power electronics | Rotational frequencies, harmonics, electromagnetic excitation |
| Auxiliary noise | Fans, air-conditioning, compressors, pumps | Rotating machinery spectra and airflow noise |
| Aerodynamic noise | Airflow around moving train | Fluid dynamics and speed-dependent scaling |
| Structure-borne noise | Track vibration travels through tunnel/ground/building, then radiates as sound | Transfer functions and wave propagation |
| Maintenance noise | Grinding, welding and engineering work | Time exposure, source level, propagation |
For metro speeds, wheel–rail rolling interaction is often a major source. Research on railway pass-by noise describes rolling noise as vibration of the wheels and track excited by their combined surface roughness.
That sentence contains the whole mechanism:
surface roughness → relative vertical displacement at contact → dynamic wheel–rail force → wheel vibration + rail/track vibration → sound radiation into air
The source is not “the rail” or “the wheel” alone.
The two surfaces interact.
Why source separation matters
Suppose measured cabin or wayside noise is high.
If wheel vibration dominates, rail grinding alone may have limited effect.
If rail roughness dominates, changing an interior panel may have little effect.
If noise is structure-borne, an airborne barrier may solve the wrong path.
Modern source-separation research therefore uses measurements and models to estimate the relative contribution of wheel and track noise.
The decision chain is:
measure receiver → identify dominant frequencies → trace transfer path → estimate source contributions → act on dominant mechanism
Noise control begins by refusing to treat every decibel as though it came from the same place.
Prompt 4 — How Can a Microscopic Rail Roughness Become an Audible Frequency?
This is one of the most elegant equations in railway acoustics.
Suppose a wheel rolls over a repeating surface roughness with spatial wavelength λ.
If train speed is v, the wheel encounters each repeating feature at frequency:
f = v/λ
Surface geometry has become time frequency.
Worked roughness example
Suppose a train travels at:
v = 20 m/s
over a repeating roughness wavelength:
λ = 0.04 m
Excitation frequency is:
f = 20/0.04 = 500 Hz
If train speed increases to 24 m/s:
f = 24/0.04 = 600 Hz
The physical roughness did not move.
The excitation frequency changed because the train encountered it faster.
This explains why the same track can sound different at different speeds.
Roughness as a spectrum
Real rail roughness is not one sine wave.
It contains many wavelengths.
Represent rail height irregularity as:
r(x)
A spatial Fourier transform decomposes it into wavelength or wavenumber components.
R(k) = ∫ r(x)e^(−ikx) dx
Wavenumber k relates to wavelength:
k = 2π/λ
When the train moves, each spatial component maps into a temporal frequency.
ω = kv
or:
f = v/λ
The wheel–rail system then amplifies some components more than others according to its frequency response.
So measured noise depends on:
roughness spectrum × contact force response × wheel vibration response × rail/track vibration response × acoustic radiation × receiver path
This is why “smooth rail” matters acoustically.
Rail grinding changes r(x).
Changing r(x) changes the excitation spectrum presented to the wheel–track system.
A new surface can temporarily sound different
SMRT notes publicly that noise can initially be louder just after track grinding because fresh grinding marks need time to smooth through operation.
That is a good reminder that maintenance does not simply move a variable monotonically towards “better”.
grinding changes surface spectrum → short-term acoustic response can change → running smooths the surface interaction → longer-term condition is measured again
A rail roughness only millimetres or centimetres long can become a tone hundreds or thousands of times per second once a train carries it through time.
Prompt 5 — Why Do Curves Sometimes Squeal?
Curves make the wheel–rail contact solve a harder geometry problem.
The outer wheel travels a longer path than the inner wheel. Wheel profiles create a rolling-radius difference that provides passive steering, but on tighter curves this mechanism may not completely satisfy the required geometry.
Lateral creepage and flange interaction can increase.
The wheel–rail contact becomes a nonlinear frictional system.
At a teaching level, friction force can be represented:
F = F(ξ, μ, N, surface state)
where ξ is creepage.
If the slope of the effective friction–creep relationship and the structural dynamics combine unfavourably, self-excited vibration can occur.
small lateral slip → friction force changes → wheel vibrates → contact state changes → friction changes again → vibration grows
This is a feedback instability.
It can create curve squeal.
Why speed matters on curves
Lateral centripetal acceleration is:
ac = v²/R
Reducing speed lowers v²/R strongly.
Suppose curve radius is fixed and speed falls from 20 m/s to 16 m/s.
ratio of lateral acceleration = 16²/20² = 256/400 = 0.64
A 20 per cent speed reduction produces a 36 per cent reduction in this acceleration measure.
LTA’s public TEL mitigation uses this same physical direction: slower operation is one of the measures used in noisier curved sections, alongside rail grinding.
But slower speed has costs:
- longer running time,
- possible timetable change,
- potential effect on headway and recovery margin,
- different energy use.
The mitigation is therefore a system optimisation, not an acoustic decision alone.
Curve radius is urban geometry
Why build a tight curve in the first place?
Rail lines must pass through a real city.
They avoid buildings, foundations, utilities and other infrastructure while trying to serve useful station locations and limit land acquisition.
LTA’s TEL explanation explicitly links the curved alignment to land and commuting constraints.
The noise therefore has a deeper geometry:
city geometry → tunnel alignment → track curvature → wheel–rail lateral mechanics → vibration → sound at passenger receiver
A sound heard inside a carriage can be the final audible consequence of a land-use decision made years before the train existed.
Prompt 6 — How Does Vibration Travel from Wheel to Tunnel, Ground and Building?
Railway vibration is best understood as a source–path–receiver chain.
SOURCE wheel–rail dynamic force PATH rail → fastener/pad → slab/track → tunnel → ground → structure RECEIVER train body, passenger, nearby building or sensitive equipment
Each part has its own frequency-dependent transfer function.
If input force spectrum is F(ω), the response at receiver can be represented:
Y(ω) = H(ω)F(ω)
H(ω) is the transfer function.
For several stages:
Htotal(ω) = Htrack(ω) × Htunnel(ω) × Hground(ω) × Hbuilding(ω)
Again, this is a conceptual linear representation. Real soil and structural systems can be more complex.
Attenuation
Vibration usually decreases with distance because energy spreads geometrically and is dissipated by material damping.
A simple amplitude model is:
A(r) = A0 r^(−n) e^(−αr)
where:
- r is distance,
- n represents geometric spreading,
- α represents material attenuation.
The exact values depend strongly on geometry, soil and frequency.
Structure-borne noise
Vibration can reach a wall, floor or building structure and make it vibrate.
That vibrating surface can then radiate sound into the room.
track vibration → ground vibration → building vibration → wall/floor radiates sound → receiver hears rumble
This is why railway vibration research separates vibration generation from secondary acoustic radiation.
Frequency dependence
Low and high frequencies do not travel identically.
Ground, track and buildings have different resonances and attenuation behaviour.
A mitigation that works beautifully at 100 Hz may do little at 20 Hz.
This is why the correct question is not:
How much vibration is there?
but:
How much vibration exists at each frequency, where did it originate, which path amplified or attenuated it, and what does the receiver experience?
Recent state-of-the-art reviews of train-induced environmental vibration emphasise exactly this coupling among vehicle, track, tunnel or surface infrastructure, soil and buildings.
A railway vibration problem can begin at a wheel and end as a room vibration because every structure in between agreed to carry part of the wave.
Prompt 7 — How Do Engineers Measure What the Ear and Body Receive?
A single overall decibel number is useful but incomplete.
Two sounds can have the same overall level and feel completely different because their frequencies differ.
Likewise, two vibrations can have the same RMS acceleration and produce different effects if their spectra and durations differ.
RMS vibration
For acceleration samples a1…an:
aRMS = √[(1/n)Σai²]
RMS represents effective oscillation magnitude over the interval.
Peak and crest factor
Peak acceleration is:
apeak = max |a(t)|
Crest factor is:
CF = apeak/aRMS
A high crest factor indicates strong impulsive peaks relative to average vibration.
This can help distinguish smooth continuous vibration from occasional impacts.
Fourier transform
A time signal x(t) can be transformed into frequency content:
X(f) = ∫x(t)e^(−i2πft)dt
For sampled data, the Fast Fourier Transform provides an efficient numerical calculation.
The spectrum can reveal:
- motor harmonics,
- wheel rotation components,
- roughness-related bands,
- track or body resonances,
- curve squeal tones,
- impact events.
Power spectral density
Power spectral density Sxx(f) shows how signal power or variance is distributed across frequency.
Variance can be recovered approximately by integrating PSD:
variance ≈ ∫ Sxx(f)df
PSD is especially useful for broadband railway vibration.
Octave and one-third-octave bands
Acoustics often groups frequencies logarithmically.
For an octave:
fupper/fLower = 2
For a one-third-octave band:
fupper/fLower = 2^(1/3)
These bands correspond more naturally to how hearing and engineering standards often group frequency effects than equal-width hertz bins.
Coherence
Suppose x(t) is track vibration and y(t) is interior sound.
Magnitude-squared coherence is:
γ²xy(f) = |Sxy(f)|² / [Sxx(f)Syy(f)]
Values near 1 suggest a strong linear frequency-by-frequency relationship.
This can help determine whether a suspected vibration source is actually connected to the receiver signal.
Transfer path analysis
When several paths contribute to cabin noise, engineers can estimate:
receiver response = Σ source contribution × path transfer function
Research separating wheel and track contributions during train pass-by uses variants of this idea, combining measurement and modelled transfer behaviour.
The measurement problem itself
A measurement can be wrong because:
- the microphone or accelerometer is poorly placed,
- background noise contaminates the signal,
- train speed differs between tests,
- the wheel or rail condition changed,
- the sample period is too short,
- the sensor is not calibrated,
- the receiver position is inconsistent.
So repeatability matters.
SMRT says train-noise measurements are taken periodically at selected locations near major residential estates so preventive maintenance can be prioritised as levels approach applicable guidelines.
A decibel reading is not the diagnosis. It is one projection of a frequency-dependent mechanical system into one receiver and one measurement window.
Prompt 8 — How Is MRT Noise and Vibration Reduced Without Breaking the Railway?
There are three basic intervention locations:
SOURCE PATH RECEIVER
1. Reduce the source
- Smooth rail roughness through grinding.
- Restore wheel roundness and profile through wheel profiling.
- Reduce curve speed where appropriate.
- Manage friction in authorised engineering applications.
- Improve rotating machinery balance and condition.
- Maintain track geometry.
Source treatment is often powerful because it prevents vibration energy from entering the rest of the system.
2. Interrupt the path
- Resilient rail pads and fasteners
- Track slabs designed for vibration isolation
- Floating-slab track where justified
- Damping treatments
- Wheel dampers or absorbers
- Acoustic barriers for airborne paths
- Tunnel and building isolation treatments
Path treatment targets the transfer function H(ω).
receiver = H(ω) × source
If the source cannot be reduced enough, make H smaller at the problematic frequencies.
3. Protect the receiver
- Cabin sound insulation
- Interior panel damping
- Building isolation
- Acoustic treatment
- Space planning and setback where applicable
Receiver treatment is useful when source or path changes are constrained.
The mitigation optimiser
Let intervention j have:
- noise reduction ΔLj,
- vibration reduction ΔVj,
- cost Cj,
- maintenance burden Mj,
- journey-time effect Tj,
- reliability consequence Rj.
A conceptual objective is:
minimise J = wL Lreceiver + wV Vreceiver + wC cost + wT journey-time penalty + wM maintenance burden + wR reliability penalty
subject to safety, service and engineering constraints.
This explains why the acoustically strongest intervention may not be the whole-system optimum.
For example:
reduce speed heavily → lower some curve-noise mechanisms → longer journey time → timetable and headway consequences → fleet and capacity consequences
Or:
grind rail more often → smoother contact and lower rolling noise → more engineering work and material removal → altered maintenance cost and rail life
So the correct question is not:
Which treatment reduces the most decibels?
It is:
Which combination removes the dominant unwanted mechanism while preserving the railway’s transport, maintenance and reliability functions?
A Complete Worked Noise-and-Vibration Example
Consider a fictional underground curve.
All values below are educational and do not represent a Singapore MRT operating section.
Step 1 — Rail roughness creates excitation
Train speed:
v = 18 m/s
Dominant roughness wavelength:
λ = 0.03 m
Excitation frequency:
f = v/λ = 18/0.03 = 600 Hz
Suppose the rail has a strong vibration response near 600 Hz.
Roughness and structural response reinforce each other.
Step 2 — Two sound sources combine
At one passenger location:
wheel–rail rolling contribution = 78 dB traction/auxiliary contribution = 68 dB
Total is:
Ltotal = 10 log10(10^7.8 + 10^6.8) ≈ 78.4 dB
The 68 dB source adds only about 0.4 dB because rolling noise dominates.
If engineers spend heavily reducing the 68 dB source by 10 dB, total barely changes.
The dominant source should be treated first.
Step 3 — Rail grinding changes roughness
Suppose treatment reduces the dominant rolling contribution from 78 dB to 73 dB.
New combined total:
Lnew = 10 log10(10^7.3 + 10^6.8) ≈ 74.2 dB
Total falls by about:
78.4 − 74.2 = 4.2 dB
A 5 dB reduction in the dominant component did not create a full 5 dB total reduction because the second source now matters more.
This is diminishing return from source competition.
Step 4 — Curve speed changes the excitation
Reduce speed from 18 m/s to 15 m/s.
The same 30-mm spatial wavelength now excites:
f = 15/0.03 = 500 Hz
If the rail response is much lower at 500 Hz than at 600 Hz, noise can fall for two reasons:
- lower dynamic curve demand,
- roughness excitation moved away from a strong resonance.
If instead 500 Hz aligns with another resonance, the result may be smaller than expected.
This shows why speed–noise relationships are not always one simple power law.
Step 5 — Measure vibration transfer
Suppose rail acceleration at 100 Hz is:
Arail = 1.0 m/s²
Measured car-floor acceleration is:
Afloor = 0.18 m/s²
Simple amplitude transmissibility at that frequency:
T = 0.18/1.0 = 0.18
In decibel amplitude ratio:
20 log10(0.18) ≈ −14.9 dB
The path attenuates that frequency by roughly 15 dB in this fictional measurement.
Step 6 — Resonance appears elsewhere
At 40 Hz, suppose:
Arail = 0.30 m/s² Afloor = 0.36 m/s²
Then:
T = 1.20 20 log10(1.20) ≈ +1.58 dB
The path amplifies rather than attenuates that frequency.
The mitigation should therefore be frequency-targeted.
Step 7 — Compare before and after
Before mitigation:
overall level = 78.4 dB strong band = 600 Hz floor resonance = 40 Hz
After grinding and a path treatment:
overall level = 74.2 dB 600-Hz contribution lower 40-Hz transmissibility reduced
The system should verify:
- receiver sound level,
- frequency spectrum,
- vibration RMS,
- passenger comfort,
- track and wheel condition,
- maintenance side effects,
- and whether the noise returns as surfaces wear again.
This last point matters.
A mitigation can decay with time.
grind → smooth → traffic → roughness grows → noise rises → inspect → grind again
Noise is therefore also a maintenance-state indicator.
The Noise and Vibration Deletion Tests
Remove the logarithm
Decibel levels are added incorrectly and source priorities become misleading.
Remove frequency
Two completely different acoustic mechanisms become one overall number.
Remove source separation
Engineers can spend money treating a source that contributes little to the received sound.
Remove the transfer path
Every receiver is assumed to experience the source identically.
Remove resonance
Large responses near natural frequencies appear inexplicable.
Remove speed
Spatial roughness is assumed to generate the same temporal frequency at all train speeds.
Remove track curvature
Curve-related lateral dynamics and squeal lose their geometric cause.
Remove measurement uncertainty
One microphone reading becomes absolute truth regardless of placement, background and operating state.
Remove maintenance feedback
A grinding or profiling intervention is assumed successful without measuring what happened afterwards.
Remove the passenger receiver
The system can optimise laboratory sound while ignoring what people actually hear and feel.
Noise mathematics fails when it measures the wave perfectly but forgets the receiver.
The MRT Noise and Vibration Paradoxes
Paradox 1 — Two 70 dB sources make about 73 dB, not 140 dB
Decibels are logarithmic.
Paradox 2 — Removing a quiet source may produce almost no audible improvement
A dominant source controls the logarithmic sum.
Paradox 3 — A tiny roughness can create a loud sound
Small geometric excitation can coincide with a strong structural resonance and become heavily amplified.
Paradox 4 — Smoother is not instantly quieter after grinding
Fresh grinding marks can temporarily alter the roughness spectrum before running smooths the surface.
Paradox 5 — A slower curve can improve noise but worsen timetable performance
Noise control and transport performance must be solved together.
Paradox 6 — A structure designed to isolate one frequency can amplify another
Vibration isolators have resonances and frequency-dependent transmissibility.
Paradox 7 — A quieter train can still create noticeable ground vibration
Airborne cabin noise and structure-borne ground vibration use different transfer paths.
Paradox 8 — More damping is not automatically better everywhere
Damping changes resonance and transmissibility, but excessive or poorly placed damping can create weight, cost and other design penalties.
Paradox 9 — Noise can be useful evidence
An unexpected new frequency or rising level can reveal changing wheel, rail or machinery condition.
Noise as a Maintenance Sensor
Noise should not only be treated as a nuisance to remove.
It can be evidence.
Suppose baseline spectrum at a location is S0(f).
Later spectrum is S(f,t).
Spectral change is:
ΔS(f,t) = S(f,t) − S0(f)
A persistent increase in one band may indicate:
- growing roughness,
- corrugation,
- wheel out-of-roundness,
- machinery degradation,
- looseness,
- or a new vibration path.
But acoustic diagnosis needs context.
Train speed changes the frequency mapping.
Wheel condition changes between trainsets.
Passenger loading changes structure response.
Background fans and station noise may contaminate measurements.
So a diagnostic model should condition on operating state:
expected spectrum = f(speed, train type, location, load, operating mode)
Then residual spectrum is:
E(f,t) = Smeasured(f,t) − Sexpected(f,t)
Repeated residual growth can become an anomaly signal for the predictive-maintenance layer.
This connects directly to How MRT Predictive Maintenance Works Using Mathematics.
The railway sometimes tells engineers that it is changing by changing its sound before it changes its service.
The Source–Path–Receiver Audit
- Who is the receiver? Passenger, operator, resident, building, sensitive equipment?
- What is the measured quantity? LAeq, peak sound, RMS acceleration, velocity, displacement, spectrum?
- At what location and time?
- What train speed and operating mode?
- What frequencies dominate?
- What is the likely source family?
- Does the frequency map to a spatial roughness wavelength?
- Is there a structural resonance nearby?
- What transfer path connects source to receiver?
- Does coherence support the suspected path?
- Are several independent sources being added logarithmically?
- Could background noise be contaminating the measurement?
- Is the condition repeatable across train passes?
- Is the issue train-specific or location-specific?
- Does wheel or rail inspection confirm the acoustic evidence?
- Would source treatment outperform path treatment?
- What timetable, energy or maintenance cost does mitigation create?
- What should be measured after intervention?
- How quickly can the benefit decay as surfaces wear?
- What evidence would prove the diagnosis wrong?
How the Mathematics Grows from School to Research
Primary Mathematics and Science
- waves and vibration,
- time and frequency,
- measurement,
- simple ratios.
Secondary Mathematics and Physics
- logarithms and decibels,
- frequency and wavelength,
- centripetal acceleration,
- simple harmonic motion,
- graphs and RMS values.
Junior College
- differential equations,
- damped oscillation,
- integration and energy,
- complex numbers and phase,
- probability and statistics.
University and Research
- structural dynamics,
- Fourier analysis,
- acoustics,
- vibroacoustics,
- transfer path analysis,
- wheel–rail contact mechanics,
- soil and wave propagation,
- finite-element and boundary-element methods,
- signal processing and machine learning,
- multi-objective optimisation.
The passenger hears one sound.
The mathematics hears a spectrum of causes.
The World Return — When the Passenger, Microphone and Track Answer the Model
A model predicts that rail roughness is the dominant source.
Rail is ground.
Noise should fall.
If it does not, something in the model is incomplete.
Perhaps wheel roughness dominates.
Perhaps a resonance still amplifies the new spectrum.
Perhaps the receiver is dominated by another path.
Perhaps fresh grinding marks temporarily increased one band.
The correct loop is:
measure receiver → decompose spectrum → infer source and path → intervene → measure source again → measure receiver again → compare predicted and observed reduction → revise diagnosis
Let predicted post-treatment level be L̂after and measured level be Lafter.
e = Lafter − L̂after
A persistent positive error means the treatment underperformed the model.
The model should change.
The same principle applies to vibration transmissibility, passenger comfort and environmental receiver measurements.
The sound after maintenance is the railway’s answer to the acoustic theory.
RFE Return — What Does Good MRT Noise Control Owe the Passenger?
It does not owe silence.
It owes a railway whose necessary mechanical energy does not create unnecessary human burden.
That means:
control dominant sources + prevent avoidable resonance + maintain wheel and rail surfaces + manage curves responsibly + isolate vibration where appropriate + measure the receiver honestly + verify that mitigation survives real operation
It also owes the surrounding city the same discipline.
Railways are embedded in neighbourhoods. Maintenance itself creates noise. Track geometry is influenced by the built environment. A good system therefore manages passenger comfort, community impact and the engineering work required to preserve both.
The RFE is not:
make every decibel disappear
It is:
keep the vibrations required for useful rail movement from becoming avoidable noise, discomfort, degradation or community harm.
Conclusion — The MRT Is a Moving Orchestra of Frequencies
An MRT carriage can sound like one machine.
It is not.
One frequency may come from a motor.
Another from a fan.
Another from a repeating rail roughness.
Another from a structural resonance.
A curve may create a nonlinear frictional vibration.
A wheel can excite a rail.
The rail excites the track slab.
The slab excites the tunnel.
The tunnel excites the ground.
The ground excites a building.
The building radiates sound again.
Meanwhile, every measured decibel sits on a logarithmic scale, every vibration has a frequency spectrum, every structure has natural modes, and every mitigation changes the transfer function rather than “removing noise” in the abstract.
rail roughness → dynamic contact force → wheel and track vibration → structural resonance → airborne and structure-borne paths → passenger or community receiver → measurement → diagnosis → grinding, profiling, isolation or control → measurement again
That is the real mathematics.
A noisy section is not merely loud.
It is a clue about geometry, contact, resonance, speed, surface condition and propagation.
The passenger hears sound.
The acoustician sees pressure spectra.
The dynamics engineer sees modes and damping.
The wheel–rail engineer sees roughness and creepage.
The maintenance engineer sees a surface that may need grinding or profiling.
The railway planner sees a curve whose speed and geometry belong to the timetable.
The city hears the combined answer.
The MRT is never silent because movement is physical. The engineering achievement is that millions of powerful mechanical interactions can become an ordinary, tolerable journey through a dense city.
Key Equations
p′(t) = P sin(2πft + φ)
Simple sound-pressure oscillation
T = 1/f
Period–frequency relationship
c = fλ
Wave-speed relationship
Lp = 20 log10(prms/p0)
Sound-pressure level
Ltotal = 10 log10(Σi 10^(Li/10))
Addition of independent sound levels
Leq = 10 log10[(1/T)∫10^(L(t)/10)dt]
Equivalent continuous sound level
m ẍ + c ẋ + kx = F(t)
Single-degree-of-freedom vibration model
fn = (1/2π)√(k/m)
Natural frequency
ζ = c/(2√km)
Damping ratio
X/(F0/k) = 1/√[(1−r²)²+(2ζr)²]
Harmonic response magnification
f = v/λ
Spatial roughness wavelength to temporal excitation frequency
k = 2π/λ
Spatial wavenumber
ω = kv
Moving spatial pattern to temporal angular frequency
ac = v²/R
Curve lateral acceleration
Y(ω) = H(ω)F(ω)
Frequency-domain transfer relation
A(r) = A0 r^(−n)e^(−αr)
Conceptual vibration attenuation with distance
aRMS = √[(1/n)Σai²]
RMS acceleration
CF = apeak/aRMS
Crest factor
X(f) = ∫x(t)e^(−i2πft)dt
Fourier transform
variance ≈ ∫Sxx(f)df
Power spectral density relation
γ²xy(f) = |Sxy(f)|²/[Sxx(f)Syy(f)]
Magnitude-squared coherence
ΔS(f,t) = S(f,t) − S0(f)
Spectral change over time
J = weighted receiver noise + vibration + cost
+ journey-time + maintenance + reliability penalties
Multi-objective mitigation model
Reader-safety note: The numerical acoustic examples, frequencies, source levels, transfer ratios and mitigation calculations in this article are educational abstractions unless explicitly attributed to a public source. They do not reproduce restricted MRT acoustic designs, internal operating limits, detailed tunnel transfer models or engineering control procedures.
Continue the MRT Mathematics Cloud
- How MRT Works | It’s Mathematics — the permanent whole-system hub.
- How MRT Timing Works Using Mathematics — speed, seconds and system timing.
- How MRT Braking Works Using Mathematics — stopping distance, braking curves and regeneration.
- How MRT Headway Works Using Mathematics — train spacing, frequency and regularity.
- How MRT Station Dwell Time Works Using Mathematics — doors, queues and passenger exchange.
- How MRT Delays Propagate and Recover Using Mathematics — delay feedback and recovery.
- How an MRT Timetable Is Built Using Mathematics — events, fleet and operating-day construction.
- How MRT Energy Use Is Optimised Using Mathematics — power, coasting and regenerative energy.
- How MRT Passenger Capacity Is Calculated Using Mathematics — train, station and line capacity.
- How MRT Routes and Transfers Are Optimised Using Mathematics — graph theory, transfers and route choice.
- How MRT Network Resilience Is Measured Using Mathematics — redundancy, cascading overload and recovery.
- How MRT Predictive Maintenance Works Using Mathematics — anomaly detection and remaining useful life.
- How MRT Track Inspection Works Using Mathematics — geometry, computer vision and track health.
- How MRT Wheel–Rail Contact Works Using Mathematics — contact pressure, friction, curves and wear.
The next natural pillar is How MRT Power Supply Works Using Mathematics: traction substations, voltage drop, current, power flow, regeneration, peak demand, redundancy, electrical faults and why the railway is also a moving electrical grid.
Sources and Further Reading
- Land Transport Authority — TEL track curvature, underground train noise, speed reduction and rail grinding
- SMRT — track maintenance, wheel profiling, rail grinding, Rail Vision and noise monitoring
- Railway Engineering Science — separating wheel and track contributions to railway rolling noise
- Urban Lifeline — review of railway vibration control and noise attenuation
- Railway Engineering Science — state-of-the-art modelling of train-induced environmental vibration
- Measurement — review of transportation noise and vibration signal processing and inference
- Journal of Sound and Vibration — wheel–track coupled resonance analysis