What a photograph of a flow shows

Almost every claim in aerodynamics is supported, somewhere down the chain, by a photograph of a flow. Smoke in a tunnel, tufts on a wing, oil streaked across a model, a plot from a pressure survey. They are the evidence.

And they measure four different things. A picture of a flow is not a picture of the flow; it is a picture of one quantity, selected by the apparatus, and the quantity is almost never stated.

Smoke records a streakline #

A smoke port is a fixed hole with dye coming out of it, and what appears downstream is the locus of everything that has passed through that hole. That is a streakline: not where the fluid is going, but where fluid that once visited one point has since got to.

In a steady flow the streakline coincides with the streamline through that point, and that coincidence is why nobody worries. It is also why the mistake survives, because the technique everyone has seen is the one case where the distinction costs nothing.

In an unsteady flow the two part company completely — the three curves through a flow are genuinely different objects — and every smoke photograph of a shedding wake, a gusting flow, or a manoeuvring model is showing something that is not a streamline and is routinely described as one.

There is a practical consequence that experimentalists know and readers of experiments often do not. A streakline photograph taken over a long exposure in an unsteady flow shows a time-averaged streakline, which is a fourth object again, and one that may resemble nothing that ever existed instantaneously. The classic smooth photographs of separated flow behind a bluff body are frequently of this kind: the real flow was thrashing about, and the picture is the average of many arrangements, none of them the one shown.

A tuft records a direction and nothing else #

A tuft is a short length of yarn taped to a surface. It lies along the local flow direction, and it is one of the cheapest and most useful instruments in aerodynamics: a wing covered in tufts shows at a glance where the flow is attached, where it has separated, and where the separation begins.

What it cannot do is say how fast. A tuft in a flow of five metres a second and a tuft in fifty look identical. The instrument has no magnitude in it at all.

That is why the tuft panel in the figure above draws every arrow the same length, deliberately. Drawing them to scale would be drawing something the instrument does not measure — which is the point of the panel, and the same caution the arrows on any vector plot deserve.

A tuft also averages over its own length and over its own response time, so it reports something smoothed in space and time, and it disturbs the flow it is in. On a laminar wing a row of tufts will trip the boundary layer to turbulence and change the answer.

An oil film records the friction, not the flow #

Surface oil-film visualisation is the technique that most often gets misread, and the misreading is subtle enough to be worth spelling out.

A thin film of oil on a model is dragged along by the shear stress at the wall, not by the flow above it. The streaks it leaves are skin-friction lines, and skin-friction lines are the field lines of a completely different vector field from the velocity.

Near a surface those two point much the same way, so on an attached flow the distinction rarely matters. It matters a great deal near separation, where the skin-friction lines converge onto a line and the oil accumulates there while the external streamlines simply lift away. The bright line an oil film shows at separation is a feature of the friction field, and reading it as a streamline gives a wrong picture of what the air is doing.

The panel in the figure above makes the point by having nothing to show. The flow drawn is ideal — inviscid — so there is no wall shear stress at all, and an oil film in it would record nothing. A technique that measures a quantity the model does not have produces a blank, and the panel says so.

A pressure tap records a scalar #

A pressure tap is a small hole connected to a transducer. It gives one number per hole, and pressure is a scalar: it has no direction in it whatever.

This is the most quantitative of the four techniques and the least pictorial. A row of taps round a section gives the pressure distribution, which integrates directly to the lift and the moment, and it is the measurement aerofoil data actually comes from. It says nothing about where the air is going.

The panel draws each tap’s reading as a line sticking out of the surface, which is a convention rather than a direction — the line’s length is the coefficient and its outward direction is chosen for legibility. A reader who took those lines for velocity vectors would have the flow leaving the body radially everywhere, which is the opposite of what the first panel shows about the same field. Two panels of one flow, disagreeing completely, because they are drawing two different quantities with the same graphical device.

What the solver computed, and how it was checked #

The four panels are one field rendered four ways, and that is the design: they cannot disagree about the flow because there is only one flow. What differs is which function of it each panel evaluates.

The field is a cylinder with circulation, computed in closed form and put through the site’s standard assertions before anything is drawn — divergence-free, tangent to the body, lift agreeing with ρUΓ.

The streamlines and the highlighted streaklines come from the same integrator with the same field, and in this steady flow they are the same curves. The figure says so in its caption rather than presenting a coincidence as a general truth.

The tuft directions are the velocity just off the surface, normalised to unit length before drawing — the normalisation is where the magnitude is deliberately discarded. The pressure lines are the Bernoulli coefficient evaluated on the surface. Nothing is sketched.

There is one more check worth naming, because it is the sort a figure of this kind can silently fail. The four panels are drawn from the same field object, in the same generator call, so there is no way for one panel to be showing a different flow from another. Had they been produced by four separate calls with four sets of parameters, a mismatched incidence or radius in one of them would have been invisible — four panels of a plausible flow, one of them a different plausible flow. Building them from one field is a structural guarantee rather than a checked property, and where a structural guarantee is available it is worth more than an assertion.

Why the steady case conceals everything #

Every one of the distinctions above is invisible in a steady flow, and that is the essay’s real point.

In steady flow, streamlines, pathlines and streaklines coincide. Skin-friction lines and near-surface streamlines nearly coincide. A tuft’s direction and a streamline’s direction coincide. Four techniques measuring four quantities give four pictures that look the same, and a reader who has only ever seen steady pictures has no reason to suspect that the quantities differ.

Then the flow becomes unsteady — which is most flows of practical interest, including every wake, every stall, every gust encounter, every rotorcraft — and the four pictures separate. Somebody who has never had to distinguish them has no framework for the fact that they now disagree.

This is a general hazard with a name it deserves. A distinction that costs nothing in the easy case is a distinction that will not be learned, and it will be needed in the hard case by people who have only ever met the easy one. Textbook fluid mechanics is full of them, and the coincidence of the three curves in steady flow is the most consequential.

The axis all four of these are bad at #

The four techniques differ in what they measure and agree in one thing: they are all pictures. They cover a surface or a plane at once and they resolve time badly — smoke diffuses, a tuft has a response time, oil accumulates over a run, and a tap sits behind a length of tubing that filters anything fast. So there is a whole half of the subject none of them can reach, and it is the half where turbulence lives, since a spectrum is a statement about time.

The instruments that reach it make the opposite trade. A hot wire is a few microns of heated wire held across the flow: moving air cools it, the current needed to hold its temperature is a measure of speed, and because the wire has almost no thermal mass it follows fluctuations to tens of kilohertz. One point, one number, no picture — and essentially every measured turbulence spectrum in the literature was obtained this way.

Its blind spot is the one worth carrying, because it is the same lesson as the oil film’s. Cooling depends on speed and not on which way the air is going. A wire in a reversed flow reports the same reading as one in a forward flow of the same magnitude, so inside a separated region it rectifies: the mean velocity it reports is biased towards zero from the wrong side, and the fluctuation level it reports is inflated by the folding. That is precisely the region a person is most likely to point it at, and the failure produces plausible numbers rather than obvious nonsense.

Which returns the essay to its own claim from the other direction. The four panels record different functions of one field at one instant; a hot wire records a different function again, over time, through a surface that maps two velocities onto one reading. No instrument here has recorded a flow. Each has recorded a projection, and the projections are not the same shape.

What the picture cannot show #

The figure has an honest limitation it shares with everything it is about: it draws four techniques and it is itself a fifth thing, a computed field. It shows what each instrument would record, which is a simulation of measurement rather than a measurement.

That is a genuine gap. A real smoke filament diffuses, and after a chord or two it is a broad grey smear rather than a line; a real tuft flutters; a real oil film runs under gravity as well as under shear, which is why models are sometimes tested inverted. None of those practical failures is drawn, and each is the reason experimental fluid mechanics is a craft.

The figure also cannot show the techniques it leaves out. Particle image velocimetry is the modern answer to all of this — seed the flow, illuminate a sheet, take two photographs a few microseconds apart, and cross-correlate to get the actual velocity field in a plane. It measures the thing the other four only approach. It is also expensive, gives a plane rather than a volume, and needs the seeding to follow the flow faithfully, which near a shock it does not.

And the panels are two-dimensional, which flatters every technique. Skin-friction lines on a three-dimensional body form patterns with nodes, saddles and foci that have no two-dimensional counterpart, and the topology of those patterns is one of the more powerful diagnostic tools available. Nothing here can hint at it.

Where the model stops #

The flow used is inviscid, which is why the oil-film panel is blank, and that blank is honest rather than convenient. A viscous field would have given something to draw there — this site’s grid solver produces one — but it would have been drawn at a resolution that could not represent the near-wall shear the technique actually responds to, and a plausible picture of an under-resolved quantity is exactly what the site’s rules forbid.

Nothing here is unsteady either, which is the limitation that matters most for the argument. The essay claims the four techniques separate in unsteady flow and demonstrates it by reference rather than by figure, because this site’s solver settles to a steady state and does not shed a vortex street. Making the claim visible would need a solver that does.

The surprise: the instrument chooses the physics #

There is a habit worth taking away from all of this, and it is not really about fluids.

A measurement technique does not record a system; it records a functional of a system, and which functional is a property of the apparatus. Smoke integrates along a history. A tuft projects onto a direction. Oil responds to a gradient at a wall. A tap reduces everything to one number.

None of these is wrong and none is complete, and choosing between them is choosing what the experiment is about. The choice is usually made on grounds of cost and convenience, and then quietly determines which features of the flow the resulting literature is able to discuss. The pictures then look like windows onto the flow, and they are windows onto one function of it, with the function unmarked.

The rule this site tries to follow is that a figure states what produced it. Every one carries a note naming its model and its regime for exactly this reason: a picture whose provenance is unstated is a picture that will eventually be read as something it is not.

That rule is doing more work than it looks. The three most-repeated wrong explanations this site takes apart — air meeting up again, Bernoulli applied where it does not hold, and the momentum sum done on one face — all survive partly because the pictures that accompany them are unlabelled. A smooth streamline photograph is compatible with almost any story about why the streamlines are where they are, and a story told next to a photograph inherits the photograph’s authority without having earned it.

Who found it, and when #

Smoke visualisation is Étienne-Jules Marey’s, from the 1890s, and his photographs of smoke past plates and cylinders are still reproduced. He was a physiologist studying motion who built the apparatus to photograph air the way he had photographed birds.

Ludwig Prandtl’s water channel of 1904 — the one that produced the pictures accompanying the boundary-layer paper — used aluminium particles on a water surface, which records pathlines rather than streaklines, and the difference between his pictures and Marey’s is precisely the distinction this essay is about. Neither man remarked on it.

The oil-film technique is post-war and largely British, developed at the RAE for transonic testing where smoke is useless. Particle image velocimetry dates from the mid-1980s and became routine in the 1990s once digital cameras and correlation hardware were cheap enough.

What is striking across that whole history is how rarely the recorded quantity is named in the caption. A century of published photographs, and the convention is still to write “the flow past a cylinder” rather than “streaklines from a rake of smoke ports, steady flow assumed”.

Van Dyke’s An Album of Fluid Motion, which is the standard collection and is on most fluid mechanicists’ shelves, is a partial exception and is worth the comparison. Its captions do generally name the technique, and reading it with attention is the fastest way to acquire the habit this essay is arguing for. It is also the fastest way to notice how much of the field’s visual intuition rests on a few hundred photographs taken between 1900 and 1980.

Where the ladder goes next #

Next rungs on this anchor: particle image velocimetry and what its assumptions actually are; schlieren and shadowgraph, which record the first and second derivatives of density and are therefore blind to a uniform flow however fast it is going; surface-flow topology, where the nodes and saddles of the skin-friction field obey a counting rule; and pressure-sensitive paint, which turns the scalar into a field and changes what a pressure measurement can be.

Then across to the three curves, which is the distinction this essay applies to instruments, and to what a flow is, which is the assumption every one of these techniques is quietly relying on.