It turns out that a fish's lateral line is a sixth sense, related to the sense of touch in the same way the senses of smell and taste are related. Along the lateral line, which stretches from gill to tail and is visible in most fish species, is a row of specialized cells that sprout microscopic hairs, each attached directly to the end of a nerve cell. The hairs grow in clusters; the size of the clusters and of the hairs themselves vary between species. In some species, the hairs are even hidden in a protective cavity under the scales. In all fish species, however, the hairs send signals to the brain when the slightest current touches them. Using this row of extraordinarily sensitive sensors, fish can tell not only when something is approaching them, they can distinguish predator from prey from falling rock, and even detect unmoving obstructions by the eddies they produce in the ambient currents.
An artificial sensor with the same capabilities would first need to be on the same scale - a hair-like structure just under a millimetre tall. It would also have to stand up from the surface it's mounted on so fluid flow could move and bend it, and it would have to trigger a signal that can be sent to a computer for processing.
A group of researchers at the University of Illinois developed a cantilever design with a piezoresistive strain gauge at the base, and manufactured it using micromachined silicon.
The strain gauge is formed by diffusion doping part of the silicon substrate with boron, to provide a semiconductive material with the desired properties. A 300nm sacrificial layer of copper is deposited then etched into the desired design using photolithography, then a 600nm layer of gold is added and etched. The back of the silicon wafer is selectively removed to create the cantilever base with the right properties. The copper is dissolved away in dilute HCl, leaving a free - but not yet vertical - artificial hair attached at one end to the cantilever base opposite the strain gauge. The whole thing is then slowly lowered on top of a permanent magnet, and the hair is deformed by the magnetic field to stand vertically. To protect and strengthen the assembly, and to insulate the electrical connections so they can operate in conductive or reactive fluids, the entire thing can be coated with a thin Parylene film of a thickness dictated by a balance between insulation and sensor stiffening.
Using the process briefly described above, arrays of these artificial hairs can be created, packed as densely as 100 per square millimetre.
The micromachining turned out to be the easy part, however. Under simple laminar (layered, so that the centre moves fastest and the fluid velocity decreases smoothly as it nears the stationary surface of an object) or plug (extremely turbulent, so that the fluid seems to move in a solid "plug" past the stationary surface) flow, an equation for the relationship between the strain measured by the sensor and the average fluid velocity could be determined. However, pure laminar and pure turbulent flow aren't that common; the most common type of flow is a mixture of laminar and turbulent flow. The laminar flow, in such a mixed flow situation, is also often called the "boundary layer" because it exists along the surface, or boundary, of an object. Past the boundary layer, the fluid abruptly becomes more turbulent but not quite "plug", with eddies and vortices making the fluid flow harder to predict on a small scale. The boundary layer can be very thin, on the order of 1mm. The hair sensors - in fish and in the lab - are thus long enough to reach past the laminar boundary layer and into the turbulent flow regime, but short enough that the boundary layer can't be ignored. Because of the different flow regimes that affect the sensor, the finite-element numerical method, where the effect of the current on very short fractions of the hair is calculated then all of the fractions are combined to provide an overall picture of the whole hair, will often have to be used to calculate the flow velocity.
And this is just one hair.
This is also assuming unidirectional flow impinging on the hair directly in line with the cantilever. Outside of laboratory test equipment, water often flows crossways and opposite the main current, creating eddies that would push the sensor around and sometimes even counteract the effect of the main current. The fish (or robot) could also be oriented in any direction relative to the fluid flow.
To properly characterize the environment - and to have a hope of identifying, or even merely detecting, underwater objects - the hairs, like the hairs in fish, have to be made in clusters of hairs of different sizes and orientations, and some serious computing power will have to be used to figure out how the different sensor readings describe even a single current's direction and speed.
Despite the fact that fish are not particularly intelligent, they can still beat our best computers when it comes to real-time analysis of fluid dynamics using an array of sensors mounted on a moving surface, much like our brains can beat any computer out there when it comes to things like analyzing light impinging on an unsteady-state detector to instantly recognize and identify complex patterns.
It looks like our underwater robots will remain clumsy and "deaf" to the subtle movement of the surrounding water for a while yet. The information carried past their skin will remain unheard until computers become powerful enough to perform the extensive analysis needed to make this as valuable a sensor system as it is for a simple fish.
References: "Design and fabrication of artificial lateral line flow sensors", Z. Fan et al, Journal of Micromechanics and Microengineering, June 2002