Continuous gas leak detection, across the whole plant at once.
A mesh of small laser nodes rings the plant. Each node is a catalogue laser, a coherent receiver on a photonic chip and a small steering mirror; it sweeps the beam over pipes, walls and corner reflectors and records, for every path, the direction, the distance and the gas absorbed along it. Edge-AI turns hundreds of paths into one continuous answer — is there a leak, which joint, how big. Methane first, because the law names it; swap the laser for the next gas.
Read the gas off the plant itself — and name the joint
A gas absorbs infrared light at a wavelength that belongs to it alone. Shine a laser locked to that wavelength along a path, measure how much comes back, and you have the amount of gas on the path, as a calibrated number. Handheld laser detectors do this today through a lens the size of a saucer, one spot at a time, in a technician's hand. We put a coherent receiver on a photonic chip and feed it from a catalogue laser, so a node the size of a fist can read the same faint return through a 5 mm mirror, sweep forty paths a cycle, and run around the clock.
[a] A chip, not a telescope
Mixing the returned light with the laser's own light on the chip makes a single-mode receiver as sensitive as a large lens for the ratio that matters, the absorption on and off the gas line. That is what shrinks a lidar to a node cheap enough to deploy in numbers.
[b] Locked to the gas line
The laser is a bought narrow-line module held on the absorption line of the target gas, so the reading is about that gas and little else, and it does not depend on temperature or background. Next gas: next laser, same chip.
[c] Surfaces, plus reflectors where it counts
The beam reads off pipes, vessels and the ground for coverage; at the flanges and valves where the rule sets the finest threshold, a corner reflector glued to the steel returns the beam far brighter. A chirped laser tells how far away the surface that answered is.
[d] A mesh in 3D, not a scanner
Each node steers in two angles and measures range, so every dwell is a point in space with a gas value. Nodes sit at two heights, under the pipework and above the vessels, and hundreds of crossing paths cover the volume between them — the raw material the edge-AI inverts into "which joint is leaking."
Where it stands: the coherent receiver chip at the gas wavelength is the hard, novel part — and it is the piece we are still developing; the laser and the mirror are catalogue parts. The mesh geometry, the absorption physics and the leak-inversion AI are what the interactive model below demonstrates on a real plant model. We do not sell this system yet.
inside one node · laser, chip, mirror
A bought laser, a receiver chip and one small mirror
Each node has three parts. The laser is a bought module: a single-frequency diode laser at the methane line, held on the line by its temperature and swept across it by its current, with the temperature control and the optical isolator in the package. The light goes by fibre into the receiver chip, a photonic die that keeps a sample of the laser as local oscillator and mixes it with the faint light that comes back from a pipe, a wall or the ground. Mixing is what lets a chip-sized aperture read a return that a handheld reads through a lens the size of a saucer. The mirror, a 5 mm MEMS part, sends the beam out and collects what comes back. Where the law asks for the finest threshold, a cheap corner reflector glued to the steel returns the beam a hundred thousand times brighter.
[a] Why this laser
Three requirements pick it. It must sit on the methane line at 1650 nm. It must be tunable by current, because the node sweeps it 1 GHz during every dwell to measure distance. And it must hold its phase over the round trip: a plain DFB diode (1 to 2 MHz linewidth) is fine for a bench at 30 m; for surfaces 100 to 300 m away the linewidth has to stay under about 100 kHz, which is an external-cavity diode laser. Both are catalogue parts with the isolator inside, so the laser is a line in the bill of materials, not a chip project.
[b] Why eight channels and sixteen detectors
Light coming back from a rough surface is speckle: a pattern of bright and dark grains. A single-mode waveguide takes in one grain, and one grain is a noisy reading. A 5 mm aperture holds about eleven grains, so the chip has eight receive waveguides, each with its own mixer, and averaging eight grains at once cuts the speckle noise by the square root of eight. Each mixer has two outputs read by a pair of detectors whose difference cancels the laser's own intensity noise: eight pairs, sixteen detectors.
[c] One moving part
A 5 mm MEMS mirror, a catalogue part with micro-radian repeatability, tilts on two axes and points the beam at about forty targets a cycle. Mirror and chip share one aperture; the mirror is also the receiving lens.
[d] Traceable by design
Every reading is a calibrated absorption on a known path, with the laser locked to the gas line, and with range, concentration and noise logged per dwell.
Where it stands: the eight-channel coherent receiver chip is the novel part we are developing. Its building blocks exist at foundries today (splitters, mixers and balanced detectors on indium phosphide; mid-infrared detector dies flip-chipped onto silicon), the lasers are bought, and coherent methane lidars built from fibre parts have flown for years. Our sensitivity figures are our own calculations, not measurements; the first bench node is what measures them. Schematic is illustrative.
how a node scans in 3D
One dwell is one point in space. A cycle is a grid of them. Two heights make a volume.
A node does not take a picture; it takes a list of measured points. At each dwell the mirror holds the beam on one target for a millisecond and the node records four numbers: the two mirror angles, the distance to the surface that answered, and the gas absorbed on the way there and back. Two angles and a distance fix a point in space; the absorption attaches a gas value to the path that ends there. The mirror then steps to the next target: left and right along the rack in azimuth, up and down the rack in elevation, about forty points a cycle. One node therefore sees a fan of paths in front of it, which is a slice of the plant. Nodes at two heights, under the pipework and above the vessels, send their fans through the same space from opposite sides, so the paths cross, and where paths from both heights carry gas, the plant model has a flange or a valve. That is the joint the system names.
[a] Two angles, one mirror
The MEMS mirror tilts on two axes, about ±15° each. Azimuth sweeps along the rack and across to neighbouring nodes; elevation covers the height of the rack from below and from above. At 30 m, ±15° spans 16 m, the height of a typical process unit.
[b] Distance from the beat
During each dwell the laser frequency is ramped by 1 GHz. Light that travelled further comes back later and beats against the local oscillator at a higher frequency, so the beat note gives the distance in 15 cm steps. The node knows which pipe, wall or reflector answered, and the point is placed at that distance.
[c] Gas on the path
The same ramp crosses the methane line, so the on-line and off-line returns come from one dwell and one speckle pattern. Their ratio is the gas along the path, in ppm·m, independent of how bright the surface is.
[d] From crossing paths to a joint
Hundreds of paths per cycle, each with a known end point and a gas value, go into the inversion. It does not paint a cloud; it checks which listed joints explain both the paths that read gas and the paths that read clean, and names the one that fits.
The range gate and dwell figures are design values from our link budget (1 GHz ramp in 1 ms, 15 cm gate, ±15° mirror), not measurements. The interactive model below shows the same mesh on a plant model; the first bench node is what turns these numbers into data.
one node · many gases
Methane first. Any gas with a laser line next.
Laser absorption spectroscopy reads a gas by the infrared light it absorbs, and every gas that EU rules ask operators to measure absorbs somewhere between 1.3 and 16 micrometres, the range single-frequency lasers cover today. Because the laser is a separate module, moving to the next gas means swapping the laser and keeping the receiver chip, the mirror, the waveform and the inversion software. Gases that share a detector band share the same chip: methane, ammonia, acid gases and carbon dioxide in the near infrared all run on one die. A new receiver die, of the same design with detectors in another material, is needed only when the gas moves to the mid-infrared. A gas with a sharp line (methane, ammonia, SF₆) is read at one wavelength; a gas with a broad band (the refrigerants) is read by a small array of lasers stepping through its band, so the fit against known spectra names the gas or the blend. Methane is first because it carries the hardest written threshold and the largest buyer, not because the node is built for methane.
Gas
Name
Laser line
Who asks for the number
CH₄
methane
3.24 / 1.65 µm
Regulation (EU) 2024/1787: leak detection and repair, 1 g/h
C₂H₆
ethane
3.35 µm
tells a pipeline leak from biogenic methane; same laser scan
NH₃
ammonia
2.23 / 1.52 µm
ammonia-fuelled ships and bunkering; SCR slip under the IED
N₂O
nitrous oxide
2.27 µm
wastewater plants (Directive 2024/3019); nitric acid under the ETS
CO
carbon monoxide
2.33 µm
stack monitoring under the IED and MCP Directive; flare efficiency
waste incineration; aluminium smelters; battery fire toxicity
SO₂ / NO / NO₂
sulphur and nitrogen oxides
5–7.3 µm
stack monitoring (IED, MCP); roadside air quality (Directive 2024/2881)
SF₆ / NF₃ / CF₄
F-gases: switchgear and fab gases
7.8–11 µm
Regulation 2024/573: switchgear, fab abatement, 2035 bans; magnesium casting
R-32, R-134a, R-125, HFO-1234yf
F-gases: refrigerants and blends
7.5–9.2 µm
Regulation 2024/573 Art. 5–6: fixed leak detection; reefers, rail HVAC, data centres, MAC
C₄F₇N / C₆F₁₂O / HFC-227ea
F-gases: SF₆ substitutes, fire suppression
4.4 / 7.5–9 µm
new MV switchgear; fixed fire-suppression leak checks (2024/573)
Three laser families cover the table, all bought as modules: indium-phosphide diodes for the near-infrared lines, gallium-antimonide diodes and interband cascade lasers for 2 to 3.5 µm, and quantum cascade lasers from 4 to beyond 12 µm, as single-frequency devices or as arrays that cover a whole band. The receiver chip follows the detector band, not the gas: one indium-phosphide die for the near infrared, one silicon die with flip-chipped detectors for the mid-infrared. Line positions from the HITRAN database and published band tables; which gases ship, and when, follows the roadmap on our partners page.
interactive 3D
Turn the plant. See the mesh from every angle.
Drag to rotate, scroll to zoom. The model auto-rotates until you touch it; double-click to reset. The glowing markers show where beams cross the gas — the moment the system reads methane.
Laser absorption along every path, turned into a plant-wide picture
[1] Ring the plant
Low-power laser nodes mount on existing structures at two heights and form a self-surveying mesh — no trenching, no fixed sample lines, no single blind spot.
[2] Read every path in 3D
Each node sweeps its beam over surfaces and reflectors in two angles and measures the range to each target and the gas absorption along the line. Hundreds of crossing paths cover the volume between the nodes, not just points on a wall.
[3] Locate & name
Edge-AI inverts the mesh of path readings back to the source and selects which known joint is leaking, how high it sits and how fast it grows — selection, not a reconstructed cloud.
[4] Alert & log
A continuous record for MRV/LDAR reporting and an alert on threshold to your control room — always-on between the handheld surveys operators run today.
what it is — and isn't
An honest picture of where this stands
It is a system that names which joint is leaking from continuous mesh readings — the answer an operator acts on.
It is not a 3D reconstruction of the gas cloud. The AI selects among known leak points; it does not paint the plume.
The chip — a coherent receiver on a photonic die, fed by a catalogue laser that is swapped for the next gas — is the enabler, and it is in development. The mesh, the physics and the inversion are what we simulate here.
Why us: the team already designs gas sensing end to end — own optics, electronics, firmware and calibration traceable to national standards.