A Spectrally™ X1 analyzer measures chemical composition inside a running process — no sampling, no reagents, no downtime. This page walks the optical chain that makes that possible, one component at a time.
Scroll
Seven components in series. The first one emits light, the last one hands a number to the plant's control system. Everything in between is there to keep one weak signal — inelastically scattered photons — intact.
Stokes-shifted light from a 785 nm source sits in the near infrared. The colour ramp on the detector is a visualisation, not what an eye would see.
A diode laser illuminates the medium at 785 nm, 600 mW at the fibre. Media that fluoresce under red light get the 1064 nm variant at 800 mW instead.
A bandpass filter cleans the diode's output down to the excitation line. A dichroic plate then splits the paths: excitation goes out to the process, returning light comes back on its own route.
A dual fibre carries light to an immersion probe mounted in the reactor loop or pipeline. The fibre run lets the analyzer sit where a person can reach it while the measurement happens where the chemistry is. No sample is drawn, no conditioning skid is built, nothing is consumed.
Most photons scatter elastically and carry no information. A small fraction returns shifted in energy by the molecular vibrations it met. That shift is the measurement.
By examining scattering — and a very particular kind, inelastic Raman scattering — we get information about chemical composition, internal structure and chemical properties. dr Bartosz Kawa, CTO · Główny Mechanik
A notch filter blocks the elastic line that would swamp everything else. A grating spreads what remains onto a back-thinned, TEC-cooled CCD — every pixel a narrow slice of the spectrum.
Intensity against Raman shift. The bands and their ratios identify what is in the medium; their size says how much. A few hundred milliseconds of integration, repeated as often as the process needs it.
The light trace — the Raman spectrum — is as unique to a substance as a fingerprint. Industry Alarm, 2026
Spectrally OS runs a model built for this chemistry: a CNN reading the full spectrum together with classical chemometric algorithms. Out comes what the process engineer actually asked for — concentrations, an acid value, a viscosity, a pass or a fail — on PROFIBUS, PROFINET or Modbus.
Every design decision below exists because a plant broke a simpler one. Ranked by how much of it competitors cannot copy — the top four came out of a value–uniqueness workshop with our COO in April 2026.
The probe carries its own reference signal, so the instrument calibrates itself and reports its own faults. If anything drifts in the optical path, the analyzer says so before a batch does.
Neural models handle non-linear behaviour and multi-component mixtures, including parameters that are not concentrations — viscosity, for instance.
Probe, spectral range, acquisition time and model are tuned to one chemistry and one process, rather than shipped as a catalogue default.
Recipes change, feedstock changes, the process drifts. Model updates for the first twelve months are part of the agreement, not a change order. (Hardware warranty is a separate term.)
In fouling media the probe retracts, is rinsed, and returns to the measurement position without interrupting continuous operation.
Measurement and logging run on a local database. No cloud dependency, no plant data leaving the site.
Three to five real samples, a preliminary model and a written report — before anyone signs for an instrument. Pilot terms available.
Where Raman scattering is too weak or the bands are silent, the platform reaches for another technique. The method follows the chemistry.
Raman does not replace HPLC as an analytical reference. It replaces waiting for one.
| Property | Raman | NIR | FT-IR | HPLC | Titration |
|---|---|---|---|---|---|
| Inline in water | Excellent | Difficult | Poor — water absorbs | — | — |
| Time to result | Seconds | Seconds | Seconds | Minutes to hours | Minutes |
| Sampling needed | No | No | No | Yes | Yes |
| Multi-component mixtures | Very good | Good | Good | Excellent | Weak |
| Substance identification | Highly specific | Less specific | Specific | Specific | Indirect |
| Reagents / lab waste | None | None | None | Yes | Yes |
The most valuable data is the data that arrives before the problem grows. Inline analyzers change the logic of the whole process. Robert Stachurski, CEO · Polska Chemia
From acquisition to a number the control system can act on — against hours or days of waiting for a laboratory result. Universal to the platform, independent of industry.
Water is a weak Raman scatterer, so aqueous streams stay measurable — where FT-IR loses the band region to water absorption. Raman bands are also more specific than NIR's overlapping ones, which matters when several components move at once. Where Raman is genuinely weak, NIR or SERS is used instead.
Fluorescence under 785 nm excitation is the usual reason to move to the 1064 nm configuration, 800 mW. The rest of the chain is unchanged.
The probe sits in a retractable holder: it withdraws from the circulation loop, is rinsed with solvent, and returns to the measurement position. Continuous measurement is not interrupted for cleaning.
The probe carries an integrated reference; every measurement is validated against it and disturbances are compensated automatically. Self-diagnostic routines detect contamination and flag deviations to the operator.
“Developing algorithms for a specific production line keeps measurement uncertainty at the level of fractions of a percent. The probe design uses a reference measurement — if anything happens in the optical system, the device detects it immediately.” — dr Bartosz Kawa, CTO
Two jobs: turn a spectrum into numbers (multi-parameter regression), and classify — pass/fail, anomaly, off-spec feedstock. Both run on a deliberate mix of CNN and classical chemometric algorithms. How the models are built.
Into the plant's own layer: PLC, DCS, MES, SCADA, LIMS, over PROFIBUS, PROFINET or Modbus. Data is stored locally; measurement and logging continue without a network connection. How the analyser is connected to a line · integration and rollout.
A one-hour workshop to find the measurement points, one to two weeks for the feasibility study, then months for installation, model calibration and validation on production data. Three to five and a half months is the usual span. The four stages.
Three to five samples from the process, a preliminary model, and a written feasibility report that stays with you whatever you decide next.
The same signal chain and the same model stack, packaged for where the measurement has to happen — a bench, a warehouse gate, a reactor loop. A model built on one instrument runs on the next without being rewritten.
Scroll
Most deployments start in a laboratory and end in the line. That path only works if the analyser that validates a method and the analyser that runs it every day share one architecture.
A benchtop analyser for identification and material characterisation, including through transparent packaging. It is where a measurement method is developed and validated before anything is mounted on a line.
LAB+ adds spectral library matching against more than 28 000 reference spectra, proprietary and public. Useful when the question is “what is this” rather than “how much of it is there”.
Incoming goods, field diagnostics, a second opinion on a suspect drum. Built-in models and libraries give a pass or a fail at the gate, through sealed packaging, without opening the container or sending anything to a laboratory.
Continuous measurement in the installation itself, feeding the control layer directly. Two measurement channels as standard, more on request; explosion-hazard zones depending on configuration, where the laser is limited to 30 mW.
An immersion probe for liquids, suspensions and slurries in reactors, pipelines and tanks. In fouling media it works with a retractable self-cleaning module — retract, rinse, return — so continuous measurement is never stopped to clean an optical window.
The models, the operator interface, the audit trail and the integration live here — the same layer whichever body the instrument came in. Model updates roll out without stopping the analyser or the line, and measurement continues when the network does not.
The family shares one signal architecture and one chemometric stack, so calibration work does not get thrown away at the step where it usually does — the move from laboratory to production.
A method developed on X1 LAB migrates to X1 INLINE without being rebuilt: same optical geometry, same detector behaviour, same model format in Spectrally OS. The feasibility study on three to five samples is therefore not a throwaway exercise — it is the first calibration of the production system.
Retractex, the self-cleaning retraction module used with X1 PROBE, is a component in that chain rather than a product of its own.
How the cabinet and probe are actually connected to a running line, and how the laboratory path feeds it: inline and laboratory.
From the 2026 product catalogue. Where a value depends on configuration, the configuration is named.
| Parameter | X1 LAB / LAB+ | X1 PORTABLE | X1 INLINE |
|---|---|---|---|
| Excitation wavelength | 785 / 1064 nm | 785 / 1064 nm | 785 / 1064 nm |
| Laser power | 600 mW @ 785 · 800 mW @ 1064 | 600 mW @ 785 · 800 mW @ 1064 | as LAB; 30 mW for ATEX |
| Spectral range | 300–1650 or 300–3500 cm⁻¹ | 300–1650 or 300–3500 cm⁻¹ | 300–1650 or 300–3500 cm⁻¹ |
| Spectral resolution | 8 cm⁻¹ | 8 cm⁻¹ | 8 cm⁻¹ |
| Acquisition time | 5–300 s | 5–300 s · integration 0.01–300 s | 5–300 s |
| Detector | CCD, back-thinned, TEC-cooled | CCD, back-thinned, TEC-cooled | CCD, back-thinned, TEC-cooled |
| Auto-calibration | — | reference integrated in the probe | reference integrated in the probe |
| Signal-to-noise | — | 547 | — |
| Wavelength stability | — | 0.01 nm/°C | — |
| Ingress protection | IP20 | IP54 | probe IP67 |
| Industrial protocols | USB | — | PROFIBUS · PROFINET · Modbus · GSM |
| Measurement channels | — | — | up to 2, expandable on request |
| Sample handling | vials, quartz cuvettes, carousel up to 25 | through transparent packaging | in situ, no sampling |
| Spectral library | LAB+: 28 000+ | built-in libraries | models + libraries |
| Power | 230 V AC · 250 W | 230 V AC · 200 W | 230 V AC · 300 W |
| Warm-up | 30 min | 30 min | 30 min |
Wetted materials: 316L stainless steel, fused silica, chemically resistant epoxy. Housing in 316L or POM-C, POM-C ESD for the ATEX version.
Reference spectra behind library matching on X1 LAB+ — proprietary plus public collections. The number that decides whether an unknown drum can be identified at all.
A one-hour workshop usually settles it: where the measurement has to sit, what the medium does to an optical window, and what the control system needs to receive.
One goes through a nozzle on a running transfer line, with the analyser bolted to a wall nearby. The other goes through a vial on a bench. They use the same optics and the same models, which is why a method proven in the laboratory can be mounted on the line without starting over.
Scroll
The measurement point is a nozzle. What arrives at the control room is a concentration, an acid value or a viscosity — not a spectrum, and not a sample bottle.
A transfer line, a reactor circulation loop, a pipeline between two units. Nothing is diverted and nothing is stopped: the measurement is taken in the flowing product at line pressure and line temperature.
The probe goes in through a standard process connection — a flange, a threaded boss or a Tri-Clamp — and reaches into the flow. Wetted parts are 316L stainless steel and fused silica, so the material question is usually settled before the meeting ends.
In media that coat an optical window, the probe sits in a retractable holder: it withdraws behind an isolation valve, is rinsed with solvent, and returns to the measuring position. Where insertion into the main line is awkward, the same probe goes into a slipstream loop instead.
The probe is automatically withdrawn from the circulation loop, rinsed with a stream of solvent, and returns to the measurement position. Gekko Photonics · Główny Mechanik
Only the probe is in the process. Excitation light travels out and collected light comes back on a dual fibre, so the analyser cabinet is mounted where a technician can actually reach it. The probe is rated for ATEX and IECEx Zone 0, with laser power limited to 30 mW in that configuration.
Spectrally OS runs the model and passes the value into the control layer over PROFIBUS, PROFINET or Modbus — one to two measurement channels per cabinet as standard. Operators see a number and a status, not a spectrum, and the batch record gets the same value on its way to MES or LIMS.
Not every measurement belongs on a line, and no model starts there. The bench path answers a different question — and it is where the method that later runs inline is built and validated.
The bench path starts with material in hand: a process sample, a retained sample, or an incoming container that has not been opened. Measurement is non-destructive and consumes nothing, so the same material can still go to the reference method afterwards.
X1 LAB takes a carousel of up to twenty-five samples and returns a result in seconds per measurement. This is the analyser that sits near production for at-line checks and in the laboratory for R&D and quality control.
With library matching against more than 28 000 reference spectra, an incoming batch is verified at the gate rather than a week later. The question here is identity and conformity, not process control — a different question from the one the inline instrument answers.
Same optical geometry, same detector behaviour, same model format in Spectrally OS. Calibration work done on the bench during feasibility becomes the first calibration of the production system rather than a discarded pilot.
The three are not competing options. Most plants end up with a bench instrument for method work and incoming goods, and inline units where a decision has to be made while the process is still running.
| X1 INLINE + X1 PROBE | X1 LAB / LAB+ | X1 PORTABLE | |
|---|---|---|---|
| Where it sits | nozzle on the line, reactor or slipstream loop | laboratory bench or at-line, near production | carried — gate, field, installation |
| Sample | none drawn | vial, cuvette or sealed container | container, measured through packaging |
| Question answered | is the process where it should be, right now | what is this, and does it meet specification | can this delivery be accepted |
| Result reaches | PLC / DCS, then MES / LIMS | report, LIMS, model development | operator on the spot |
| Mechanical scope | process connection + cabinet + fibre run | bench space and a socket | none |
| Protection | probe IP67, ATEX / IECEx Zone 0 | IP20 | IP54 |
| Channels | up to 2 per cabinet, expandable | carousel up to 25 samples | one at a time |
The mechanical and electrical questions are short. The ones that decide the outcome are about the medium.
Composition range, temperature, pressure, pH, solids, and whether it coats a window. Fouling behaviour decides whether a retractable holder is needed.
Which nozzle, what connection standard, how much insertion length, and whether the flow at that point is representative of the batch.
Zone classification, cable routing to a location where the cabinet can be serviced, and mains supply.
Which protocol the plant speaks, which tag the value should land on, and who owns the alarm when the analyser reports a deviation. The value is only useful if somebody is accountable for acting on it.
The current method and its results for the same material — without them there is nothing to calibrate against or validate with.
Where 21 CFR Part 11 applies, it is handled as a project requirement rather than a checkbox.
Mounting time reported for the water and wastewater segment — mains or battery supply, no reagents and no wear parts. Process installations in chemical plants are scoped individually; this figure belongs to that segment.
A P&ID extract, the connection standard at the measurement point, and three to five samples are enough to say whether this works on your line.
From the tanker at the gate to the effluent leaving the site. Most plants do not need all eight — but seeing them together is the fastest way to work out which two or three would change a decision that is currently made on faith.
Scroll
The same substance is measured as a delivery, as a stored raw material, as a reacting mixture, as a finished product and finally as whatever leaves in the water. Different instrument, different question, one analytics layer.
The delivery is measured at the bay — through the sampling port or straight through a sealed transparent container, without opening it. The question is narrow and expensive: is this the material on the certificate. A wrong tanker discharged into a storage tank contaminates everything downstream of it.
Incoming quality control on stock: identity checks on drums and IBCs, re-checks after long storage, and verification when a label is ambiguous. Built-in spectral libraries handle identification; where the answer needs to be quantitative it goes to the bench.
The benchtop analyser does two jobs: routine release measurements with a carousel of up to twenty-five samples, and the development work that turns a chemistry into a validated model. Library matching against more than 28 000 reference spectra answers “what is this”; the models answer “how much”.
At-line means at the line: the analyser stands a few steps from the vessel and the operator measures without walking to the laboratory. The charge, the premix, the solvent drum about to go in — verified in the minutes that are actually available before a batch is started.
An immersion probe in the reactor or its circulation loop, an analyser cabinet on the wall, and a value on PROFIBUS every few seconds. This is where the endpoint gets decided, where a deviation shows up in hour two instead of at the end, and where the payback is calculated from.
The blend that leaves the vessel is checked at-line before filling, while rework is still cheap and the material is still in a tank rather than in a thousand containers. The measurement is the same one the laboratory would run, taken where the decision happens.
A final identity and conformity check on what is about to go on a truck, with the result written into the batch record. It is the cheapest place in the plant to catch a mislabelled pallet, and the most expensive one to miss it.
Continuous monitoring of the works outfall — the first Polish deployment of this kind, with MPWiK Wrocław, watches for petroleum-derived substances in wastewater. Water is a weak Raman scatterer, which is exactly why the substances in it stay visible. Continuous detection of microplastics and PFAS is in development, using SERS.
Read the third column first. If nobody is currently making that decision — or it is being made on a number that arrives too late — that station is worth a feasibility study. The rest are not.
| # | Where | Instrument | Decision it changes |
|---|---|---|---|
| 01 | Goods-in bay | X1 PORTABLE | Accept or reject a delivery before it is discharged |
| 02 | Raw material store | X1 PORTABLE | Release stock to production, or quarantine it |
| 03 | Laboratory | X1 LAB / LAB+ | Release a batch; build and validate the method |
| 04 | At-line, before the reactor | X1 PORTABLE | Start the batch, or correct the charge first |
| 05 | Reactor and circulation loop | X1 INLINE + X1 PROBE | Hold, dose, or call the endpoint — while it still matters |
| 06 | After the reactor, before filling | X1 PORTABLE | Pack it, or rework it while it is still in the tank |
| 07 | Finished goods | X1 PORTABLE | Dispatch, with the evidence in the batch record |
| 08 | Works outfall | X1 INLINE | Report, or intervene before a discharge becomes an incident |
Every station in the map runs the same signal architecture and the same model format in Spectrally OS. A model calibrated in the laboratory at point 03 is the model running at point 05, and the identity library used at point 01 is the one behind point 07. That is the part a plant cannot assemble from three vendors.
Where to start is a question for the workshop: the four stages. How the reactor point is physically connected: inline and laboratory.
Of the substances found in the chemical industry are detectable with the platform — the figure the company states publicly, and the reason one measurement architecture can cover a whole site rather than a single unit.
Two patterns, from the deployments so far.
When the pain is variability and waste inside the process. A resin producer running inline measurement for six months reported raw-material losses down 10 %, waste down 12 % and batch variability held below 1.5 %. Payback in that segment lands at 6–10 months.
Those figures belong to the resin segment. Water payback is 18–24 months; cosmetics savings were measured as an annual figure rather than a payback. Numbers do not travel between segments.
When the pain is what arrives on site. A portable unit at goods-in gives pass/fail in minutes through sealed packaging, removes the sampling delay, and needs no installation work at all — which makes it the least disruptive way to find out whether Raman sees your chemistry.
It is also the usual first exposure: a demonstration on your own material before any conversation about process installation.
Where the losses are, which decisions currently wait for a laboratory result, what the medium does to an optical window, and what the control system needs to receive. One hour is usually enough to mark two or three points on the plan — and to strike the ones that would only add instruments.
Send a block diagram or a P&ID extract and we will mark the points where a measurement would change a decision — and the ones where it would only add an instrument.
A detector returns a noisy vector with a fluorescence slope and the occasional cosmic ray. Turning that into a number a process engineer can act on is a modelling problem — and it is where most of our engineering sits.
Scroll
Five stages. The first two are housekeeping that decides whether anything downstream can work; the middle two are the model itself; the last one is what keeps a number trustworthy six months after commissioning.
A vector of pixel counts carrying the Raman bands, a fluorescence slope from the medium, shot noise, and now and then a cosmic ray that looks exactly like a very sharp peak. None of it is a concentration yet.
Baseline correction, spike rejection, normalisation. Every measurement is validated against the probe's built-in reference, and disturbances are compensated automatically rather than left for the model to guess at.
Every measurement is validated and disturbances are compensated automatically. Self-diagnostic algorithms detect contamination and inform the operator about deviations. dr inż. Maciej Jaworski · PIPC
A convolutional network takes the full vector rather than a handful of chosen peak windows. That is what makes non-linear behaviour and heavily overlapping bands tractable — and why parameters that are not concentrations at all, viscosity among them, can be predicted.
Classical algorithms are chosen where they win: small calibration sets, interpretable loadings, fast validation against a reference method. The mix of both families is deliberate, and it generalises better than either one on its own.
The value leaves with its error bar, an agreement check against the reference method, and the analyser's own diagnosis of the optical path. Recipes drift; models are updated for the first twelve months under the contract, and re-calibrated when the process changes.
Purely classical models are cheap to validate and blind to non-linearity. Purely neural models absorb non-linearity and demand data. Real process chemistry needs both, chosen case by case.
Full-spectrum input, non-linear response, overlapping bands, multi-component mixtures, and derived properties such as viscosity that no single band encodes.
Compact calibrations from a limited number of samples, loadings a chemist can inspect and argue with, and quick validation against titration or HPLC.
Faster, more accurate models that generalise better than a single-family approach. In a feasibility study on alkyd resins, a CNN over the whole spectrum outperformed linear regression on the same data, reaching R² above 0.99 for both acid value and viscosity.
Calibration, validation, maintenance, re-calibration. A model that is never touched again is a model that quietly stops being right.
Starts from feasibility samples — usually the client's R&D material plus Gekko laboratory work — and the reference values that go with them.
Predictions checked against the established reference method on the client's own process, not on a public data set.
Further training on production data once the analyser is running, which is when the awkward cases show up.
Triggered by what actually changes in a plant: a new recipe, a different feedstock supplier, modified process parameters. Model updates are covered for the first twelve months under the agreement, and they roll out without stopping the analyser or the line. The hardware warranty period is a separate commercial term.
Figures from feasibility studies on real process samples. Each one is a specific matrix and a specific reference method — not a platform-wide claim.
Full matrices, bands and methods: feasibility results.
Prediction uncertainty for alkyd resin viscosity during synthesis — a rheological property, read from a Raman spectrum. This is the number that tends to end the argument about whether Raman only measures concentrations.
Non-linear response, overlapping bands, a property that is not a concentration. Those are the cases the hybrid stack exists for.
Every application below started as a feasibility study on real process samples — a named matrix, a named band, a number against a reference method. Nothing here is a brochure claim about what Raman can do in principle.
Scroll
A Raman spectrum is only useful if a specific band moves with a specific quantity. These four are documented on real plant material — and two of them sit close enough together to explain why a model, not a peak-pick, does the work.
In an aqueous sodium-salt solution, the band at about 799 cm⁻¹ tracked the target product as it formed. Across the reaction stages the model followed it from roughly 90 % to above 99 %, which is precisely the region where an endpoint decision gets made.
During surfactant synthesis, the whole 820–1140 cm⁻¹ group grows as ether bonds form in the ethoxylated and propoxylated products. Conversion is followed continuously, with no sampling and no laboratory step in the loop.
Nitrate nitrogen in liquid fertiliser streams sits at about 1044 cm⁻¹; total nitrogen in granulated product with an organic coating at about 1050 cm⁻¹. With 8 cm⁻¹ resolution those two are practically the same channel — which is the clearest argument for letting a model interpret the spectrum rather than reading a peak height off a chart.
At about 1050 cm⁻¹ an unwanted isomer in an FMCG synthesis was tracked down from roughly 8.5 % to trace level. Around 780 cm⁻¹, marker bands and intensity ratios separated lighter from heavier refinery fractions. Same neighbourhood of the spectrum, different chemistry, different model.
The difference an inline analyser makes is not really accuracy. It is how many times per batch anybody gets to look.
A sample is drawn, walked to the laboratory, prepared, analysed. The number that comes back describes a batch that has already been made. If it is out of specification, the decision is about what to do with the product, not about the process.
An inline analyser returns a few to a dozen measurements per minute — over a thousand points in a multi-hour process. The trend becomes an object you can steer by rather than a pair of end points.
A few to a dozen measurements per minute — over a thousand measurement points in a process lasting several hours. Główny Mechanik, 2026
A drift that a laboratory result would have revealed at the end is visible in hour two, when there is still process time left to react.
The most valuable data is the data that arrives before the problem grows. Robert Stachurski, CEO · Polska Chemia
At a European synthetic-resin producer running inline measurement for six months: raw-material losses down 10 %, waste down 12 %, batch-to-batch variability held below 1.5 %. Those figures belong to that segment and that installation — resin economics are not water economics.
Feasibility studies from the 2026 catalogue, organised the way an engineer arrives at them: I need to follow a reaction, I need to know what is in it, or I need to release a batch.
Reaction progress, material transformation, stream identity — measured where it happens.
What else is in there, and how much of it — in matrices that defeat single-parameter sensors.
Release decisions that used to wait for titration or HPLC. The two alkyd figures are the ones worth arguing about: an acid value without titration, and a rheological property read from a spectrum.
Water and environment work includes the first Polish deployment of continuous wastewater monitoring for petroleum-derived substances, with MPWiK Wrocław. Microplastics and PFAS detection is in development, using SERS.
Send three to five samples from the actual process. The feasibility report says what is measurable, at what uncertainty, and what it would take.
Four stages from a first conversation to a system the plant relies on. The second one — a feasibility study on real samples — exists so that the capital decision is made with evidence rather than optimism.
Scroll
The sequence is the same whether the answer turns out to be an analyser, a different measurement point, or a process change that needs no instrument at all.
About an hour, online or on site. Where the losses come from, which deviations hurt, what a bad batch costs, and which measurement points would actually change a decision. No instrument has been chosen at this stage.
Real process material, measured, modelled, and compared against the reference method. The report states what is measurable, at what uncertainty, and what it would take. It stays with you whether or not anything is ordered.
Measurement configuration, points, integration design, and a payback calculation built on the plant's own cost of waste, rework and downtime. Sometimes the honest recommendation is a process change rather than an instrument.
Installation, model calibration on production data, validation, operator training, and continued model development as the process evolves. Model updates are covered for the first twelve months under the agreement.
A measurement that stops at a screen in a cabinet changes nothing. The result has to arrive in the layer where the plant already makes decisions.
No sample line, no conditioning skid, nothing to consume. The probe sits in the reactor loop or the pipeline and looks at the process itself.
Model inference, pass/fail classification, deviation alarms, and an audit trail of which model version produced which number. Everything is written to a local database on the plant's own hardware.
The result appears where operators and control loops already look, over PROFIBUS, PROFINET or Modbus. That is the difference between an instrument that reports and an instrument that participates.
Spectrally does not only measure — it helps to steer. Robert Stachurski, CEO · PIPC
Values flow on to MES, SCADA or LIMS as release documentation and compliance evidence. When the network is unavailable, measurement and logging keep running locally and synchronise afterwards.
Specification and price are properties of the platform — quote them anywhere. Payback, savings and reduction percentages are properties of a segment and an installation. Resin economics are not water economics, and we will not pretend otherwise on a sales call.
How much more the global instrument houses ask for a comparable process-analytics installation. The capital range above is the whole reason a mid-size chemical plant can consider inline Raman at all.
| Segment | Payback | Savings | Other measured effects |
|---|---|---|---|
| Resins — continuous chemistry | 6–10 months | +EUR 180 000 / year | OEE improvement, batch repeatability |
| Resins — European producer, 6-month observation | not measured | — | losses −10 %, waste −12 %, variability <1.5 % |
| Cosmetics — emulsions | not measured | +EUR 100 000 / year | mixing time −12 %, rework −8 % |
| Cosmetics — incoming quality control | not measured | qualitative | pass/fail in minutes, sampling delay removed |
| Water and wastewater | 18–24 months | — | ppm accuracy; first Polish water deployment |
| Nitrogen fertilisers — UAN / DEF (illustrative) | est. 8–14 months | — | off-spec −50–60 % (est.), DEF first-pass ~95 % (est.) |
| Specialty fertilisers — ATS (illustrative) | est. 8–14 months | — | off-spec −45–55 % (est.), QC workload −30–40 % (est.) |
| Silicones (illustrative) | not measured | — | off-spec −6–9 % (est.), reactor time −5–7 % (est.) |
The shortest path from “we think this might be measurable” to a documented answer.
Then that is what the report says. The platform includes SERS for analytes below plain Raman's sensitivity and NIR where Raman bands are weak — and there is a separate, technology-neutral audit of a plant's analytical quality-control system, priced as a fixed fee, whose recommendation is sometimes a process change and no new instrument at all.
One sample, four analytical methods, one stopwatch. The point is not that Raman is better than a titration — it is what each answer costs by the time it arrives.
A process sample leaves the line. Whichever method takes it, the chemistry is identical — what differs is how much has to happen before anybody can act on a number.
Reliable and well understood, and it consumes the sample along with reagents that then have to be disposed of. Weak on multi-component mixtures, and it cannot run continuously.
The most specific answer on this chart, and the method our own models are validated against. It also needs sampling, preparation and a queue. The delay is what we compete with — not the accuracy.
Also fast, and useful where Raman bands are weak — which is why NIR sits inside our own platform at about 5 % of the portfolio. Its bands overlap more, so it struggles when several components move at once.
Molecular vibrations are specific to the substance, water barely scatters, and no sample has to be drawn. About five seconds to a value, nothing consumed, repeated as often as the process needs it.
The most common objection we hear is that Raman measures composition, not properties. It is a fair objection, and the answer is a chain of physics you can follow link by link — not a black box.
At the start of a polycondensation the vessel holds mostly free monomers. Nothing about that mixture is viscous yet, and no single chemical concentration would tell you what the batch will pour like at the end.
As the reaction proceeds, monomers link into chains and the distribution of chain lengths shifts. Viscosity follows that distribution — and so do the molecular vibrations, because the bonds themselves are what changed.
Many adhesives and resins are made by polycondensation, and some polymerise while curing. That is exactly what we see — the change in chemical bonds. dr Bartosz Kawa, CTO · Główny Mechanik
Overlay the spectrum from the start of the batch on one taken later and the peak positions barely move — the ratios between them do. That shift is small, systematic, and far too subtle to read off a chart by eye.
A CNN reads every channel rather than a few chosen peak windows, with classical chemometrics alongside it where a compact, inspectable calibration wins. The model is fitted on this plant’s own samples against its own reference values.
In a documented feasibility study on alkyd resins the model reached R² above 0.99 for viscosity through the whole synthesis, at a prediction uncertainty of ±0.25 Pa·s — and the acid value came out of the same spectrum at about ±0.2 mg KOH/g, without a titration.
A production sample is a mixture, and the bands of its components sit on top of one another. This is the case where a single-parameter sensor gives up and a model does not.
A pure substance has a clean fingerprint: a handful of bands in known positions with known relative intensities. Measured alone, in a laboratory, it is a solved problem.
Add three more and the bands start sharing space. Some overlap almost completely, and the intensity at any given wavenumber now belongs to more than one substance at once.
The instrument never sees the components separately. It sees their sum, plus whatever the matrix adds. Reading a peak height off this curve would give you a number that belongs to no single component.
Calibration uses real process mixtures with reference values, not a library of pure substances. The model learns how the overlap behaves as the composition moves — including the non-linear part, which is where classical single-band approaches break down.
This is the everyday case in our feasibility work: silicone contamination in PA66 recyclate quantified across a 0.3–1.0 % calibration range, and ammonium nitrogen predicted in industrial water even though its signal overlaps the water bands.
Illegal discharges are not steady. They are short, they are timed badly on purpose, and hydrocarbons travel on the surface rather than mixing in. Three buckets a day is not a monitoring strategy — it is a lottery.
Petroleum-derived hydrocarbons are less dense than water, so they move along the surface rather than mixing through the profile. Where the sample is taken from decides what the laboratory will find — before any analysis happens.
Petroleum hydrocarbons, with a density lower than water, flow along the surface; illegal discharges are intermittent — which is why sampling does not catch them. dr Karolina Orłowska, CSO · Industry Alarm
A day with three events, each lasting minutes. Nothing about them is steady, and none of them announces itself. In the aggregate figures at the end of the month they may not appear at all.
Scheduled sampling at 06:00, 12:00 and 18:00 lands in the quiet windows between the events. The results come back clean, the file says compliant, and the discharge that happened at 04:04 is not in any record.
An analyser in the channel measures the flow itself, without drawing samples and without reagents. Water is a weak Raman scatterer, so the substances carried in it stay visible rather than being drowned by the matrix.
Water is invisible to us — we focus on the substances in it, without distorting the picture. dr Bartosz Kawa, CTO · INNPoland
With MPWiK Wrocław — continuous monitoring of wastewater for petroleum-derived substances, mounted in a day, no reagents and no wear parts. Continuous detection of microplastics and PFAS is in development, using SERS.
An hour on your process comes first. Everything after that — feasibility, architecture, payback — follows from what that hour finds.