One reactor, one measurement point — this is a standard that works well for simple batch syntheses, but becomes insufficient when the process has several reaction zones, several parallel lines, or when we want to simultaneously monitor the raw material, mixing, and endpoint. Multi-probe architecture In process Raman, this is the answer to such requirements: one or several analyzers servicing multiple probes, synchronized with the DCS and operating within the same chemometric layer. At Gekko Photonics, we design and manufacture process Raman analyzers in Poland, in inline, laboratory, and portable variants — multi-probe capability is a standard element of our Spectrally X1 INLINE offering, not an exotic add-on.

This article organizes when it is worth using a multi-probe process analyzer, what architectures are available, where the real design trade-offs lie, and what to consider before configuring a system. The reference points are typical scenarios from process chemistry, petrochemistry and resin production — without being tied to a specific industry.

When one probe is not enough

The decision for a multi-probe architecture is almost always a derivative of the process topology, not an ambition to „measure more because we can.” The most common reasons why a process engineer needs more than one Raman measurement point:

• Reactor with several zones — a distillation column where we want to see the composition on several trays; a reactor with a temperature gradient and different composition at the bottom/top; a crystallizer with a dissolution point and a nucleation point.
• Parallel process lines — several identical batch reactors, several filters or crystallizers. A dedicated analyzer for each would be wasteful, while one central analyzer with several probes provides equivalent coverage.
• Input, middle, output — measurement of raw material before the reactor (incoming QC, batch verification), during the reaction, and on the final product. Three probes → full quality audit from gate to tanking.
• Redundancy — in critical applications, two parallel measurements of the same point (from two probes) allow detection of drift in one of them and transition to degraded mode without stopping the process.
• Calibration transfer — parallel measurement of the same sample with a reference (calibration) probe and a process probe, to maintain the credibility of the chemometric model over time.

All these scenarios share one characteristic: we do not want to purchase and maintain N analyzers if they can be serviced centrally. CAPEX cost, maintenance cost, model consistency, and a unified data layer all argue for a single system controlling multiple probes.

Multi-probe architectures — an overview

Application engineers distinguish three main approaches to multi-channel Raman measurement in process. Each has its place; the choice is a derivative of the required measurement cadence, number of channels, fiber optic length, and tolerance for signal loss.

1. Time multiplexing — one spectrometer, fiber optic switch

The most widespread architecture. A central analyzer with one laser and one CCD detector is connected to a fiber optic multiplexer (most often piezoelectric or MEMS), which sequentially directs the laser beam and collected spectrum to successive probes. Switching times in the process class are typically below 150 ms — for Raman measurements, where acquisition time is several to several tens of seconds, the switching cost itself is negligible.

Pros: one set of optics, one set of wavelength calibrations, one service point. Cons: measurement cycle = (acquisition + warming) × N, so for 4 probes and 30 s acquisition, a full iteration takes ~2 minutes. If the process requires a response faster than the cycle, one must either reduce the number of probes, shorten the integration time, or switch to a different architecture.

2. Spatial multiplexing — several detection channels in parallel

The second approach is a spectrometer with a detection system that simultaneously images several fiber optic inputs onto different areas of the CCD matrix. The laser beam can be passively split (splitter) or have several sources, and each probe has its own „path” on the detector.

Pros: parallel acquisition, no switching losses, short cycle. Cons: lower laser power per probe (with passive splitter) or need for several lasers (more expensive), more complex calibration, limited number of channels (usually 2–4), greater sensitivity to cross-talk between channels. Makes sense when very fast, simultaneous measurement is needed at 2–3 points (e.g., inlet/outlet of a continuous reactor).

3. Multiple independent analyzers + software layer

Distributed architecture: each location has its own analyzer, and integration occurs at the software layer (local SCADA, MES, or dedicated chemometric platform). Each analyzer operates independently — full laser power, full measurement frequency, redundancy in case of failure of one.

Pros: maximum performance, no fiber optic length limitations (each analyzer is located close to its probe), independent calibration paths. Cons: higher CAPEX, more service points, need for model synchronization and versioning at the software layer. This approach dominates in large plants with geographically dispersed installations — e.g., several distillation units in petrochemistry.

Design trade-offs — what really hurts

Each of the three architectures requires a trade-off between cost, cadence, and reliability. From our engineering practice, the following issues recur most often:

Fiber optic length

785 nm Raman spectroscopy tolerates long fiber optic runs well — typical installations work with lengths up to 100 m without significant signal loss, maintaining a signal-to-noise ratio in the process class. This is a real advantage over other PAT techniques, where the spectrometer-sample distance is severely limited. In practice, this means one analyzer in a control cabinet can service probes scattered around the perimeter of a batch reactor, in a compact continuous line, or on column walls.

Above 100 m, effects of fiber optic intrinsic fluorescence, losses at connections, and the need for fibers with higher dopant purity begin to matter. For such applications, a distributed architecture (variant 3) is more sensible than forcing extreme lengths on a single analyzer.

Calibration per channel

Each channel (probe + fiber optic + media interface) has its own optical throughput and spectral response. A chemometric model built on probe A will not transfer 1:1 to probe B without calibration transfer. In practice, we manage this through:

• Calibration of each probe against the same reference sample during startup.
• Use of an internal reference in the probe (standard band, e.g., from the window material), which allows automatic drift correction per channel.
• Periodic validation procedures in the service cycle — checking each probe's response on a single calibration sample.

Synchronization with DCS and control loop

In a time-mux architecture, the control loop must know which probe the currently reported composition refers to. Addressing via PROFIBUS/PROFINET requires designing the data structure: one set of variables per channel, timestamp, and probe identifier. Without this, the operator controls the process based on an averaged signal — which is usually worse than having no measurement.

Probes in difficult media

Multi-probe capability increases the total number of points where product can deposit on the probe window. For viscous media, polymers, resins, and deposits, we use Retractex self-cleaning modules — the probe is periodically retracted, the window is flushed with a cleaning medium or inert gas, and then returns to the measurement position. In installations with 3–4 probes in difficult media, such a function is practically a prerequisite for stable measurement over a monthly horizon.

Designing a multi-probe system — checklist

Before ordering an analyzer and probes, we typically go through a list of questions with the client to organize the project:

1. How many measurement points and are all critical for the control loop, or do some have a monitoring role? Critical points usually go to spatial multiplexing or have a dedicated analyzer; monitoring points — time-mux is sufficient.
2. Required measurement cadence per point. A 30-second control loop requires a different architecture than a ten-minute quality report.
3. Physical distances between probes and the analyzer cabinet. Above 100 m — consider two analyzers instead of one.
4. Probe operating environment — explosion hazard zone, temperature, pressure, medium aggressiveness. ATEX configuration is available as a variant of the Spectrally X1 INLINE with a 30 mW laser; we standardly work with 600 mW outside the zone.
5. Need for self-cleaning — which probes will be located in sticky/greasy media that foul the window? There, Retractex.
6. Network architecture — whether the plant already has PROFIBUS/PROFINET, or whether we need GSM as a backup channel (e.g., for alarm signaling to maintenance).
7. Consistency of chemometric models — whether all probes measure the same analyte (one model for all) or different analytes (separate models per probe).
8. Validation plan — who verifies, how often, and on what samples that each channel maintains its calibration.

A well-thought-out checklist is a stronger argument for selecting a specific architecture than a catalog specification. In our projects, we usually finalize it in a workshop with the client's process and maintenance engineers — this is the moment when it typically becomes clear that, instead of 6 probes in one multiplexer, a configuration of 2 + 2 + 2 on three analyzers with an OS layer as a common dashboard is more sensible.

Multi-channel calibration vs. chemometric model

The most common design misunderstanding is the assumption that one chemometric model will serve all probes „because it's the same analysis.” The reality is that channels differ in throughput by a few percent, which can be significant for a PLS model in a narrow calibration range. Two technical paths we use:

• Common model + calibration transfer — one base model, with an offset/scaling correction per channel determined during the commissioning phase. Used when probes measure the same analyte under the same conditions (e.g., parallel batch reactors).
• Separate models per channel — independent calibrations built during the feasibility phase on samples from each location. Used when probes observe different parts of the process (e.g., raw material vs. post-reaction product), where value ranges and interferences differ.

In both cases, models live on the platform Spectrally OS — with versioning, drift monitoring, and a validation log per channel. This is where multi-probe operation ceases to be a hardware problem and becomes an operational problem that we want solved in software.

Gekko Photonics solutions for multi-probe architectures

In our product family, we build multi-probe systems on three pillars. The Standard Spectrally X1 INLINE supports up to 2 probes on one analyzer (expandable upon request after feasibility), operates with a 785 nm laser and 600 mW power, communicates via PROFIBUS, PROFINET, or GSM, and the fiber optic cable length to each probe can reach 100 m. The TE-cooled back-thinned CCD detector provides stable response across the full 300–1650 cm⁻¹ window with 8 cm⁻¹ resolution — sufficient for most process applications.

For larger installations — four, six, or eight probes — we deploy a distributed system: several X1 INLINE units in separate cabinets, each handling 1–2 probes in close proximity, and the aggregation layer is built by Spectrally OS with a central dashboard, chemometric models (CNN, PLS, PCA), a library of ~28,000 reference spectra, and alarms per channel. This is also the scenario where we implement redundancy — two analyzers on the same critical point with automatic switching in case of signal loss.

We support model validation and calibration transfer with Spectrally X1 LAB — a benchtop analyzer with a 25-sample carousel operating in the same spectral window as the INLINE. This ensures that reference samples measured in QC remain comparable to the measurement in the reactor, and a model developed in the lab transfers to the process without costly recalibration. Where mobile verification is required, Spectrally X1 PORTABLE — in a suitcase, for incoming QC and field reference measurements.

The entire product family is available on the analyzers page, and a practical comparison of laboratory and process configurations is discussed in detail in the article Laboratory vs. Process Raman.

FAQ — frequently asked questions about multi-probe architectures

How many probes can be connected to a single Raman analyzer?

In the standard configuration of the Spectrally X1 INLINE, this is 2 probes with the possibility of expansion. For a larger number of points, we recommend a distributed architecture with several analyzers on the network, managed from the Spectrally OS level. Attempting to connect 8+ probes to a single analyzer via an aggressive multiplexer usually degrades the measurement cycle and complicates service.

Does each probe need a separate chemometric model?

Not always. If the probes measure the same analyte under the same conditions, a common PLS/CNN model with individual calibration transfer correction per channel is sufficient. If the probes observe different parts of the process (e.g., raw material vs. reaction), we usually build separate models. We determine the recommendation during the feasibility phase based on the client's samples.

How long can the fiber optic cable to a probe be in a process architecture?

For 785 nm, typical lengths are up to 100 m without significant loss of signal-to-noise ratio. Beyond this limit, the fiber's own fluorescence and splice losses become significant. In geographically larger installations, it is more sensible to place several analyzers locally than to force extreme lengths on a single one.

Does time-division multiplexing slow down the measurement?

Yes, the measurement cycle increases linearly with the number of probes — for 4 channels and a 30 s acquisition, a full iteration takes ~2 minutes. This is acceptable in most process applications, but for control loops faster than the cycle, we consider spatial multiplexing or a distributed architecture.

Does Gekko Photonics deliver turnkey multi-probe systems?

Yes. We implement the full stack: configuration of probes for the specific chemistry, integration with PROFIBUS/PROFINET, chemometric models in Spectrally OS, validation per channel, and service. A standard multi-probe project is launched in 3–5.5 months from the design workshop — with a feasibility phase on the client's samples as the first step.

Next step — let's discuss your process topology

The choice of multi-probe architecture should not be a catalog decision. At Gekko Photonics, we start with a 30-minute conversation with an application engineer, during which we outline the process topology, the list of measurement points, cadence requirements, and environmental constraints. After signing an NDA, we perform a test measurement on the client's samples within approximately two weeks — this is the moment when it becomes clear whether the chemistry is actually distinguishable by Raman and whether the chosen architecture makes sense. Next, we design the system, select the probes, models, and validation scheme per channel.

Leave your contact on the contact page — we will get back to you with a workshop agenda and a list of information we need from your side to move from concept to feasibility.