sensor systems for bioprocess monitoring

How to Choose Sensor Systems for Bioprocess Monitoring

admin 14 min read
How to Choose Sensor Systems for Bioprocess Monitoring

Begin with the process decision, not the sensor category

Sensor-system selection should begin with the process decision you need the measurement to support. Process Analytical Technology (PAT) uses timely measurements of critical process parameters to design, analyze, and control manufacturing processes. That definition places the process objective before the detector label.

A practical selection sequence is:

  1. Define what you need to know about the culture, media, vessel, or product-related state.
  2. Define what decision the information must support, such as observing a process change, comparing a run with a reference, or informing a control action.
  3. Trace how the sample or signal will reach the detector and how the result will reach the person or system using it.
  4. Set the evidence needed to decide whether the result remains interpretable through the intended process duration.

This sequence prevents a common category error: treating a sensor that produces a frequent signal as automatically suitable for the variable, process condition, and decision you care about.

Write the selection brief in terms of measurements and decisions. State the target variable, the expected range, the required response time, the acceptable sampling or installation approach, the action associated with an alarm or change, and the reference method used for comparison. If you cannot describe those items, postpone technology comparison and clarify the process question first.

Define the measurands the process actually needs

Bioprocess measurands fall into several useful layers. Basic process conditions include pH, ORP or redox potential, dissolved oxygen, dissolved carbon dioxide, temperature, and conductivity. PAT can also address capacitance, off-gas oxygen and carbon dioxide, substrates, and metabolites. Depending on the PAT tool, measurements can provide information about biomass, biomass composition, substrates, and metabolite concentrations.

Turn those layers into an explicit measurement list:

  • Process conditions: Which physical or chemical conditions must remain visible throughout the run?
  • Biomass and viability: Do you need a direct or inferred indication of biomass, viable cell density, or cell state?
  • Substrates and metabolites: Which nutrients, by-products, or metabolic changes matter to the decision?
  • Product-related variables: Do you need information about product titer or another product attribute, or will a laboratory method remain the reference?
  • Context variables: Which temperature, pressure, gas, medium, or solids conditions could affect interpretation?

The list should distinguish a variable you measure directly from a variable you estimate from other signals. That distinction affects calibration, validation, alarm interpretation, and how you describe the result to process and quality personnel.

A comparison of bioprocess technologies lists Raman spectroscopy for glucose, lactate, glutamine, viable cell density, and titer. For the listed Raman entry, the comparison gives a glucose RMSEP of 0.2–0.5 g/L, an installed cost of $70K–$160K, and high calibration effort requiring 5–15 batches for PLS. Treat those figures as the conditions of that comparison, not as a general specification for every Raman system or process.

Define the output before choosing the instrument. A useful output statement names the variable, its units or qualitative state, the expected concentration or operating range, the time resolution required for the decision, and whether the result is a direct measurement or a model estimate. This statement gives you a consistent basis for comparing unlike technologies.

An infographic shows the listed Raman entry with an installed cost of $70-160 K, glucose RMSEP of 0.2-0.5 g/L, and PLS calibration effort of 5-15 batches.
The comparison pairs a broad installed-cost range with glucose accuracy and substantial calibration effort.

Trace how the sample becomes a measurement

Measurement access mode determines where the sample is exposed to the sensor, how much manual handling is involved, and what equipment must be changed. In-line sensors are inserted into the process, on-line measurements use a bypass loop, at-line measurements analyze samples removed from the process nearby, and off-line measurements send a sample to laboratory analysis.

Compare the access modes against the same criteria:

Mode Sample path Handling and equipment implications Boundary to confirm
In-line The sensor is placed directly in the bioreactor or process. The measurement occurs at the process location, so installation, sterilization, material compatibility, and sensor maintenance must fit the vessel. Confirm that the sensor can remain representative and non-contaminating under the actual process conditions.
On-line A sample moves through a bypass, is measured automatically, and may return to the bioreactor. The approach can support constant monitoring and real-time control, but it requires a specifically designed or modified bioreactor and adds setup complexity. Confirm bypass design, sample residence, return path, sterilization, and the effect of the loop on the measured result.
At-line A sample is removed and analyzed nearby; automated sampling can be part of the arrangement. The analyzer may be close to the vessel while the sample still follows a defined transfer and handling path. Confirm sampling frequency, transfer conditions, operator involvement, and how quickly the result can inform the decision.
Off-line A sample is taken from the bioreactor in sterile conditions and analyzed in a laboratory after preparation such as filtration or dilution. The method requires clear SOPs and skilled personnel; preparation or handling problems can reduce result accuracy. Confirm the delay, sample stability, preparation method, and laboratory reference procedure.

For mammalian cell culture, one process-monitoring account describes processes running 10–21 days and states that a metabolic shift detected four hours late can cascade into a batch deviation that off-line sampling would reveal only the next morning. Use that example to test your own decision path: identify the latest acceptable result, then determine whether the access mode and analysis route can meet it.

Off-line laboratory measurements can serve as reference measurements for monitoring and validating in-line or on-line process analyzers. A laboratory method therefore does not automatically compete with an in-process sensor. It may provide the comparison needed to understand whether an in-process signal remains representative.

Match the access method to the vessel and process

Vessel architecture and process duration constrain the access method. For process runs lasting several weeks or even months, in-line and on-line sensors can measure without manual intervention. That capability matters only when the sensor, installation, sterilization state, and maintenance plan remain compatible with the vessel throughout the run.

Single-use bioreactors were described as well established in industry in a 2011 account, while single-use sensing technologies were described as becoming more common for critical parameters. That account identified temperature, pressure, dissolved oxygen, and pH as common single-use measurements. A separate review identifies a need for viability sensors in single-use bioreactors because invasive probes may not reliably preserve sterile-barrier integrity.

Non-invasive spectroscopic methods can be implemented in single-use bioreactors and can prevent cross-contamination, but the plastic reactor material can challenge measurement by affecting electromagnetic-wave transmission. Standardized ports such as the Ingold port are also not available for single-use applications. Confirm the bag material, optical path, port arrangement, sterile boundary, and installation method before treating non-invasive access as a practical solution.

One PAT account describes continuous measurement of glucose, lactate, viable cell density, dissolved gases, and product titer instead of relying only on samples withdrawn every 12–24 hours. Use the comparison to ask what your process actually requires: continuous observation, a defined sampling interval, or a laboratory result at selected stages. The answer should follow the process decision rather than the marketing label attached to the instrument.

Record these vessel and process constraints before selecting a sensor:

  • reusable or single-use vessel;
  • available ports, tubing, bypasses, and optical windows;
  • sterile connection and sterilization requirements;
  • expected run duration and maintenance access;
  • culture medium, gas environment, pressure, temperature, and solids;
  • allowable sample withdrawal, return, or product loss;
  • installation steps that must occur before or after autoclaving.

A system that fits the analyte list but cannot fit the vessel is not a complete candidate.

Compare what each technology observes and infers

Detector technology determines how the process reaches the result. Optical, electrochemical, and mass-based detectors can be considered according to the application requirements. The relevant question is not which family sounds most advanced; it is which interaction with the sample produces information that matches the required measurand and operating conditions.

Spectroscopic sensors can provide information about the molecular, physical, and metabolic state of cells during continuous process monitoring. Because on-line sensors and spectroscopic methods can use no reagents, require no sampling, and consume no analyte, they show potential as non-invasive in-situ techniques for viability determination. Confirm which biological or chemical property the spectrum supports in your process rather than treating a spectral signal as a direct reading of every variable you care about.

Soft sensors use available on-line data to estimate variables that are difficult to measure directly or are measured only at low frequency. Soft sensors can be model-driven or data-driven; chemometric techniques such as partial least squares (PLS) are data-driven soft sensors. A soft-sensor output is therefore an estimate whose validity depends on the input data, model scope, reference method, and process conditions.

For every proposed output, ask four questions:

  1. What physical, chemical, or biological interaction generates the signal?
  2. Is the requested variable observed directly, or inferred from other measurements?
  3. Which reference method establishes the relationship between signal and result?
  4. What process changes could move the sample outside the relationship used to interpret it?

Keep direct measurements, spectral observations, and model estimates separate in the data system and in operating procedures. That separation makes later troubleshooting more precise.

A flow infographic shows detector interaction, spectral observation, and soft-sensor estimate converging on signal interpretation. The result classifies an output as a direct measurement, spectral observation, or model estimate.
A clear signal-to-output distinction prevents inferred results from being treated as direct readings.

Specify the data path before accepting real-time claims

Real-time language does not identify a single timing specification. One comparison described in-line analytics as continuous measurements every 30 minutes and off-line analytics as low or periodic measurements 1–2 times daily. That comparison described in-line measurements as providing data in milliseconds, while off-line methods could take hours or days. Treat these as context-dependent descriptions, not interchangeable promises.

Ask for four separate timing values:

  • Acquisition interval: how often the instrument collects a signal.
  • Reporting interval: how often a result is written or displayed.
  • Latency: how long the system takes to turn the signal into an available result.
  • Control-loop interval: how often a control system can use the result.

A supplier discussion should state the process conditions, signal processing, averaging, alarms, and communications associated with each value. Do not infer a control capability from the word “real-time” alone.

Integration requirements include the route from sensor to software, the data format, time synchronization, alarm handling, electronic records, and replacement of field devices. Choose sensors and actuators compatible with digital standards such as Profibus, Foundation Fieldbus, or Ethernet when those standards fit the control architecture. Record the actual interface, protocol, data fields, and ownership of maintenance rather than accepting compatibility as an undefined feature.

For cultivated-meat production, a source published on February 4, 2026 described regular calibration as essential for meeting GMP requirements and food-safety regulations such as Regulation (EC) 853/2004. Keep that statement within its cultivated-meat context. If your process operates under another regulatory framework, map the applicable data-integrity, calibration, and record requirements with qualified personnel instead of transferring the example unchanged.

Your integration brief should show the complete path: process or sample, sensor, signal conditioning, model or calculation, historian or control system, alarm, operator display, and retained record. Mark every handoff where units, timestamps, filtering, averaging, or model status could change the meaning of the result.

Make calibration and reference evidence part of the comparison

Calibration compares an instrument measurement with a known reference standard and adjusts the device so its readout aligns with that standard. A calibration plan should therefore name the reference, the operating range, the process state, the interval, the acceptance rule, and the response when the result falls outside the rule.

Calibration coverage must represent the measurements you expect to make. Our calibration set must span the range of expected concentrations, including the extremes. For an inferential model, our predicted values cannot be more accurate than the reference method used for training; the reference method’s reproducibility determines the model’s accuracy floor. Ask for the reference method, its reproducibility, the calibration range, the number and distribution of samples, and the conditions represented by those samples.

Bioreactor calibration timing depends on the sensor and the process state. Generally, calibrate pH sensors before installation and autoclaving with two reference buffers covering the expected operating range. Calibrate pO2 and optional turbidity sensors after autoclaving and dispensing the culture medium because those parameters depend on actual operating conditions. Follow the applicable instrument instructions for other sensors and document which step occurred before or after sterilization.

During early process development, use frequent off-line sampling to correlate in-line readings with laboratory reference measurements and build predictive models for automated control. The comparison is useful only when the samples, process states, handling, and reference method are representative of the intended application. Define how disagreements will be investigated before relying on the model output.

A demonstrated 3D-printed microfluidic platform monitored pH, pO2, pCO2, glucose, and lactate during microbial and mammalian cell cultivation. Its sampling unit withdrew small cell-free volumes for at-line analysis, and the study demonstrated sampling after an incubation time of 96 h. The platform’s online glucose measurements were compared with manually sampled culture broth measured by a CEDEX Bio analyzer using an enzymatic assay that photometrically monitored NADPH at 340 nm. To enhance glucose-measurement accuracy, the study recommended calibration with both the designated calibration buffer and the specific growth medium in use.

Before accepting calibration evidence, request answers to these questions:

  • Does the calibration cover the complete expected concentration range, including extremes?
  • Does it represent the actual medium, temperature, pressure, gas condition, and solids?
  • Which samples were held out for independent checking?
  • How does the reference method handle filtration, dilution, degassing, or other preparation?
  • What result triggers recalibration, model review, or suspension of the output?

These questions connect the numerical result to the process conditions in which you intend to use it.

A timeline infographic shows pH calibration before installation, followed by pO2 calibration after autoclaving and dispensing culture medium, and optional turbidity calibration after autoclaving.
Calibration timing should match each sensor’s dependence on installation and operating conditions.

Test maintenance, drift, and model failure modes

Bioprocess conditions place demanding requirements on sensors. Only a small fraction of new sensors are suitable for fermentations because harsh process conditions, interfering media, contamination constraints, and long runs without recalibration can invalidate an otherwise promising design. In-situ sensors should be sterilizable and should not leach components into the culture or otherwise contaminate the bioprocess.

Signal drift can result from temperature or pH changes, surface fouling, or changes in the sensing interface over time. Drift complicates calibration and can cause analyte-quantification errors in continuous monitoring. Matrix interference and surface fouling can dominate drift and produce false or unreliable sensor responses in real-world samples, so the maintenance plan must address the sample matrix rather than only the instrument enclosure.

Multiple PAT sensors can support a broad process understanding, but purely data-driven models risk oversimplification and potentially unreliable correlations. A hybrid approach combining mechanistic and data-driven modelling can enable real-time model refinement and improve interpretation of unexpected events. We monitor residuals for model maintenance; a rising spectral residual indicates that the instrument is detecting something the model was not trained on.

A comparison table lists optical dissolved-oxygen sensors with a stated accuracy of ±1% air saturation, an installed cost of $2K–$8K, and minimal calibration using a two-point calibration. Those values belong to the listed comparison entry and do not establish performance, price, or calibration effort for every optical dissolved-oxygen sensor.

Our OrionIR® held its spectral stability across a repeatability test of more than 100 days. That result is evidence about the stated repeatability test, not a universal conclusion about every sample matrix, installation, model, or bioprocess.

Reliable bioreactor measurement requires suitable technology, representative installation, controlled signal processing, calibration, adequate response, sound control loops, alarms, trustworthy electronic records, and lifecycle management. Convert those requirements into named responsibilities:

  • inspect and clean the sensing surface;
  • monitor drift, residuals, and data-quality flags;
  • define calibration and recalibration triggers;
  • retain alarms, model versions, reference results, and maintenance records;
  • test the response after a sensor, software, or process change;
  • state when an output must be treated as unavailable or investigational.

A maintenance plan is part of the measurement design, not an after-installation task.

Use qualified review when evidence or requirements remain unclear

Qualified review becomes necessary when the measurement will support a consequential process, quality, or regulatory decision and the evidence chain remains incomplete.

Use a review gate when any of these conditions apply:

  • an adaptive model changes during process operation;
  • the target variable is inferred but the reference method or calibration range is unclear;
  • the proposed installation affects a sterile boundary, single-use vessel, or product-contact path;
  • the timing specification does not separate acquisition, reporting, latency, and control-loop intervals;
  • the alarm, record, model-status, or requalification requirements have not been assigned.

Bring the unresolved question to the appropriate process, engineering, quality, regulatory, or technical specialist. A short review before selection can prevent an instrument comparison from becoming a premature implementation decision.

Prepare a supplier discussion brief

A supplier discussion brief should describe the process stage and decision first, then provide the technical conditions needed to judge fit. Include:

  • target measurands and whether each is direct or inferred;
  • expected concentration or operating range, including extremes;
  • required accuracy, reporting interval, latency, and control-loop use;
  • stream or culture composition, temperature, pressure, and presence of solids;
  • vessel type, ports, sampling route, sterile boundary, and allowable sample handling;
  • reference method, calibration plan, model inputs, residual checks, and validation evidence;
  • data interfaces, records, alarms, maintenance access, cleaning, and recalibration triggers.

This brief lets you compare the measurement path, not only the detector name.

To start testing, we require the stream composition, concentration range, accuracy, temperature, pressure, and presence of solids. Supply those details together with the vessel, sampling, integration, calibration, and maintenance questions so the technical discussion has a defined starting point.

Your next decision is whether the proposed system can produce an interpretable result, through a compatible sample path, at the time and operating conditions your process requires. If the evidence does not answer that question, identify the missing test or specification rather than treating an unresolved assumption as a capability. A qualified process, engineering, quality, or regulatory professional should decide individual suitability and applicable requirements for your operation.


What should you confirm before comparing sensor systems?

What should I confirm about sensor timing before using a system for process control?

Ask separately for the acquisition interval, reporting interval, result latency, and control-loop interval. These values can differ by instrument and implementation, so do not treat “real-time” as a complete timing specification. Confirm the effects of signal processing, averaging, alarms, and communications under the intended process conditions.

How can I tell whether a bioprocess measurement is direct or model-based?

Identify the physical, chemical, or biological interaction that generates the signal. A direct measurement observes the target more directly, while a spectroscopic or soft-sensor result may estimate it from other signals. For an inferred result, verify the reference method, calibration range, model inputs, residual monitoring, and process conditions represented by the model.

What information is needed to assess a sensor for a single-use bioreactor?

Confirm the bag material, available ports or optical paths, sterile-boundary requirements, sampling or bypass arrangement, and installation method. Non-invasive approaches may reduce cross-contamination concerns, but plastic permeability can affect optical measurements, and single-use vessels may lack standardized ports. Verify compatibility with the actual vessel rather than assuming that a sensor suitable for a reusable bioreactor will transfer directly.

How should sensor performance be checked during a long bioprocess run?

Define a lifecycle plan covering calibration, cleaning or inspection, drift and residual monitoring, alarms, reference measurements, recalibration triggers, and records of maintenance and model versions. Use representative reference testing to identify fouling, matrix effects, or process conditions outside the model scope. If the signal becomes unreliable or the model status is uncertain, specify when the output must be treated as unavailable or investigational.


Next step

Have a question for Paeonia Innovations?

Get in touch