Why a Soft Sensor Project Has No Single Price Tag
The cost of integrating soft sensors into a bioprocess platform is driven less by the model itself and more by four structural choices the buyer controls. First, the reference-assay program that calibrates the model — the Frontiers 2021 review (Cárdenas et al., doi:10.3389/fbioe.2021.722202) stresses that "accuracy of the reference and online measurements limits the accuracy of the resulting soft sensors."
Second, the modeling software stack: chemometric packages such as SIMCA-online, Process Pulse, or CMET, or in-house builds in environments like LabVIEW. Third, the integration layer — directly in the control level (e.g. PLC), inside the SCADA/PAT platform (MFCS, Eve, BioXpert, SIMATIC SIPAT, synTQ, xPAT, Lucullus PIMS), or in a cloud environment (MindSphere, Predix). Fourth, qualification, documentation, and the ongoing recalibration budget the same review explicitly calls out as necessary in practice. Each choice has a distinct cost profile, which is why a single price tag is misleading — the real answer is a configuration decision.
The same Frontiers review defines three automation layers for any bioprocess measurement system: field (sensors and actuators), control (programmable logic controller / process control system), and supervisory (SCADA and other data management). These layers dictate how data flows through the infrastructure. A soft sensor can, in principle, sit on any of these layers, but the review notes that "online implementation of soft sensors requires at least communication between field and control level and in most cases also supervisory level." Software-only deployments that bypass the control layer are not the standard case in practice. The buyer's first scoping decision is therefore where in the automation stack the soft sensor will live, because that decision locks in the integration effort, the validation scope, and the lifecycle cost — not which algorithm to use.
The Four Cost Buckets That Make Up an Integration Budget
A bioprocess soft-sensor budget is the sum of four cost buckets, and most overrun risk lives in the buckets buyers underweight at the proposal stage. Mapping every quote to these four lines is the fastest way to compare competing vendor proposals.
- Reference assays and offline data. Because the Frontiers review states that "the accuracy of the reference and online measurements limits the accuracy of the resulting soft sensors," the offline analytical program — cell density, titer, HPLC, BOD/COD, or spectroscopic reference scans — is a real CAPEX and recurring OPEX line, not a free input. The buyer must size the number of reference samples per batch and the analytical instrument time those samples consume.
- Modeling and chemometric software. The buyer chooses between a commercial PAT platform (SIMATIC SIPAT, synTQ, xPAT, Lucullus PIMS) and an indirect implementation in chemometric or analyzer software (SIMCA-online, Process Pulse, CMET), per the Frontiers review. Licensing models for these two categories differ, and only the vendor quote will give the per-seat, per-site, or per-prediction number.
- Integration architecture. The review documents three implementation paths: direct in the control level (PLC / process control system), direct in the SCADA or data management system (MFCS, Eve, BioXpert, SIMATIC SIPAT, synTQ, xPAT, Lucullus PIMS, LabVIEW), or in the cloud (MindSphere, Predix). The architecture choice is also a vendor-lock-in choice, because each platform pulls data from a specific historian or sensor bus.
- Qualification, documentation, and recurring recalibration. The review states that "when soft sensors are implemented in an industrial environment, they must first undergo an intensive functional and risk assessment (qualification)." It further states that "the maintenance or recalibration of soft sensors — just as for hardware sensors — is necessary in practice to preserve the quality of their prediction performance." Both lines belong in the budget from day one.
For bioprocess plants that also run a biological wastewater treatment train, the same four-bucket model applies to inferential sensors on the WWWT side. A plant already running an MBR membrane bioreactor system will already own much of the supervisory layer that a soft sensor needs to land on.
Build vs. Buy: Which Soft-Sensor Path Fits Your Plant?
The Frontiers review implicitly defines four integration paths. Placing the project on the right path is the single biggest cost lever a buyer controls. The matrix below summarizes where each path sits on cost, internal labor, and qualification effort; specific license and integration fees are not stated in the public literature and must be requested from each vendor.
| Path | Where it lives | Best fit | Cost shape | Main buyer risk |
|---|---|---|---|---|
| A — Embedded in an existing PAT platform | Inside SIMATIC SIPAT, synTQ, xPAT, or Lucullus PIMS | Plant already runs one of these platforms | Lowest incremental cost; data plumbing and qualification templates are reusable | Vendor lock-in to one PAT platform |
| B — Chemometric software add-on | SIMCA-online, Process Pulse, or CMET layered on top of existing SCADA | Plant has SCADA but no full PAT platform | Moderate; software license plus chemometrician labor | Validation effort per the review's qualification requirement |
| C — In-house in LabVIEW or open-source | Custom build on the supervisory or control level | Skilled internal modeling team; novel analyte | Lowest software license cost; labor dominates | Same SCADA-to-PLC communication and qualification documentation still required, per the review |
| D — Cloud deployment on MindSphere or Predix | Cloud platform, with model shared across plants | Multi-site program or fleet-scale model | OPEX-heavy, per-prediction or per-data-volume | Data-governance and IT-security review not detailed in the literature and must be scoped by the buyer |
The review's key point for build paths is that even an in-house LabVIEW or open-source soft sensor still needs "communication between field and control level and in most cases also supervisory level" plus the same qualification documentation as a vendor product. Internal labor — not license cost — is what drives a build path's total.
Matching the Cost Profile to Your Bioprocess Type
The same Frontiers review notes that soft-sensor input data "can compose differently depending on the organism (bacteria, yeast, filamentous fungi, mammalian or insect cells, etc.) used in USP and the techniques used in DSP." These variations dictate the necessary hardware and preprocessing requirements. A mammalian-cell DSP line fed by chromatography and mass-spectrometry inputs has a different cost profile than a bacterial USP fed by spectroscopic inputs, because the input data shape drives probe CAPEX, preprocessing compute, and reference-assay program scope.
Spectroscopic inputs (Raman, NIR, UV-Vis) add probe CAPEX and a "complex preprocessing" step that the review flags as a limit on prediction frequency — compute and data-engineering cost rises with prediction rate. Biosensor and free-floating wireless sensor inputs shift cost from CAPEX into per-batch consumable OPEX, with a maintenance schedule closer to a disposable than to an inline probe. For biological wastewater treatment specifically, the same four-bucket model applies, and benchmark cost profiles from municipal WWWT plants are a reasonable proxy when internal data is missing. Buyers running an MBR plant operation and maintenance guide workflow will already have the historian, sensor bus, and SCADA plumbing that a soft sensor needs as inputs.
Lifecycle OPEX: Recalibration, Drift, and Sensor-Fault Tolerance
The project does not end at go-live. The Frontiers review explicitly states that soft-sensor recalibration is "necessary in practice to preserve the quality of their prediction performance," and the buyer should treat that as a planned annual line, not an emergency line. The same review identifies "fault tolerance" and "sensor fault detection" as core remaining challenges, meaning the soft-sensor design itself must include redundancy and a fault-handling path — which adds both software and documentation cost at the design stage, not just at maintenance time.
For spectroscopic soft sensors, the review flags that prediction frequency is limited by "complex preprocessing" and computational power, which sets a real cost ceiling on cheap hardware: storage and compute scale with prediction rate. For cloud-hosted soft sensors on MindSphere or Predix, OPEX becomes a per-prediction or per-data-volume subscription that the buyer must model against the chosen prediction frequency, not against the one-time integration effort. A DaaS vs CapEx TCO comparison framing is the right structure for that subscription line, and a parallel review of treatment-train equipment cost — covered in the CMP wastewater equipment cost comparison — gives a useful cross-check on the magnitude of recurring spend the plant already carries for similar process equipment.
One practical lever on recurring cost is putting the soft sensor on a PLC-controlled chemical dosing skid that already streams process data to the supervisory layer: the soft sensor can then consume that data stream with no new sensor CAPEX, which compresses the lifecycle number back into the OPEX-only band.
Frequently Asked Questions
What budget line items should we expect in a soft-sensor integration project?
Four: the reference-assay program (offline analytics labor and instrument time), the modeling and chemometric software license, the integration architecture effort (control, SCADA, or cloud), and the qualification documentation plus recurring recalibration line. The Frontiers 2021 review names each of these as required in practice; the public literature does not state price levels, so a defensible budget must be built line by line from vendor quotes and internal labor estimates.
How do we choose between an in-house build and a vendor PAT platform?
Match the choice to the platform already on the plant floor. If SIMATIC SIPAT, synTQ, xPAT, or Lucullus PIMS is already licensed, the lowest-risk path is to embed the model there (Path A in the build-vs-buy matrix). If only a SCADA layer exists, a chemometric add-on such as SIMCA-online, Process Pulse, or CMET (Path B) is the next step. The review's qualification requirement applies equally to in-house LabVIEW builds, so internal labor — not license cost — is the swing factor on a build path.
What recurring OPEX should we plan for after go-live?
Plan an annual recalibration line, a sensor-fault-detection and redundancy design cost at the build stage, and — for cloud deployments on MindSphere or Predix — a per-prediction or per-data-volume subscription. The Frontiers 2021 review explicitly flags recalibration and fault tolerance as "necessary in practice" and as "core remaining challenges," respectively; a defensible OPEX line should be sized off prediction frequency and storage volume, both of which the review ties directly to computational power limits on spectroscopic inputs.
How does bioprocess type change the cost mix?
It changes the input data shape, which drives probe CAPEX, preprocessing compute, and reference-assay scope. A mammalian-cell DSP with chromatography and mass-spectrometry inputs will carry a different profile than a bacterial USP with spectroscopic inputs, per the Frontiers 2021 review. For biological wastewater treatment trains, the same four-bucket model applies and benchmark cost profiles from municipal WWWT plants are a reasonable proxy when internal cost data is missing.
