A soft sensor, also called a virtual or inferential sensor, is a model-based estimate of a process variable — biomass concentration, a metabolite level, or a product titre — that substitutes for or supplements a physical measurement that is slow, expensive, or unavailable in real time. In the patent literature this shows up as claims combining a state-estimation or uncertainty-modelling method with a control or AI classification. This dataset's search string specifically pulls records that pair soft-sensor language with state-estimation, biomass-estimate, or prediction-interval terms, which is why the results lean toward control-engineering assignees rather than bioreactor hardware makers alone.

The ranked field covers 35 assignees, with one filer well ahead of the rest at 11 records against 5 at fifth place and 3 at tenth. The five most active filers together account for 61.7% of all 60 records in scope, and the ten most active for 88.3%, which makes this a concentrated field with a long tail of smaller and individual filers behind the leaders. Recent-year momentum data, however, shows even the largest corporate filers with zero records in the latest tracked year, a pattern that likely reflects publication lag as much as any genuine pull-back.

The only reliable year-over-year comparison in the data is 2021 to 2024, where filings fell 89% from 9 records to 1. Filing peaked earlier, in 2018, at 11 records. Counts for 2025 and 2026 are still filling in because publication typically lags actual filing by around eighteen months, so those recent years should be read as incomplete rather than as evidence the field has gone quiet.

Control and regulating systems (IPC class G05B) appear in 68.3% of the 60 records, making it by far the dominant classification, followed by AI-based computing methods (G06N) at 35.0%. Bioreactor and enzyme apparatus (C12M), the class most specific to bioprocessing hardware, appears in only 15.0% of records. That gap suggests most patent activity in this space is written as general estimation and control methodology that is applicable to bioprocesses, rather than as bioprocess-specific equipment claims.

Given that generic control and AI classifications dominate the corpus while bioreactor-specific apparatus claims (C12M, 15.0% of records) are comparatively thin, the more open ground sits in claims that tie inferential estimation methods explicitly to bioreactor hardware, calibration routines, or uncertainty-bounded metabolite prediction. Sub-areas like hybrid mechanistic-machine-learning estimators for biomass and cross-scale transfer learning for soft sensors show limited dedicated coverage relative to the field's control-theory core. A freedom-to-operate search focused on these narrower combinations is more likely to surface open claim space than a search on soft-sensor estimation broadly.