A soft sensor, also called a virtual or inferential sensor, is a model trained on process data that estimates a value a physical sensor would otherwise measure directly, such as a concentration or quality parameter in a bioprocess. This dataset scopes to filings that combine that sensing concept with model-training mechanics like training data curation, cross-validation, parameter tuning or regularization. It spans both classical process-control soft sensors and newer AI-native inferential models, which is why the technology composition leans heavily toward AI/computing IPC classes alongside control-systems classes.

The ranking of 100 assignees is dominated by a small group at the top: the five leading assignees together hold 52.5% of the 815 records in scope, and the single largest holds 251 records on its own. Below the top handful the count drops quickly, with fifth place at 17 records and tenth place at 10, leaving a long tail of assignees with only a few filings each. The mix includes industrial-IoT platform companies, automotive and aerospace OEMs, chipmakers and process-instrumentation specialists, reflecting how widely the underlying training techniques have spread beyond their original process-control context.

Filing volume peaked at 144 records in 2019 and the most recent fully comparable period, 2021 to 2024, shows a decline from 89 to 73 records, an 18% drop. That is a real cooling from the 2019 peak rather than a collapse, and it should not be read as the field losing relevance. Publication typically lags filing by around 18 months, so the 2025 and 2026 figures in the raw trend are still incomplete and will rise as more applications from those filing years publish.