Virtual sensing¶
Inferring a sensor-like reading of a target quantity from other measured signals through an inferential model when direct measurement is unavailable or impractical for the intended task.
Core Idea¶
Virtual sensing uses measured physical signals and a model to infer a sensor-like reading of a different target quantity. It is useful when the target cannot be measured directly, or direct readings are impractical for the intended task. The estimate needs testing against representative target data before anyone can rely on it; testing is a credibility question rather than a condition for naming the method.[ref-e9dfbb5c831a][ref-4b2c43a0c24f]
Scope of Application¶
In a cement kiln, Bao Lin and colleagues estimated free lime and nitrogen oxide from recorded process signals. Free-lime laboratory assays were about two hours apart, while process measurements were recorded every ten minutes. In a vehicle study, Fen Lin and Youqun Zhao compared estimates of side-slip angle from yaw rate and lateral acceleration against measured side-slip angle. The vehicle paper's available original abstract discusses possible guidance for stability-control design; it does not establish deployment.[ref-e9dfbb5c831a][ref-4b2c43a0c24f]
Clarity¶
The input sensor reads one variable; the virtual sensor infers another. A reported “free-lime measurement” may be a laboratory assay or a model-based estimate, and those have different limits. Likewise, a tested side-slip estimator is not automatically an installed control sensor.
Manages Complexity¶
Ask five questions: What target is wanted? Which signals are physically measured? What model maps them to the target? What data test the mapping? What inferred reading comes out? This short map works across different engineering systems without requiring the same algorithm or sampling interval.[^ref-e9dfbb5c831a]
Abstract Reasoning¶
Trace the target backward through its model and input signals. Then check whether target observations support the model under conditions relevant to the intended task. If those conditions change or the input signals degrade, the inferred reading may need renewed evaluation; a successful comparison in one study is not a universal accuracy guarantee.[^ref-e9dfbb5c831a]
Knowledge Transfer¶
The same relation appears in kiln process monitoring and vehicle dynamics, with different targets and models. General statistical estimation is related, but this named method has the extra sensor-substitution role: measured auxiliary physical signals supply a reading for another target.[ref-e9dfbb5c831a][ref-4b2c43a0c24f]
Example¶
In Bao Lin and colleagues' kiln study, free lime is the target, process measurements are inputs, the fitted inferential model maps those inputs to the target, laboratory values help validate it, and the more frequent estimated value is the sensor-like output. The study does not make its particular algorithm or data-cleaning choices mandatory for every virtual sensor.[^ref-e9dfbb5c831a]
Neighborhood in Abstraction Space¶
Virtual sensing sits in a sparse region of the domain-specific corpus (98th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.
Family — Unclustered & Miscellaneous (2551 abstractions)
Nearest neighbors
- Floor Effect — 0.77
- Uncertainty analysis — 0.76
- Least-Squares Adjustment — 0.75
- Iterative reconstruction — 0.75
- Probability Bounds Analysis — 0.75
Computed from structural-signature embeddings · 2026-10-08
Not to Be Confused With¶
A direct physical sensor measures the target itself; a virtual sensor infers it from other physical measurements. A statistical estimator is a rule that may lack the sensor-substitution task. The current catalog leaves Virtual sensing a provisional unparented root: its broader inference skeleton resembles the live Estimation Prime, but that Prime's explicit uncertainty and decision-adequacy conditions are not necessary for every virtual-sensor design.
[^ref-e9dfbb5c831a]: Bao Lin, Bodil Recke, Jørgen K. H. Knudsen, and Sten Bay Jørgensen, “A systematic approach for soft sensor development,” Computers & Chemical Engineering 31 (2007), 419–425, DOI 10.1016/j.compchemeng.2006.05.030; especially abstract, introduction, §§4.1–4.2, and outlier discussion on p. 424. https://skoge.folk.ntnu.no/prost/proceedings/npc07/DTU/dtu09.pdf [^ref-4b2c43a0c24f]: Fen Lin and Youqun Zhao, “A Comparison of Two Soft-Sensing Methods for Estimating Vehicle Side Slip Angle,” SAE Technical Paper 2007-01-3587 (2007), original author abstract reviewed; full paper and numerical results not reviewed. https://saemobilus.sae.org/papers/a-comparison-two-soft-sensing-methods-estimating-vehicle-side-slip-angle-2007-01-3587