All services
Soft-sensor proof of concept
Estimate a lab-measured quality between samples, using the signals you already record.
From $2,000 for a typical scope
Discuss this service“Lab results come too late”
When a quality result arrives hours after the sample, the unit has already run on old information. A soft sensor uses live temperatures, pressures and flows to estimate that quality between samples.
A good fit when
- A lab-measured quality drives a real operating decision
- Samples are too infrequent to catch changes in time
- An online analyser is missing, unreliable or too costly
How the work runs
Timing is a planning estimate, confirmed after the first stage.
| When | Work | You can review |
|---|---|---|
| Week 1 | Tag mapping, timestamp alignment and lab delay checks | Data-quality report and go/no-go decision |
| Weeks 2–3 | Agree the test period and score simple estimates first | Baseline report |
| Weeks 3–5 | Fit candidate models and test on the frozen period | Error, coverage and failure review |
| Weeks 5–6 | Review results with your process engineers | Decision note |
What we need
- Historian tags around the unit, with units and sample rates
- Lab results with true sample times, not entry times
- Sensor faults, startups, shutdowns and maintenance records
- Any existing rule of thumb operators use
What you keep
- Data dictionary and reproducible quality checks
- Soft-sensor model and source code
- Comparison with simple baselines and current practice
- List of conditions where the estimate should not be trusted
Agreed before we start
- Useful accuracy against lab results
- Time between the estimate and the action it supports
- Share of operating time the estimate is valid
When we’d stop
We stop or narrow the work if lab timestamps can’t be recovered or the available signals carry too little information about the quality.