Soft sensors for predictive quality control
Predict CQAs in real time with hybrid physics-AI models. Built on your existing process data, they flag a batch heading out of spec and provide recommendations on how to correct it.
Predict quality in real time. Correct the batch before it goes out of spec.
Soft sensors predict critical quality attributes between lab results, correct themselves as each result comes in, and recommend the adjustment that keeps the batch in spec.
How it works: soft sensors turn the data you already have into quality predictions.
Process tags, off-gas and any installed PAT probes stream in from the historian. Lab results calibrate the model.
Mass and energy balances form the backbone. AI fits what the physics doesn’t capture, so each prediction comes with a confidence interval.
Predictions feed monitoring, forecasts, alerts and control, and are written back to the historian.
Predictive quality control with real-time visibility into every CQA.
Understand how your CQAs are trending, identify deviations, and receive recommended corrections to maintain on-spec production.
Traditional PAT vs. Basetwo.
Traditional process analytical technology (PAT) needs a dedicated sensor for every quality attribute you want to measure. With Basetwo, one model predicts them all from the process data you already collect.
soft sensor
Continued process verification (CPV) reports, generated automatically.
Stop assembling CPV reports by hand. Basetwo pulls batch data from historian, MES and LIMS, trends every CPP and CQA against its limits, flags signals for review and generates the report as a PDF, every month, quarter or N batches.
Proven in production at leading manufacturers. Explore the case studies.
Predict product purity in real time and adjust reflux before off-spec distillate is collected.
Track cell growth and titer between daily counts and adjust feeds before a batch drifts.
Predict melt index from process tags and correct grade transitions before off-spec product builds up.
Questions PAT and quality teams ask
Usually not. Most soft sensors start from tags you already log, such as temperatures, flows and off-gas. Where an NIR, Raman or FBRM probe is installed, its signal is used as an input and the hybrid model makes it less sensitive to recalibration.
Fewer than a purely data-driven model, because the mass and energy balance carries most of the structure. A few dozen paired samples across normal operating variation is a typical starting point, and the model reports where it is under-calibrated.
It can produce the evidence, but RTRT is an agreement with your regulator, not a software setting. Most teams start with the soft sensor as in-process monitoring and decision support, and build the validation record toward RTRT from there.
The confidence interval widens and the sensor flags the inputs that are out of range. It never reports a confident number it can’t back up, and those periods are the samples the next retrain asks for.
Send us a month of process data and lab results. We'll build your first soft sensor on it.
