Case Study

Soft sensor and sensor fusion temperature measurement

At Cambridge Aerothermal, we have been developing machine learning and sensor fusion techniques to get measurement systems with even better accuracy and instruments with error detection and correction. Furthermore, our sensor fusion instruments can provide a high-fidelity correction of a sensor that is erroneous.

We use a single point sampling system where a sonic probe (Case Study 1) measures total gas temperature by (i) the Sonic Method, (ii) O2 TBGA and CO2 TBGA. For example, the uncertainty (3σ) for a specific measurement for the Sonic Method is ±1.8%, O2 TBGA is ±4% and CO2 TBGA is ±3%. By combining the result, shown in Figure 1, our Sensor Fusion method has an overall uncertainty of ±1.4% which is better than each instrument alone. Also, in Figure 2, the Sensor Fusion method can detect an erroneous instrument, in this case, CO2 TBGA. The Sensor Fusion uncertainty becomes ±1.6%, which is still better than each instrument alone as well as being able to estimate the CO2 TBGA providing a ‘virtual’ CO2 measurement.