An Instrument Control System Using Predictive Modelling

We describe a system for providing early warning of possible error to an operator in control of an instrument providing results in batches from samples, for example, chemical elements found in soil or water samples. The system has the potential to be used with any form of instrument that provides multiple results for a given sample. The idea is to train models for each measurement, using historical data. The set of trained models are then capable of making predictions on new data based on the values of the other measurements. This approach has the potential to uncover previously unknown relationships between the measurements. An example application has been constructed that highlights the difference of the actual value for a measurement from its predicted value. The operator is provided with sliders to attenuate the sensitivity of the measurement perhaps based on its importance or its known sensitivity.