Laboratory Update of Inferential

With well-integrated information systems it is relatively easy to automatically update the inferential with the latest laboratory result. Any difference between the laboratory result and the value of the inferential at sample time can be used to update the bias term in the inferential calculation. 

If we do need to update automatically then we need to separate the bias error from the total error. The CUSUM technique offers an effective solution. In this case CUSUM is the cumulative sum of differences between the inferential and the laboratory result. Table 9.2 presents an example calculation. 

There is no need to ever reset the CUSUM to zero. However is important to record that a correction has been made so that subsequent estimates of the CUSUM gradient do not include values collected before the correction. 

If the error has even a small random component, the performance index (( ) will always reduce if automatic updating is implemented. The index measures only random error. If there was no random error then the index would have a value of 1 - no matter how large the 

Whether the correction should be fully automated is debatable. It is certainly important to monitor random and bias errors frequently but automatic correction is not a substitute for a poor inferential. Its inclusion may disguise a problem. Indeed this is exactly the situation with our predicted stock values. The prediction was quite simply yesterday s value with automatic updating based on today s value. 

---