Skip to main content

If a critical input for the model goes out of specification and is dropped from the model, how do you monitor the change in accuracy and precision of the PEMS? If a second critical input has to be dropped (or more) at what point do you say a good prediction is no longer possible?

Non-critical parameters are, by definition, not important in maintaining accuracy at the required level (usually this is within 10%). We determine the critical parameters using an input failure test and analysis. This is done each time the model is adjusted.  The input failure test is conducted similar to the daily calibration, only we fail various combinations of inputs and look at the deviation in the prediction (usually on a mass emission basis) taking into account the pollutant and diluent.

 

If a single input failure produces greater than 10% deviation in the predicted mass emission rate, it is deemed critical. Failures in the critical inputs, will generate an alarm and invalid flag that can be used to screen the data and can account for monitor downtime. If a combination of inputs yields a deviation greater than 10%, these inputs are also deemed critical.

 

It is important to note that we have much flexibility in the use of inputs in our model, so by properly designing the model and using reliable inputs, we can minimize the critical inputs and reduce monitor downtime. When the process goes outside of the envelope (or finds a gap in the data within the envelope), the PEMS status is flagged invalid even if the prediction is accurate.

 

Correction to this problem is either to operate within the envelope or add data to the model to extend the envelope (or fill the gap) such that the PEMS status is valid. We always try to build models up front that are robust – that includes all of the normal operating conditions such that monitor downtime is   less than 1% per quarter.