| OCR Text |
Show BENEFITS OF NONLINEAR REGRESSION MODEL BASED PEMS Regardless of the modeling method employed, achieving the EPA 95% up-time with a PEMS for demonstration of continuous compliance requires a stringent quality assurance program. Since PEMS are computer software programs which rely on sturdy process sensors, achieving the required up-time is almost assured. In order to further assure that the up-time requirement is met, the ability to detect sensor failures, and, if possible, compensate for those failures is necessary. Sensor validation is a crucial step to ensure accuracy and reliability of the emissions model. If sensor drifts or failures go undetected, the process could be out of compliance without detection, or the system could determine emissions that are higher than the proper values. Either case is unacceptable for an accmate, reliable system. When a nonlinear regression model based PEMS is created, a Sensor Validation System can be also be created. A Sensor Validation System is based on an accurate models of all of the sensors used in the emissions model as a function of other sensors in the plant. These sensor models enable the accurate prediction of process data so that in the event of a process sensor failure or drift, the PEMS can continue to determine emissions accurately. This validation and data reconciliation function ensures that the PEMS will continuously provide accurate monitoring values in a reliable and robust fashion. Sensor validation is based on solid principles that are commonly used in all measurements, essentially, checking one sensor measurement against an independent measurement. This is exactly how calibration of a sensor is done; a sensor is compared against an independent reference value and if it is off, it is adjusted or replaced. This is the same concept behind a Sensor,Validation System. It is based on independent checks of each sensor, but the reference in the case of the sensor validation system is a calculation based on data from the other sensors. Consider the following: Sixteen temperature transducers located at various positions on a process. The temperature measurements of each transducer is correlated and interdependent. If, for example, these transducers are located on a pipe and one sees that ttansduCf'I A reads 134 degrees , C reads 135.2 degrees, but Breads 57 degrees, then if B is located between A and C on the pipe, it is evident that B is wrong. In fact, B should be measuring around 134.6 degrees. Note that with the infonnation provided here, it is unclear as to whether B has failed or whether A or C have failed. Based on common sense, since A and C agree, the assumption is that B has failed. The same is true in calibration of any instrument. If you compare an instrument to a reference, and you see that there is a significant deviation, it could be that the instrument is off, or 8 |
