Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

Feature Report

THE MATURATION OF A TECHNOLOGY:

Predictive
���������������������������������������������

����������

Emissions
��������������������������
��������������������� ��� ����
����������
������������������� �� �����

Monitoring
����������������

FIGURE 1. Neural
networks learn patterns
to predict emissions from process-sensor-input data

To be successful,innovative environmental technology and regulations


must evolve together. This technology’s success is based on
its effectiveness and its cost advantages
Paul S. Reinermann, III, Pavilion Technologies, Inc.

A
ir pollution is monitored by curate as the hardware-based CEMS 60 (New Source Performance Stan-
many different regulatory over the entire range of operation, and dards), whereas more cumbersome
programs, some of which to cost 50% less. and less favorable PEMS require-
require that emissions be Since then, 250 PEMS have been in- ments have been established for emis-
measured continuously. This stalled worldwide. The effectiveness of sion units that are subject to 40 CFR
task is often accomplished PEMS has been proven in many types Part 75 (Acid Rain Program — Utility
by directly measuring air pollutants of process units, combustion configu- Boilers and Turbines). The main dif-
with continuous emissions monitor- rations and fuel types (see box, p. 51). ference between these two regulations
ing systems (CEMS). In recent years, lies in the initial certification require-
innovative technology has enabled a PEMS AND REGULATIONS ments — a one- to three-day test under
new approach that predicts emissions PEMS can only predict certain pollut- Part 60 and a 30-day test under Part
from process variables. Industry wel- ants. The most prevalent pollutants 75. Note that the quality and the ac-
comes predictive emission monitoring predicted are nitrogen oxides, followed curacy of the PEMS are not enhanced
systems (PEMS) because they reduce by carbon monoxide (CO). Regulations because of the increased certification-
emissions-monitoring costs. Regula- generally require that emissions be test duration.
tory agencies are accepting these new normalized to a common oxygen (O2) Most state agencies and non-U.S.
systems, as proof of their accuracy and or carbon dioxide (CO2) basis. For ex- agencies have adopted the Part 60
reliability is demonstrated. ample, measurements of NOx in parts- PEMS requirements published by
The onset of PEMS was in 1992, per-million (ppm) may be normalized EPA in January 1996. The Part 75
when a chemical engineer used neural- to percentage of O2. By far, the usage PEMS requirements were viewed as
networks (Figure 1), a pattern-learn- of O2 is more customary than CO2. impractical by the Texas Environ-
ing technology, to accurately predict Pollutants such as sulfur dioxide mental Quality Council. Hundreds of
nitrogen oxides (NOx) emissions from (SO2) or specific organic compounds are PEMS have been established under
a natural-gas-fired boiler. This model related to the sulfur content in fuels or Part 60 PEMS requirements whereas
worked like no other first-principles the level of the organics in a material. less than ten PEMS have been estab-
model or regression model previously The prediction of these pollutants is lished under Part 75. The main reason
built — it was accurate over the entire done by mass-balance calculations and for the difference is that PEMS under
operating range of the boiler, whereas is not considered by regulatory author- Part 60 cost significantly less than
the other models were accurate only ities to be part of a PEMS. under Part 75, primarily due to the
in specific operations. Within a year, cost of certifying the PEMS.
the neural-network-based NOx model Regulations
evolved into the first PEMS that was ap- PEMS are more compatible with some Regulatory acceptance
proved by the U.S. Environmental Pro- regulatory programs than with others. In the U.S., in order for the EPA to
tection Agency (EPA) as an alternative Practical PEMS requirements have accept PEMS, the agency required as-
to direct CEMS measurements. This been established for emission units surance of the quality and reliability
first PEMS was proven to be as ac- that are subject to EPA’s 40 CFR Part of PEMS’ outputs over a long term.
50 CHEMICAL ENGINEERING WWW.CHE.COM JULY 2006
TABLE 1. GENERAL INFORMATION
NEEDED FOR PEMS FEASIBILITY
EVALUATION
������������������ Contact In- Operations
����������� formation Environmental
Information technology (IT)
Type of Boiler
Emission Turbine
Unit Process heater
Combination
Other
��������������
Burners Number of burners
Air register and other burner
controls
����� �����
������������ �������� Types and Natural gas
��������� Ratios of No. 2 fuel oil
Fuels Refinery or process gas
FIGURE 2. Execution of the DOE creates the dataset that is needed for Coal
building the PEMS emissions and sensor-validation system models Wood waste
Process Fuel flow
Sensors Fuel quality
PROVEN PEMS APPLICATIONS Available Air flow
Process Units: Turbines with duct burners, other turbines, engines, boilers, Temperatures
olefin furnaces, crude oil heaters, styrene superheaters, methanol reformers System In- DCS and data historian
Configurations: Single burner, multiple burners, low NOx and dry, low tegration
NOx burners, flue gas recirculation (FGR), steam injection Pollutants NOx, O2, CO, CO2, SO2, Particu-
and Regu- lates
Fuels: Natural gas, No. 2 fuel oil, other fuel oils, refinery gas, hydrogen-rich
lations US EPA 40 CFR Part 60 or Part 75
fuels, varying but measured gaseous-fuel quality, waste wood EU NEN-14181
Permit Requirements

Initially, the EPA used the CEMS sion source (also called the emission gas and oil, either alone or in combina-
quality-assurance requirements as a unit in EPA terms) must be evaluated. tion (effectively three fuel types that
template. For the most part, however, A general checklist of information that produce different level of pollutants),
fitting PEMS into CEMS require- makes up such an evaluation is given then nine days are required for data
ments is like putting a square peg into in Table 1. gathering, two for validation and two
a round hole. As a better fit, two key Contact information. Operations for certification. Also, the time that is
PEMS technologies were created: personnel know the emission unit, needed to build the model doubles.
1) A realtime sensor validation system the burners, the fuels, and the avail- Another critical step is ensure that
that evaluates all sensors which input able sensors. They do not necessarily needed process sensors are available.
into the PEMS (this is discussed in understand the nuances of the envi- In general, most classical instrumen-
more detail later) ronmental regulations nor the system tation packages for boilers, turbines,
2) The daily electronic data assess- integration issues, so contacts who are and process heaters contain sufficient
ment tests of the PEMS, where a experienced in both environmental sensors to support the PEMS. For
known set of values is put into the sys- and information technology (IT) are process or refinery gases, fuel quality
tem to verify the PEMS accuracy also necessary. For these key people, monitors need to be present because
their names, phone numbers and fluctuations in fuel quality affect pol-
PEMS PROJECT EXECUTION email addresses should be obtained lutant levels. The presence of the nec-
The decision to use a PEMS is based and made readily available. essary process-sensor data in the dis-
on two factors: technical feasibility Details on the emission unit. In- tributed control system (DCS) or data
and cost effectiveness. The technical formation about the emission unit, historian also needs to be confirmed.
viability of a PEMS for a particular including the number and type of System integration. The PEMS soft-
application should be assessed be- burners, fuel types and ratios, and ware must be able to access the pro-
fore economics are even considered. available process sensors is needed cess-sensor data from either the DCS
Once the technical aspects have been and can usually be obtained from or data historian. For optimum per-
confirmed, regulatory issues must be the operations group. The simplicity formance, design your system to read
evaluated. In some states, EPA re- or complexity of the emission unit is process sensor data from the DCS and
gions, and regions of the world, PEMS a function of the number of different have the PEMS write values back to
have not been introduced. And despite configurations of burners and fuels. the DCS, using information that the
their technical soundness, PEMS have The simpler the emission unit, the less PEMS obtained from the DCS by the
not been accepted in some locations. time is needed to gather data to build, data historian. This design takes ad-
Experienced PEMS vendors will be validate and certify the PEMS. For ex- vantage of the store-and-forward ca-
able to guide you in achieving techni- ample, a natural-gas-fired boiler with pability of most DCSs and provides a
cal sense and regulatory acceptance. a single burner requires three days for backup of PEMS data in case of DCS/
data gathering, one day for validation data-historian communication down-
Step 1. Analyzing viability and one day for certification. If the time. Regulatory agencies expect con-
To check technical feasibility, the emis- same boiler instead fires both natural tinuous monitoring systems to have
CHEMICAL ENGINEERING WWW.CHE.COM JULY 2006 51
����� ����

Feature Report ����� ���������

������������� �����
���������
����
��������������
������

���
data availability for 95% of the oper-
ating time. Proper system integration ����
����
eliminates most known PEMS down- ������������
��������������
time. ���������
Remote access to the PEMS for
periodic or requested evaluation is ��� ����
��� ���� ����� ���
a must to reduce response time and ���������� �����
����
the overall cost of PEMS ownership.
The most effective link is by Virtual FIGURE 3. This preliminary evaluation FIGURE 4. Another step in evaluating
Private Network (VPN). System secu- of a PEMS shows a good r-correlation a PEMS is to check if the predictions are
rity comes into play when considering within ± 7.5% of actual values
remote access. The views and proce- At this stage of the
dures on remote access vary by com- project, accuracy of the �����
���������
pany and change with time as person- emissions data is more ��������������������
nel, products and perceptions change. important than that of ��������������
����
The location of the PEMS server, with the process sensors. For ���������������
reference to the firewall, can make a instance, a good and ad- ����������������

��������
difference in allowing remote access. If equate sampling location ���������������������
����
located outside the firewall, generally on the stack is necessary
remote access can be granted, while if to obtain representative
located inside the firewall, remote ac- samples that are not com- ����
cess may not be possible. promised by possible pol-
lution stratification in the
Step 2. Experimental design stack. Adherence to EPA ����
According to six-sigma methodology, a reference test methods ����
��� ���� �����
design of experiments (DOE; Figure 2) for gathering the emis-
�������������������������
is a structured, organized method for sions data ensures good
determining the relationship between emissions-data quality. In
factors (X) affecting a process and the practice, utilizing dual (re- FIGURE 5. This neural-network PEMS has learned
outputs of that process (Y). In the case dundant) EPA reference the correct gains without the use of additional soft-
ware functions
of PEMS, the DOE is key to creating a test methods guarantees
dataset that reflects the relationships highly accurate emissions
between process operating conditions data and helps avoid having to repeat through the experimental design. If
(X) and pollutants being emitted (Y). portions of the DOE if one reference any planned experiments are missed,
The overall accuracy and robustness test method fails. they will need to be made up, which
of the PEMS is predicated on the ex- will increase costs. The PEMS engi-
ecution of the DOE. Step 3. Executing the DOE neer or operations superintendent
Pushing the operating window. The coordination of the DOE with should work with each operating shift
A properly designed DOE will push plant operations is key to minimizing to explain the DOE and the ramifica-
the operating window of the emission any economic impact that could arise tions of not following it.
unit. The design needs to be created from varied operations of the emission Emissions data sources. When
by an expert PEMS engineer and re- unit. For example, when executing the replacing a hardware CEMS with
viewed by operations for process con- DOE to create a PEMS on a boiler, if a PEMS, a reliable source of data
formity and safety. While an emission the facility has more than one steam is needed. If the CEMS data are of
unit may be capable of a wide operat- source, then the other steam source questionable quality, it is advised to
ing range in many instances, the ac- should be allowed to “swing” to en- not use them. If CEMS data are to be
tual operating range is constrained in sure that enough steam is provided to used, ensure that the CEMS has been
practice. For example, a boiler may be the facility and no fuel is wasted. At shown to meet accuracy requirements
able to operate in the range of 25 to a petroleum refinery, the effect of the of within 7.5% relative accuracy im-
100% of rated capacity, but in reality crude oil slates and the ability to han- mediately before the DOE. It is rec-
only operates between 75 to 100%. dle downstream disturbances of the ommended that consideration also be
Ensuring good data quality. Prior products from the crude heater need given to having the EPA reference test
to conducting the DOE, all process sen- to be considered. method operational during the DOE,
sors should be in good working order. Waiting for the right window of op- in order to supplement and confirm
Calibration of the process instruments portunity to execute the DOE is para- the CEMS data.
is desirable but not necessary. Repeat- mount to containing costs. Execution In practice, most PEMS are created
ability of the process instruments is of the DOE is a 24-hour-per-day exer- for emission units that do not have
more important than the accuracy of cise. All operating shifts need to be in CEMS installed. In such cases, firms
the process sensors. tune with moving the emission unit specializing in stack testing using in-
52 CHEMICAL ENGINEERING WWW.CHE.COM JULY 2006
�������������������������� �� �
���� � � �����
�����
�� ���� � � �� ���
��������������������� � ���� � �� ��� ������

��� �� ��
����� ��� ��� ��� ������ �
��������������������������

����������
���� � �
�������������� � � � � ��� �
��� ��
����������
������������
���
���
��������������������
�����
��������������������

��

��
��

����
������������������

����
����
��

� ���
��
��

�� �
�����

��
��������� ��
��
����� ��
��

��
�����

��

��
� ���

��

��

�� �� ��
���

�� ���

�� ��
��

�� ��
��

�� ��
��
��
�������������������������� �

�� ���
��

��
����� ��

��

��

��
����������

��
� ��
�����


��������������������������

��������������

��
��
������������
������������������ FIGURE 7. 3-D plots are helpful to ascertain the relationship be-
�������������������� tween combined process parameters and emissions
����������
operations with suffi-
�������������������� technique is to use the process sensors
cient accuracy.
��������������������� that have a known relationship to the
�����
Periodic DOE re- pollutant and to add sensors until the
view. After the first accuracy becomes sufficient.
day of experiments, Evaluating offline accuracy. An
�������
����� the PEMS engineer offline evaluation of the PEMS is per-
�����
should evaluate ex- formed throughout the PEMS model-
� � ��� ecution of the DOE. building exercise. Two criteria for ac-
The engineer should curacy are used: 1) an r-correlation
access and combine analysis of the PEMS predictions to
FIGURE 6. For a complex system, software functions the process data and the reference-test-method; and 2) the
can force the neural network to learn the observed gains. sensor data to ensure overall accuracy of the PEMS predic-
In this case, the top graph shows incorrect learning that is
corrected on the bottom that the DOE is being tions in comparison to the findings of
followed, that process the actual reference-test-method.
strumental EPA reference test meth- data and emissions data are being col- The r-correlation coefficient is a
ods must be hired. As mentioned lected properly, and that the emission number that is a measure of the
earlier, dual instrumental EPA-refer- unit is behaving as expected. strength and direction of the correla-
ence-test methods should be utilized In some cases, adjustments to the tion between two variables, which in
to guarantee accurate pollutant data. DOE will be made. For example, if a this case is between the results from
The stack-testing firm needs to en- certain change in operation had no the PEMS and the EPA reference test
sure that the emission analyzers are effect on emissions, then it does not method. The closer r is to 1, the stron-
kept in a climate-controlled environ- need to be repeated later, and time ger and more positive the relation-
ment and are properly calibrated. and money can be saved. In most ship. The r-correlation coefficient for a
Warm-up of the analyzers and stabil- cases, however, the DOE proceeds as PEMS must be higher than 0.8.
ity of the testing-trailer environment scheduled with daily evaluation of its The r-correlation analysis does,
are needed. While the equipment can execution and performance. however, have some statistical limi-
be installed and producing numbers tations. For one thing, the dataset
within one hour, approximately four to Step 4. Building the model is invalid for r-correlation analysis if
eight hours are necessary to achieve Engineers with expertise in this area auto-correlation is present, or if the
stable, repeatable and accurate read- should be employed to build PEMS signal-to-noise ratio of the data from
ings. At a minimum, initial three-point models. The selection of the right the reference test method is too high.
calibrations and twice daily calibra- input sensors for the PEMS is criti- Also, there is a special limitation if the
tions should be performed. cal. Originally, neural-network PEMS emissions data do not change signifi-
Process data source. The facility’s were built following the “kitchen soup” cantly while a key input parameter
data historian is the source for process approach — all available process sen- (generally fuel flow) is varied: in this
data. Data historians can be set up to sors were initially used and then pro- situation, there is simply not enough
compress the data or ignore the data cess sensors were eliminated until a movement for the r-correlation test to
if sensor readings do not vary signifi- good mix of sensors and PEMS accu- have a meaningful result. When this
cantly. These features must be turned racy was achieved. This worked well, lack of change occurs, agencies will
off during the DOE because otherwise, but resulted in PEMS that used ten waive the r-correlation test require-
the process data will not reflect actual or more process sensors. The current ment. Figure 3 depicts a good r-corre-
CHEMICAL ENGINEERING WWW.CHE.COM JULY 2006 53
Feature Report

lation for PEMS predictions to refer- FIGURE 8. An installed


ence-test-method results. PEMS generally resides on
a server and communicates
The second check is designed to en-
directly with the DCS and/
sure that the accuracy of the PEMS or data historian
emission model, as required by regula-
tions, will be achieved. The actual test
reference margin of ±7.5% is plotted in
comparison to the actual PEMS pre-
��������������������
dictions. The predictions should fall ������������ ����������������
within this range for at least 90% of all ���������
data points to ensure that the PEMS ������������
�����
will meet the relative-accuracy require-
ments of 10% of the emission standard
or 20% of the actual emission value. necessary to force the neural network If the firebox temperature is used in
Figure 4 depicts a comparison between to learn the observed gains. the emissions model, then one or more
the actual test reference margin of Another evaluation of model behav- of the other temperatures are used by
±7.5% to the actual PEMS predictions. ior is to ascertain the relationship be- the sensor-validation model for the
Good predictions versus good mod- tween combined process sensors and firebox temperature. The validation
els. While good predictions are impor- emissions, especially two key process model for the firebox-temperature
tant, of greater importance are good sensors. This can be done with a three- sensor will also include other process
models that combine accuracy with dimensional (3-D) plot as shown in Fig- sensors, such as fuel flow, as deter-
model behavior. A model’s behavior ure 7. This graph represents the effects mined by the engineer.
is judged by evaluating the combined of fuel flow and steam-injection rate To develop appropriate sensor-vali-
effect of individual sensor gains with upon NOx emissions from a turbine. dation models, the engineer utilizes
process-sensor interactions on emis- As expected, the highest fuel flow com- the model-building and model-analy-
sion predictions. If the emissions bined with the lowest steam-injection sis techniques, as well as software
model reflects the gains correctly, then rate results in the highest NOx emis- functions that were used to develop
the emissions model will be robust sions. Alternately, the lowest fuel flow- the emissions model. While the devel-
and effective when in operation over rate combined with the highest steam- opment of the emissions model can be
the long term. injection rate results in the lowest NOx rather straightforward, the develop-
During the development of PEMS emission rate. ment of the sensor-validation models
emission- and sensor-validation mod- can be complicated because the rela-
els, the engineer must determine if Step 5. Sensor validation tionships may not be well-understood.
the neural network has learned the The dataset used to create the emis- Once sensor validation is modeled,
correct gains. If the neural network sions model is also used to create the engineer needs to determine the
has not, then the engineer can use one models of each sensor as a function of sensitivity or tolerance of the differ-
or more modeling techniques and soft- the other process sensors. These sen- ences between the predicted sensor
ware functions to influence the learn- sor models are combined with other output and the observed sensor out-
ing of the neural network. software to form the sensor validation put that will label a value as suspect
Figure 5 illustrates the gains for a system of the PEMS. and therefore not useable in the emis-
NOx-emissions model on a natural- When developing the sensor mod- sions model. The tolerances are read-
gas-fired turbine with steam injection. els, the redundancies of instrumen- ily selected by an experienced PEMS
This neural-network PEMS learned tation and the relationships between engineer. In some cases, such as in re-
the correct gains without the use of certain sensors should be considered. sponse to similar and numerous false-
other software functions. More process sensors are used by the sensor validation-system alarms, the
Figure 6 shows wrong neural-net- sensor-validation system than by the tolerances are adjusted after installa-
work learning (top) as compared to emissions model because redundant tion of the PEMS.
correct neural-network learning (bot- information, which is required for the When a sensor’s input to the emis-
tom) for NOx-emissions models that emissions model, is not necessary for sions model is suspect, then the sen-
were built from the dataset for a coal- the system. All sensor inputs into the sor-validation system will generate
fired boiler. While the output data for PEMS are evaluated by the sensor- a reconciled value for the emissions
both NOx emissions models were sat- validation system, not just the sensors model. For example, if a boiler is firing
isfactory, the neural-network model used in the emissions model. fuel at rated capacity and the process
on the bottom graph reflects the gains Combustion units monitor several O2 value is 21%, then the process O2
observed and will give performance temperature zones or points. For a meter is either in calibration or read-
superior to that of the model on the boiler, the firebox, economizer inlet, ing ambient air. When this occurs, the
top, both initially and long term. For economizer outlet and stack tempera- sensor-validation system will provide
a complex system, such as this coal- tures are usually monitored. These a reconciled sensor value to be used in
fired boiler, software functions are temperatures are closely correlated. the model.
54 CHEMICAL ENGINEERING WWW.CHE.COM JULY 2006
GENERAL REGULATORY ��������������� �������������� ���������������
REQUIREMENTS FOR PEMS ��������� ���������� ����������
�������� �������������
Submit PEMS Plan Ongoing PEMS ����������
�������
• Technical feasibility Quality Assurance ����

• Operating range of PEMS Requirements ������� ��������


��������
• Annual �������
Initial PEMS Certification comparison test
Requirements • Quarterly or �������
• Comparison test against more frequent �������
reference-test methods PEMS electronic
- EPA methods 3A, data assessment �������
7E, and 10 • Minute-by-minute �
- Test at three operating operation of � � � � � � � � � �� ��
conditions sensor-validation �����
• Sensor-validation system
system evaluation FIGURE 9. PEMS make good economic sense in comparison
to CEMS. Here total cost of ownership (TCO) is compared

The ability of the sensor-validation not been corrupted or replaced. initial relative accuracy for over 130–
system to reconcile sensor values and By this time, the PEMS personal NOx PEMS was 6.4%, and the ongo-
maintain good PEMS accuracy must computer (PC) has been installed, and ing relative accuracy was 3.2% — an
be evaluated. The EPA expects the fol- communications with the DCS and/or improvement of 50%. The main reason
lowing offline evaluation: data historian has been tested by the for the improvement was the ability
1. Determine the overall accuracy of plant’s IT department and the soft- to tune the models by augmenting the
the PEMS for the dataset that was ware vendor’s system-integration spe- original dataset with new data.
used to create the PEMS. Express cialist. Once the output tags from the PEMS continue to make good eco-
this in terms of relative accuracy as PEMS have been created in the DCS nomic sense in comparison to CEMS.
defined in 40 CFR Part 60, Appendix and/or data historian, then the PEMS Actual costs for NOx PEMS were com-
B, Performance Specification 2 can be installed and placed online pared to CEMS for a natural-gas-fired
2. Artificially fail the most important (Figure 8). boiler in the EPA’s “CEMS Cost Model,
sensor used to predict emissions. Version 3.” The total cost of ownership
This can be done by analysis of Step 7. Certification and (TCO) is shown to be substantially
the relationship between process documentation less for the PEMS than for the CEMS
sensors and emissions. Using the Once the PEMS is online, the stack- (see Figure 9).
reconciled values generated by the testing firm is brought back to vali- The neural-network based idea of
sensor-validation system, assess the date the PEMS and then certify the the chemical engineer in 1992 was the
effect on the accuracy of the PEMS PEMS as required by the EPA or other birth of a technology now recognized
over the entire dataset. If relative applicable regulatory agency. A report around the world. The relatively rapid
accuracy is less than 10%, then one of the certification test results is sub- acceptance of PEMS can be attributed
failed sensor can be tolerated mitted to the authorities and a PEMS to cost advantages found by industry
3. Retain the first artificially failed site guide is issued to assist in the un- and the accuracy that satisfies regula-
sensor. Artificially fail the next most derstanding of the PEMS. tors. Both are impressed by the reli-
important sensor and evaluate the Requirements for PEMS have been ability that these systems can offer. ■
effect on the relative accuracy of the established worldwide (see box). While Edited by Dorothy Lozowski
PEMS. If relative accuracy is less the format of the documents, the length
than 10%, then two failed sensors and breadth of the certification tests Author
can be tolerated and the timing of the submittals and Paul Reinermann is director
4. Repeat this procedure for the next tests vary, regulatory agencies have of environmental and regula-
tory affairs at Pavilion Tech-
most important sensor until relative the same, basic expectations of PEMS nologies, Inc. (10415 Morado
accuracy is greater than 10% performance and documentation. Circle, Austin, Texas 78759;
Phone 512-438-1575; Email
preinermann@pavtech.com).
Step 6. PEMS integration THE FUTURE OF PEMS At Pavilion, he assists in the
sales, marketing, business de-
Once the sensor-validation system PEMS are getting better over time velopment and project execu-
tion associated with Pavilion’s
and emissions model are created, the and continue to out perform regula- environmental product suite,
PEMS online software can be inte- tory requirements. The EPA recently consisting of Pavilion’s PEMS and Real-time En-
vironmental Management application. He holds
grated with the DCS and data his- evaluated initial and long-term PEMS a B.S. deegree in chemical engineering from the
torian. At this point, the electronic- performance by analyzing relative-ac- University of Cincinnati (1982) and has worked
exclusively in the field of air emissions testing,
data-assessment test is set up. This curacy-test audit results. The PEMS monitoring, compliance and enforcement since
test works by passing known sensor passed all audits and the relative graduation. Paul has been with Pavilion for over
11 years. Prior to Pavilion, he worked as an en-
inputs through the PEMS, gener- accuracy of the PEMS improved sig- vironmental consultant performing air emissions
studies, as EPA’s Chief of the Air Compliance
ating outputs and then comparing nificantly from the initial certification Monitoring Section enforcing air-pollution-control
these outputs to the known outputs. tests to the ongoing certification tests. permits and rules, and as an environmental spe-
cialist for an owner of independent power plants,
The electronic-data-assessment test In comparison to the requirement of overseeing a myriad of permitting and compliance
ensures that the PEMS models have less than 20% relative accuracy, the activities.

CHEMICAL ENGINEERING WWW.CHE.COM JULY 2006 55

You might also like