PROCESSWEST Magazine Online
News

Techology Profile: Why experts believe PEMS deserves a second look


November 6, 2016  


Print this page

It’s no secret fossil fuel energy producers are under the gun with so much emphasis on global climate change. Much of that pressure is the expectation for a significant reduction of emissions. And that means improving emissions monitoring. Two Western Canadian engineers believe there’s already an existing monitoring technology that can play a critical role in advancing the measurement and control of emissions in a cost efficient way.

That technology is known as predictive emission monitoring systems (PEMS), and it’s being used in a variety of areas around the world — including more than 18 states and local authorities in the U.S., along with Australia, Sweden, the Netherlands, Italy, France, Norway, U.K., Saudi Arabia and the United Arab Emirates, where PEMS has been either approved or is in the process of being validated.

The current emissions monitoring standard in Canada for stationary sources is called continuous emission monitoring systems (CEMS), which has been used to measure flue gas emissions since the 1970s. Simply put, a CEMS continuously extracts gas samples from a specific spot in a processing facility — for instance, at a natural gas or oil-powered electrical generating unit. The gas sample is filtered, transported, conditioned and sent to a gas analysis system where gas concentrations are measured, recorded and stored as data for further use.

Typically, CEMS has been used for pollutants such as NOx, CO, and SO2, although CEMS are also being used for hazardous emissions such as mercury (Hg), hydrogen chloride (HCl) and hydrogen fluoride (HF). Despite its wide use, CEMS has significant downsides — specifically, high capital costs, considerable maintenance and frequent operational failure under severe weather conditions.

The consequences can be significant because even small errors resulting from operational failure can lead to inaccurate reporting, missed deadlines, permit violations, financial penalties and, ultimately, lost profit. To reduce cost and improve data continuity, PEMS was developed to replace, or to operate in parallel with, CEMS by deriving emission concentrations from process and ambient condition data.

While CEMS takes a sample of media coming out of the emissions stack and analyzes it to determine what is coming out of stack, PEMS examines the characteristics of the source and ambient environment to predict the emissions through the use of algorithms. This software-based emission monitoring 38 PROCESSWest October 2016 PROCESSWest October 2016 39 takes readings from the processing plant’s existing fuel flow, heat value and mass-density monitoring equipment to calculate emissions.

It can also provide feedback to plant operators so they can optimize operations for both energy savings and emissions reductions. In addition to more accurate analysis, the parametric form of PEMS has noticeably lower installation and operating costs. For instance, according to CMC Solutions, a supplier based in Farmington, Mich., a PEMS can be configured, delivered and installed in 30 to 45 days. That includes a computer equipped with the DAS/PEMS software and interfaced to the boiler control system.

The installation and start up is typically a one-day process. In comparison, delivery for a CEMS takes 12 to 14 weeks, while installation can take seven to 14 days, depending on complexity and location. It also requires various skilled trades (electrical, mechanical, computer) to install ports, probes, umbilical trays, CEMS racks, gas cylinders, cylinder racks, gas tubing runs, drain and exhaust lines and interconnecting wiring. CEMS installation may cost more than the CEMS equipment, and normal start-up is two to four days, depending on complexity.

Time will tell if advances in artificial intelligence will create newer predictive PEMS systems that can be implemented as quickly and as inexpensively as a parametric PEMS. PEMS on natural gas turbines have been commercially available since the late 1980 and have been in operation at some B.C. compressor stations since the mid-1990s.

However, in Alberta and the rest of Canada, there’s been a noticeable general reluctance to adopt PEMS technology. Ke (Duke) Du, assistant professor of mechanical and manufacturing engineering at the University of Calgary’s Schulich School of Engineering, believes it’s the black box nature of the technology that is proving to be a major obstacle in mainstream acceptance of PEM. “Black box” applies to a device, object or even system whose inputs and outputs are visible, but whose internal structure or operation remains concealed.

“PEMS has been accepted in Europe, the Middle East and many states in the U.S., but in Canada the acceptance is low with only some cases in B.C. and a few in Alberta,” he says. “The PEMS model in Alberta is typically a black-box concept, so the environmental agencies are concerned how the technology inside works. “One of our objectives is to transform that black-box technology into open source so anyone is able to understand the technology used in PEMS and to encourage others to develop their own.

To a large extent, that basic technology, essentially, the methodology of data mining, is already open and accessible.” Du is teaming up with his engineer colleague Roger Ord of SNC-Lavalin Inc. in Victoria to dissuade the black-box perspective. The two of them are seeking to establish a pilot study to demonstrate current PEMS capabilities in Alberta and develop new PEMS expertise and/or products. Ord, who is director of acoustics, air quality and climate change for environment and geoscience, infrastructure, at SNC-Lavalin, agrees the black-box issue is preventing PEMS advancement.

“Proving the technology and the results used in the analysis process is not completely straight forward,” Ord says. “For instance, manufacturers of turbines that employ PEMS use a combustion algorithm, but the algorithms aren’t publicly disclosed because they’re considered intellectual property. In order to be granted a permit for emission controls, the provincial regulator must verify the accuracy of the control system. “Because of the black box situation, the only way to obtain a permit for PEMS is to run it in parallel with a CEMS to prove the actual numbers. So, it’s not necessarily just the owners of emitting facilities that are reluctant to install PEMS.

It’s also due to the low regulatory comfort level in Canada.” And that is the point of the pilot project, adds Ord. It’s to create an open software algorithm environment that can grow the awareness of how the technology works for all regulators and have them play an advisory role. But there is also another objective. “The technology itself has remained quite static because of the lack of regulatory acceptance,” says Ord. “But it has the potential to apply to a much broader range of emissions sources than is now happening.”

The most typical applications for PEMS are in gas turbines, boilers, furnaces and combustion engines. For those applications, emissions monitoring is quite predictive and simple. For a natural gas turbine, for instance, a PEMS is comprised of sensors for monitoring the ambient environment, including relative humidity, temperature and pressure, along with existing process information. There aren’t any CEMS components so it’s much less expensive.

“But the PEMS also has potential for such applications as duct burners, dryers, regenerative thermal oxidizers, process heaters, crude heaters, flares and incinerators,” says Ord. “For more complex sources, more analytical types of inputs are included — for instance, to measure the type of fuel being used. The PEMS will have to be tweaked, or be sophisticated enough to learn to predict emissions from those types of sources.”

Du and Ord have outlined the purpose of the pilot project, which they are moving forward in the following points:
• To provide an objective test of a PEMS accuracy and precision from first principles on an Alberta emission source of consequence;

• To provide hands-on experience to potential PEMS users, developers and regulators as to how it works, its strengths and weaknesses;

• To confirm the potential costs related to technology implementation, and;

• To assess PEMS potential application to other emission sources (i.e., flares, incinerators) and contaminants (i.e., CO2, CH4, VOCs) The first principle method referred to in the above objectives is one of the methodologies to be considered during the pilot study. It is based on thermodynamics, including reaction kinetics, stoichiometry, energy and mass balances to predict emission by combining the following parameters:

• Data available from the standard installed sensors,

• Physical insight (like geometrics) and reaction kinetics, and;

• Statistical methods for calibration on test measurements. A second methodology is called the datadriven model. It’s a mathematical model that fits a given set of input parameters to a corresponding set of measured emissions. There is no formulation of physical properties or relations in the models. It’s a model in which:

• The model needs to be trained on all levels of measured loads, and; • The calibration and validation are performed randomly during the entire measurement campaign. The data-driven model itself can use several types of technologies, including neural networks, multivariate analysis and statistical hybrid methods.

Ord points out neural networks have become increasingly regarded as one of the most advanced forms of artificial intelligence. Neural networks use what is known as back-propagation algorithms and were pioneered by Prof. Geoffrey Hinton with the University of Toronto’s Department of Computer Science. Hinton directs the program on Neural Computation and Adaptive Perception for the Canadian Institute for Advanced Research, and was primarily responsible for developing Google’s artificial intelligence deep learning program, which has now led to enormous interest and just as enormous amounts of financial investment in the technology community.

Multivariate Data Analysis (MDA) refers to any statistical technique used to analyze data that arises from more than one variable, according to CAMO Software AS, an Oslobased software company. MDA essentially models reality where each situation, product, or decision involves more than a single variable.

CAMO was founded in 1984 by a group of Norwegian scientists, and is considered a pioneer and leader in multivariate data analysis. Du and Ord’s pilot project is scheduled for a one-year period during which they’ll examine a number of emission sources and develop related algorithms using the above approaches and select the one which exhibits the greatest accuracy.

However, Du’s and Ord’s efforts won’t end with the pilot project. A further objective is to establish a network of (PEMS) excellence to increase the awareness of PEMS capabilities and appropriate usage in Canada and expand the application of PEMS by leveraging new technology and other capabilities.

To do this, they are looking for support from industry, government and academic institutions. Ord says they have already been approached by some businesses and government groups that could act in advisory roles.

About the author: Ernest Granson is a Calgary-based writer and editor, and a contributor to PROCESSWest.