Failure of the large, high-performance filter banks in a gas turbine’s inlet can have two negative effects. Firstly, compressor blades can erode and have their lives shortened as dust passes through the turbine and impacts with them at high velocity. This adds to the plant’s service requirements and, in extreme cases, can cause catastrophic breakdown of the blades. Secondly, particulates can build up more quickly on the blades and cause a gradual decline in the turbine’s efficiency. This increases the fuel consumption of the plant as it works harder to produce the same power output.
Epsilon combines multiple images to remove unwanted reflected background noise in the duct
While traditional online monitoring can gauge the overall performance of the filter banks, identification of the precise cause of any degradation has required an offline visual inspection of each individual filter. If a potential ingress of dust is identified, for example by pressure differential instruments, each filter in the bank must be physically inspected to locate the problem — a time-consuming task when the inlet duct on a typical large gas turbine commonly approaches 350 m2 in area and contains over seven hundred pairs of filters. Also, a pressure-drop monitor will only help identify alarm conditions and provides no information on which filter has failed.
A better way to measure the condition of individual filters is continuous online monitoring for particles that have breached the bank. This method is better able to predict filters’ future performance, and although visual inspection will normally identify the location of a catastrophic failure of a filter, it may well, unlike continuous online monitoring, miss the simple filter degradation that occurs over time and precedes their catastrophic failure. Inlet filters operate at varying degrees of efficiency, gradually deteriorating at a rate that is not always linear.
Etr-Unidata’s non-intrusive monitoring system, Epsilon, quantifies how much dust is getting through the filter bank and identifies which section of the inlet filter is failing. It measures sub-ambient particle concentrations in real-time, at the levels associated with inlet air downstream of a filter bank, over the entire bank. It employs a laser scanning system and camera-based digital imaging to provide high-resolution spatial information on particulate distribution within the duct.
There are four key parts to Epsilon, none of which are intrusive, so the system poses no threat to the integrity of the plant: a laser whose power output depends on the minimum size of particle to be monitored; a computer-controlled projection system that comprises a small angled mirror and a precision stepper motor; several digital cameras, the exact number depending on the geometric and physical constraints of the inlet duct; and computer hardware and software to analyze data and provide a graphical interface for the user. The key data required for measurement of filter performance is the trend in the count of particles and their spatial distribution.
Typical configuration of the Epsilon laser and single detector
By reflecting the laser off a small angled mirror and onto the filters, the system is able to cover the whole of the inlet duct, even in very large ducts such as that of the GE Frame 9A gas turbine. A stepper motor rotates the mirror so that the system sequentially projects a fan of laser light beams in a single plain across the whole duct.
The system mounts on the outside of the inlet duct wall as close as possible to and on the clean side of the inlet filters. It projects the laser into the duct through the wall via a small optical window whose location is critical. Windows installed in a slightly different position allow the digital imaging to occur. The exact location of all of the optical windows is installation-dependent because, in general, gas turbine plants have unique geometries. The windows are normally installed during an outage period, the preferred time for system set-up, although this is not critical. No air purging system is required to keep the windows clean and free from dust or particulate build up because they reside in the duct wall on the clean side of the inlet filter. Even in the event that dust builds up on the surface of the window, no interference with the monitoring would occur as the digital camera is focused well beyond the window and into the duct. No hardware is located inside the duct.
Dust particles that pass through the laser beam scatter its intense light in all directions. This light is detected by the digital camera receiver unit, which creates 20 images of particulate presence per second. Imaging software then integrates the data from up to 18 000 images per full duct scan to allow the system to determine the particle count at thousands of points across the duct, count the individual particles (generating extremely accurate continuous read-outs) and map the particles for proximity or distribution profiling.
To create a meaningful image of particulate distribution from the captured images requires several steps. First, the image has to be corrected for the fish eye distortion that the cameras’ wide-angle lenses create. This adjustment occurs during set-up and is similar to, but a much more sophisticated application of, the technique that removes the keystone effect in LCD projectors. It only has to be made once for each camera in the installation and creates the correct rectangular shape of the duct.
Epsilon isolates the centres of scattered particulates using the Hough transformation technique to detect those particles that have been illuminated by the laser beam
Second, each laser scan has to be located within the image. Random and background noise then has to be removed from each frame to ensure that only the light scattered by the individual particles is counted by the system. Advanced digital processing software removes the noise, such as that caused by reflected light from the duct walls or internal framework. The system takes multiple high-speed images of each laser position so that noise, which is random in nature, can be cancelled out by combining individual images. This process also increases the light signal from the particulate.
A technique called Hough transformation then follows to detect the light generated by individual particles as they interact with the laser. This signal algorithm relies on the parameterization of the centres of the scattered laser light. The system plots the information it has extracted from the digital image and applies the Hough transformation algorithm to determine whether the scattered centres lie on a straight line or not. The technique thus distinguishes between random background scatter and scatter on a straight line, as caused by the laser. The system repeats this sequence for each position of the laser to build up an image of particulate distribution in the duct.
The graphical interface sits in the control room and consists of a high-resolution grid display that relates to the inlet filters. If a filter breakthrough occurs, an individual filter cell on the display changes colour in a traffic-light way as an alarm condition occurs. The operator would then be aware of which filter or group of filters has started to fail. The user interface also allows the operator to zoom in on the area where the breakthough has occurred and identify more precisely the location of the problem. The magnitude of the particulate breakthough can then be assessed and appropriate actions taken. Epsilon lets operators base their decisions on filter condition using real-time information on actual particulate levels inside the duct rather than having to rely on an indicative pressure-drop measurement.
The user display sits in the control room and shows a real-time particle distribution map
Epsilon’s ability to locate and focus on areas of filter deterioration and provide information on the on-going level of particulate in the duct has the potential to make huge savings in plant maintenance through reduced down time and better filter management. The system’s extreme sensitivity allows it to quickly identify and locate any potential filter or seal failure and hence prevent significant levels of particulate from entering the inlet duct and damaging the turbine blades.
It has also been suggested that Epsilon will give those operators who “wash” their compressor blades online to manage the rate of particulate build-up a much improved ability to determine when the cost of washing falls below the costs associated with the efficiency drop. This will, over time, allow the plant to run at higher efficiency. The resulting fuel savings will reduce the plant’s carbon emissions.
The system will also give operators in general an improved ability to monitor the efficiency drop associated with particulate build-up on the compressor blades and let them better optimize their maintenance cycles, especially if washing the compressor blades is a major determinant in the outage cycle.
Particulates in inlet air can burden plant with significant costs, which the Epsilon system can reduce. These arise because filters have to be replaced earlier than necessary, failed turbine blades have to be replaced and because the turbine consumes extra fuel as plant efficiency declines. There is also the opportunity cost of the extended down time that follows a failure. Operators therefore build an insurance policy into their maintenance scheduling, replacing filters earlier than necessary, and probably having more and longer outages than would be necessary if there were a reliable means to monitor the performance of individual filters and the build-up of particulates on the compressor blades.
The Epsilon system can also be of use in extreme situations, such as those that occur in the Middle East, where plant operators use reserved pulsed-air jet cleaning to manage high levels of ambient dust. Pulsing can cause individual filters to be dislodged, which will allow high levels of dust to enter the inlet filter system and ultimately damage turbine blades. Applications such as this require frequent visual inspection of the inlet filter house.