Donald Lyon, Monitran Technology, UK
Within the industry there is a continuous drive towards predictive maintenance, achieved through monitoring the health of plant equipment and machinery. As has been the subject of many a technical paper, by continually monitoring the condition of equipment and machinery it is possible to receive the earliest possible warning of potential failures, and thus avoid unnecessary downtime. It is also possible to squeeze as much life as possible out of parts and components, as opposed to replacing them on a set schedule irrespective of use.
However, there is one piece of equipment an essential item in many industrial applications that is difficult to cater for in terms of implementing maintenance based on its condition or health: the turbine be it gas or steam driven.
The majority of turbine failures occur because of blade defects; either developing over time or as a result of instantaneous damage caused by a foreign object. The earliest warning of a failure takes the form of blade vibrations, caused by dynamic loads on the blade. These loads can be generated by various mechanisms such as rotor imbalances, varying blade tip clearances (usually caused by non-concentric casings or orbiting of the main shaft), distortions in the inlet flow (usually caused by irregular intake geometries) and stationary vanes or struts up or downstream of the rotor blade.
During the manufacture of a turbine, a common method of assessing blade vibration relies on the use of optical probes mounted in the blade casing assembly. The principle on which the optical system operates involves the focusing of a narrow laser light beam onto the passing blade tip. As the blade tip enters the path of the light beam, light is reflected back to a photo sensor.
The intensity of the reflected light rises very rapidly as the blade passes and in the absence of any structural vibration, the time for the tip of a particular blade to reach the optical probe, called the ‘blade arrival time’, is dependent on the rotational speed alone. However, when a blade is vibrating its arrival times are dependent on rotational speed and any displacement due to vibration.
Such optical systems produce clear, accurate results and for turbine development it would be fair to say they are the de facto standard. However, because a clear optical path is required between the turbine casing and the blade tips, the use of optical probes for blade vibration analysis is not a realistic proposition for the long-term, in-field monitoring of blade health because contamination from dust and exhaust gases rapidly degrades signal quality.
There exists therefore a strong motivation to find an alternative to optical probes one which produces optical system quality continuously. Thankfully, research conducted in the aerospace industry, where jet engine health is of paramount importance, has led to the development of such a solution.
Eddy current sensing
In addition to the use of optical systems, blade arrival times can also be measured using other non-contact methods, including the use of capacitive, high frequency pressure and eddy current transducers. In 2002, a series of trials were conducted, at QinetiQ’s turbine test facility in Farnborough, UK, to compare these three options against an industry standard optical-based system. The trials revealed that of the three technologies the eddy current sensor showed the best promise for further development.
Eddy current sensors are most commonly used for non-contact proximity and displacement measurements. Measurement accuracy is high and ruggedized versions are often used in contaminated environments. In a turbine application, a major advantage the eddy current sensor has over the other probe types is that it is possible to take blade passing data through the turbine casing; though there is significant attenuation of the target signal.
In QinetiQ’s trials, bench tests were initially conducted on a standard off-the-shelf eddy current sensor. The sensor chosen had a reasonable range, while having overall dimensions of an acceptable size: in particular the sensor head was not sufficiently large as to cause mounting problems or to run the risk of detecting more than one blade simultaneously.
The maximum range of an eddy current sensor is equivalent to about half the diameter of its coil. For the trials, a range of up to 7 mm was required, so a sensor with a coil diameter of 12 mm diameter was chosen.
The preliminary tests were carried out on a small rotating test rig which used a simple shield mounted between the sensor head and the rotor to act as a casing. The shield could be moved to adjust the gap between the rotor and the shield, and the gap between the shield and the sensor.
For the early tests the material used for the shield was aluminium, and it was demonstrated that it was possible to detect blade passing events through casing materials up to 2 mm thick. Next, it was decided that a more realistic evaluation of the sensors should be carried out by mounting the sensors in a steel pocket to evaluate how they might perform when mounted in an engine casing.
It was found that the pocket design and material had an effect on the sensor performance, and that through-casing measurements did not produce good quality signals when a shield was fitted. To overcome these problems, the pocket was redesigned with an air cavity around the sensor head. Also, the sensor coil was optimized for improved resolution, signal to noise ratio and range
Tests were then carried out with varying tip gaps, from 0.5 mm to 2.0 mm. Although the signal was small in amplitude, it was of reasonable quality, with very well defined peaks when the blades passed by.
real engine testing
With a greatly improved bench tested sensor it was decided to embark upon a series of engine trials and, as chance would have it, a 24-blade AE3007 engine was on test at Rolls-Royce Indianapolis. A number of tests were carried out, with the engine running at 8700 rpm, on the first stage fan rotor of the engine. Figure 1 shows a schematic of the test cell and the instrumentation configuration. Four sensors were mounted into the fan casing; within ‘pockets’ which were manufactured in stainless steel and with varying base thicknesses in order to test the sensor’s ability to measure blade passing data.
Figure 1: A diagram of the set up for the fan tests
A number of tests were conducted to compare shield thicknesses and Figure 2 shows a sensor output through a 1.0 mm shield. It can be seen that there are significant differences in peak values [blue trace], indicating variation in blade tip clearance with the casing. Note: the red trace is a ‘once per revolution’ signal measuring the engine low pressure shaft speed using a speed encoder.
Figure 2: The sensor’s output through a 1.0 mm thickness casings following reduction of the tip clearance
Similar results were obtained using 1.25 mm and 1.5 mm shields except that all peaks were slightly lower. It was still possible to count all the blade passings, despite the extra thickness of shield.
After the acquisition of blade passing data through a shield, the ends of the pockets were machined off to allow blade passing data to be taken without any intervening material between the sensor head and the blade tips. Figure 3 shows one of the modified pockets prior to being fitted back on to the engine, and Figure 4 shows data from one sensor taken at full speed. Not surprisingly the output is much larger with no shield present and the blade passing is clearly visible. In addition the base noise level is reduced in comparison to the ‘through case’ data. In order to be of use in a blade tip timing system, data from sensors must be adequately monitored and conditioned so that the outputs to any turbine health monitoring system would be both reliable and repeatable.
Figure 3: The sensor pocket machined to allow direct line of sight from the sensor to the blade tip
Figure 4: The sensor pocket machined to allow direct line of sight from the sensor to the blade tip
Although four sensors were fitted into the engine fan casing, each was tested in a slightly different configuration. All the sensors successfully survived the engine running throughout the testing with no degradation of the signals becoming apparent. This shows evidence of the high mechanical integrity and immunity from contamination displayed by the sensors. It also indicates that for fan blade applications the sensors are sufficiently robust to be used without any form of metallic shielding.
Measuring real-time DATA
The eddy current sensors used in the tests proved sensitive to the tip clearance. However, the pulse shape generated by each blade is not significantly altered, suggesting that a repeatable amplitude independent triggering technique would be possible.
Although the sensitivity to clearance of the eddy current sensor offers useful additional information that is not provided by optical probes, amplitude variations lead to increased uncertainty in the determination of blade arrival times when using simple threshold triggering. The solution was to develop a triggering technique more suited to the eddy current sensor output signal.
Blade arrival times calculated from an amplitude-corrected signal showed promise and it was decided to embed the correction into the signal-conditioning hardware, so it could be performed in real-time. The signal conditioning system produced a square wave output, which was fed into an acquisition system that used a threshold triggering system. Figure 5 shows the raw and output signals from the signal conditioning system, with the output pulse occurring on the falling edge of the raw signal.
Figure 5: Falling-edge triggering technique, producing a relatively square-wave output
The conditioning was performed using analogue electronics however a digital signal processor (DSP) based system is currently under development that will improve the accuracy of the timing pulse. Also, with tip timing systems generating terabytes of data, even for just a few hours worth of recording, a DSP will be able to autonomously process in real-time what would otherwise require several days of analysis.
Accordingly, a DSP-based system will be able take the blades’ time of arrival data and process it into blade deflection information, and output real-time blade health in a typical traffic light format: with green indicating that all blades are fine; amber indicating that at least one blade has exceeded a predetermined limit of deflection or vibration; and red indicating imminent failure – and that the turbine should be shutdown.
Effectiveness for tip-timing applications
During the trials the eddy current sensors demonstrated their effectiveness for tip-timing applications. They also proved to be extremely robust and suitable for through-casing applications.
Further development of the signal conditioning electronics, specifically in terms of allowing the triggering correction to be carried out in real time, will deliver eddy current probe based tip timing acquisition and analysis on par with optical systems – but with the benefit of contamination tolerance in long-term use. The users (the maintenance engineer) need not be an expert in the field of tip timing. A DSP-based system will remove the need for traditional tip timing acquisition and analysis systems, which are very expensive.