Technologies

The covalent chemical bonds between atoms in all small molecules can vibrate at known specific resonant frequencies determined by the atomic weights of the atoms and the strength of the bond e.g. the C-H bond in methane (Fig. 1).  If light of frequency close to the resonant frequency of the bonds passes into the gas, it will excite the vibration, be absorbed and the transmitted power will decrease.  The absorption will be strongest and transmitted power will be lowest when the light frequency exactly equals the resonant frequency.  Imagine a situation where laser light is incident on such a gas and we tune (vary) the frequency (Wavelength) of the laser output and measure the transmitted light by a photodiode (Fig.2).  The transmission signal will decrease as we approach the resonant frequency, reaching a minimum at that frequency and then increase after passing it (Fig. 2).  The resulting signal is referred to as an absorption line of that gas.  The line-shape and strength (depth) is determined by the gas concentration, C, the pressure, P, and the temperature, T.  Simplistically, for fixed temperature, the depth is strongly dependent on concentration and the width is dependent on pressure. Hence, for fixed and known temperature we can measure C and P or for fixed and known P we can measure C and T.  If we measure more than one absorption line with different temperature dependence, we can measure all three. 

The vibrations of the molecular bonds can combine with molecular rotation at varying speeds / energies leading to many absorption lines associated with particular vibration-rotation combinations. The study of such absorption line is known as spectroscopy [1-3]. Figure 3 shows the rotation-vibration absorption spectra of CO₂ and water at ambient temperature and 600°C at light wavelengths just less than 2000nm.  Note that this can result in overlaps between line sets at specific frequencies and that we need to choose the lines we target for measurement to ensure no cross-species interference (Note that the CO₂ line selected for measurement in Fig. 4 does not overlap with any water line). 

Figure 4 shows simple measurement systems with a laser beam passing through the gas to be measured, including one where the light is delivered to the target measurement zone over an optical fibre.  The laser output frequency / wavelength is then tuned over the region of an absorption line and the transmitted power is measured by the photo-diode, recording the line-shape function. The measurement process is as follows.  Firstly we obtain from the literature or, if unavailable, we measure the spectroscopic parameters (line-strength and pressure broadening coefficients) of the target gas as a function of temperature over the temperature range of the measurement zone.  Using these, we develop an accurate model of the line-shape of the target absorption line as a function of concentration, pressure and temperature.  The model is then tested against laboratory measurements for known concentration, temperature and pressure.  In the in-situ real measurements one of temperature or pressure will be known and we record the measured line-shape function.  For known temperature (pressure), concentration and pressure (temperature) are varied in the model until the simulated line-shape and strength (depth) matches the measured line and the values that result in a match are the measured values (See Fig. 5).

 If we can measure two lines with different temperature dependence, we can determine temperature from the ratio of the signal amplitudes (absorption strengths) – see Figure 6.

In TDLS [4], the laser is tuned repetitively over the line by applying a saw-tooth (repeating ramp) current to the diode laser (Fig. 7) and recording multiple measurements of the line-shape which can be averaged over many scans to clean up the signal to noise ratio (SNR).  The laser output power increases with the increasing current in each ramp phase and the final line-shape is recovered by normalising, point by point, the recorded signal to the background intensity signal. The measurement process using line-shape models is as described above (Section 2.1).  This is referred to as direct TDLS as the line-shape is recorded directly at the output of the photodiode [5-8].  However, this approach has significant disadvantages in terms of improving signal to noise ratio.  Since the signal is recovered at low frequencies, at or near DC, it suffers from 1/f noise, laser intensity and phase noise and receiver noise (probably thermal).  More significantly, for gas turbine engine measurements, any noise arising from beam propagation issues in a turbulent environment such as dynamic beam wander off and on the receiver, beam break-up or mechanical vibration are particularly destructive.  

To beat the noise issues an advance on direct TDLS was developed referred to as wavelength modulation spectroscopy.

In WMS the centre wavelength of the laser is repetitively scanned over the line using a saw-tooth current function, as before, but with a high frequency sinusoidal modulation (dither) current superimposed on it (Fig. 8, 9). As the laser centre wavelength traverses the line, the sinusoidal current dithers the laser wavelength (frequency).  The interaction of the beam subjected to wavelength modulation (WM) with the absorption line-shape generates an intensity modulation (Fig. 9) at multiple harmonics of the fundamental modulation frequency [4, 9-16]. 

Transmitted signals at the modulation frequency are strongest and purest when the laser centre frequency is at a linear point of the line-shape and the higher harmonics are strongly generated when the centre frequency is passing over the highly non-linear regions. The intensity modulation harmonic signals arising from the wavelength dither are isolated in frequency and their amplitude measured by a lock-in amplifier (LIA – Fig. 8).  The output of the LIA then traces of the amplitudes of the WM induced intensity modulated (WM-IM) signals. For the LIA set to record at the fundamental modulation frequency the output trace (1f signal) is a close approximation of the first derivative of the line-shape and, if set to record the second harmonic (the 2f signal), the output (2f signal) is a close approximation of the second derivative of the line-shape (Fig. 9) and so on.

In the WMS approach we can choose the modulation frequency and hence the frequency detected by the LIA to minimise the noise and maximise SNR.  Figure 10 shows the measured noise spectrum on a recovered TDLS signal during a measurement on the exhaust of a gas turbine engine. If we make the measurement using a WMS approach to capture 1f or 2f signals above 400kHz we can, to a large extent, suppress the influence of all the noise sources.  The information we want is recovered from a high frequency signal and thus we avoid the worst excesses of lower frequency 1/f noise, laser intensity noise and receiver noise, and suppress the noise arising from mechanical sources (turbulence in the beam path, vibrations etc.).  Figure 11 shows a comparison of recovered signals from different approaches the black signal is from direct TDLS, the red is a 1f signal and the green signal is a 2f WMS signal from the LIA.  This illustrates clearly the power of 2f WMS to recover signals with high SNR from extremely noisy photo-diode outputs and why it became the preferred choice for many applications.

Notwithstanding the above, 1f and 2f TDLS-WMS presents some difficulties.  There is no intensity referencing and the signal amplitudes are strongly dependent on system parameters in a very complex way.  As a result, it is mostly used with calibration of recovered signals to those from known gas mixtures.  However, as system and component parameters tended to drift with time the integrity of stand-alone instruments was always questionable. In its simplest form, based on measurement of signal amplitude, they also suffer inaccuracies arising from varying pressure and temperature.

Since 2007 there have been many significant advances in TDLS-WMS to yield calibration free techniques that are very robust and ideally suited to stand-alone instruments in the field [14-16].  As well as the WM-IM signals carried by the beam onto the receiver, a direct IM signal is present from the intensity modulation of the laser by the high frequency sinusoidal drive current (known more widely as the residual amplitude modulation or RAM signal). The result is phase related distortion of the recovered IM and WM-IM signals at 1f or 2f with accompanying in accuracies. Due to the nature of the laser modulation, the WM-IM signals are phase delayed relative to the direct modulation IM signals.  Recognising this led to techniques to isolate the direct IM and WM-IM signals by phase selection or a phasor decomposition approach [14, 16].  These led to calibration free approaches based on the phase isolated direct IM signals, which enabled direct recovery of the absorption line-shape [14, 16].   Extraction of gas concentration and pressure (T known) then followed the same measurement procedure as for direct TDLS. The techniques required accurate, in-situ, characterisation of the system parameters [17] and refinements of the models particularly for high temperature use [18].  

A particularly useful calibration free approach [15] that provides intensity referencing normalises the WM-IM 2f signal to the 1f signal (takes the point to point ratio of the 2f to 1f signals).  The two signals contain the same common-mode optical noise components, including from the laser and the beam path generated noise from turbulence and vibration.  Hence, the normalisation process largely eliminates such noise from the processed signals vastly improving the SNR. The resulting signals, though, are strongly dependent on system parameters particularly the laser modulation characteristics, such as relative IM to WM-IM phase and the laser wavelength tuning coefficient.  This approach is made more accurate by the new understanding of phase related IM distortion and laser tuning, how to characterise them in-situ on stand-alone instruments and take account of them in the signal analysis [14, 16-22]. Even with such complex and careful signal generation, recovery and processing techniques, the recovered signals are uniquely related to C, P and T. Hence, with accurate models of the 2f/1f signals, accounting for system parameters and the variations of key spectral parameters, we can again simply fit the models to the measured signals by varying C and P / T in the model (with T or P known). This approach was used in a study of CO₂ in the exhaust plume of a GTE [23]. Measured 2f/1f signals for CO₂ in that study for two different thrust levels are shown in Figures 12a and 12b along with the theoretical simulations of these signals for those concentrations 

The excellent SNR confirms the efficacy of the normalisation in eliminating the common-mode noise.  The technique was used to investigate CO₂ concentration in the exhaust plume of a GTE as the thrust was cycled.  The results are presented in Figure 13 along with the engine temperature measured by a thermocouple.  The strong correlation in trends is evident as the engine thrust is cycled, again indicating the efficacy of the approach.

A further calibration free technique with intensity referencing and noise cancellation [24] was developed by the Strathclyde team in 2018.  This technique normalised the 1f WM-IM signal to the 1f isolated pure IM (RAM) signal, recorded in quadrature, giving excellent results for SNR and CO₂ measurement integrity [24] in measurements made on a GTE combustor at Cardiff University.  A sample measured signal compared with a simulated signal fitted to it is shown in Figure 14a.  For comparison the 2f/1f signal derived from the same measurement is shown in Figure 14b.  This  RAM normalised 1f-WMS approach [24] and the 2f/1f normalisation approach [15] will be used in LITECS.

By exploiting the absorption of light at carefully chosen wavelengths, Chemical Species Tomography (CST) yields images of the spatial distribution of a target species of molecules within engineering processes. Images can be obtained at thousands of frames per second (fps), and in favourable cases it’s also possible to obtain images of the spatial distribution of temperature. Since CST uses collimated light beams, it operates in a similar way to medical Computed Tomography scanning using X-rays, except that it doesn’t normally involve any moving parts. In the literature, CST is often referred to as “TDLAS tomography” (see below), or “hyperspectral imaging”, among other names.

CST for engineering applications became feasible following the advent of diode lasers and optical fibre technologies in the communications industry in the 1980s. Practical implementation of high-speed CST exploits the technique of Tunable Diode Laser Absorption Spectroscopy (TDLAS), which has been developed for many gas sensing applications in non-imaging mode. Typically, spectral absorption is measured along several dozen separate optical beams, which demands bespoke electronics. In common with X-ray CT, the reconstruction of images in CST uses mathematical methods from the field of Inverse Problems, and it also borrows heavily from data inversion methods used in other modalities of Industrial Process Tomography.

An overview of the development and status of CST is given in [1], and its challenges are unique:

• The target molecular type is usually in a mixture of gases, often at high temperature and/or pressure, demanding exquisite spectroscopic knowledge;
• Sophisticated optical access engineering is required in order to launch many light beams through the measurement subject, along with complex multi-channel Data Acquisition (DAQ) technology;
• With many fewer beams than X-ray CT, and by measuring an absorption spectrum rather than just attenuation at a single photon energy, the image reconstruction process is particularly challenging;
• Reconstruction of meaningful images requires that the spectral absorption measurements have very low noise content, i.e. high Signal-to-Noise Ratio, SNR, despite the difficult environment.

After the first demonstration of high-speed CST two decades ago [2], many research groups around the world have used it to image species and/or temperature distributions in a wide variety of engineering processes.

In LITECS, we are focusing on gas turbines and working to establish CST systems to provide:

1. Independent images of CO2 concentration and temperature (using absorption spectroscopy of water), in the exhaust plume;
2. Independent images of temperature and the concentrations of both CO and CO2 in the combustion zone;
3. Exemplar applications of the above under a variety of turbine combustion conditions, with various fuels.

The motivation for this research is to facilitate the development of improved environmental performance of gas turbines, which are deployed in a variety of industrial settings, notably in aviation and in power generation.

Automotive Compression Ignition Engine:

To help develop better diesel fuels, we worked with Royal Dutch/Shell to image the in-cylinder fuel distribution in an adapted Volvo D5 engine, running typically at 1,200 rpm. This 31-beam system exploits absorption at 1700 nm by the C-H bonds in the dodecane fuel molecules, compared with non-resonant attenuation at 1651 nm. Both wavelengths are fixed, to enable high frame rate, viz. 3,000 frames per second (fps). The image shown here [3] was taken at 27 crank angle degrees (CAD) before top dead centre, and 23 CAD after start of liquid fuel injection. The cylinder diameter is 81 mm and typical spatial resolution in the images is 9 mm. Each of the fuel lobes from the central 7-hole liquid fuel injector was clearly located, and their motion in the in-cylinder swirl flow was tracked by images every 3 CAD.

Swirl Flame Dynamics:

Swirl combustion allows very lean gas turbine operation, for improved fuel efficiency and reduced CO2 output. However, combustion stability is then a key safety issue. Working with Beihang University in Beijing, the dynamics of swirl combustion were imaged for the first time, using water absorption at 1343.3 nm and 1391.7 nm, measuring over the whole absorption peak in each case [4]. With 60 beams yielding temperature images at 83 fps, a precessing hot ‘crescent’ zone was revealed in the flame, as shown here. Flame extinction by the so-called Lean Blow-Out mechanism was shown to occur over a period of approx. 60 ms, suggesting that online control may be viable, thus enabling closer, but safer, approach to the lean limit.

Towards CO₂ Imaging in Civil Aero Engine Exhaust:

We have developed a system to image CO₂ in the exhaust plumes of large civil aero engines, in collaboration with Rolls-Royce, Royal Dutch/Shell, INTA (Madrid), and DAS Photonics (Valencia). The measurement system comprises 126 beams and, even though the exhaust plume is expected to be only about 1.4 m in diameter, the hostile environment means that the beams travel 7 m across the measurement plane [5]. The measurement method uses the so-called 2f/1f technique of TDLAS, requiring great sophistication in the DAQ system [6]. The system is now installed in Madrid and has taken data behind several engines. The example image shown here [7] is from a so-called “phantom” test, where a known distribution of CO₂ was produced by the combustion products from two gas burners (outlined here by the smaller two white circles). The task of image reconstruction faces several additional challenges here, since the beams are arranged to provide excellent spatial resolution in the plume region (i.e. within the largest white circle), with diameter only about 25% of the beam length. In this example, we have used a novel and highly efficient method [8] where 64 discrete cosine basis functions describe the concentration distribution, and the iteration process incorporates physical bounds and converges rapidly. The burner plumes are reliably located (the concentration units are arbitrary). Compared to the conventional Tikhonov method, our images are quantitatively and qualitatively superior.

Preparatory Research for LITECS:

Drawing lessons from the above projects, a benchtop system has been built in Edinburgh for CST of water, enabling research and development of many aspects of the technique. It uses 32 beams, as shown on the left, operating at the same two wavelength regions as in the swirl flame case above. This CST system is built upon commercial off-the-shelf (COTS) electronic sub-systems to create a sophisticated and flexible prototype data acquisition system [9] that enables investigation of various data sampling and multiplexing approaches, different spectral measurement techniques, and optimised signal processing methods. Similarly, a number of different image reconstruction approaches are under investigation, e.g. to enhance the properties of temperature distribution images obtained from the 2-wavelength technique [10].

The image on the rightabove here shows the concentration (mole fraction) distribution of water generated by two small burner flames, with the 32 beam-paths superimposed, and some of the surrounding Near-IR launch and receive sub-systems. The 2f/1f spectral measurement technique was used, and image reconstruction was achieved by the conventional Simultaneous Algebraic Reconstruction Technique (SART).

1. McCann H, Wright P, and Daun KJ, Chemical Species Tomography, Ch.5 in  Industrial Tomography: Systems and Applications, (Ed. Mi Wang), Woodhead Publishing, 2015, DOI: 10.1016/B978-1-78242-118-4.00005-8
2. Hindle FP , Carey SJ, Ozanyan KB, Winterbone DE, Clough E, and  McCann H, Measurement of gaseous hydrocarbon distribution by a Near Infra-Red absorption tomography system,  J. Electronic Imaging 10 (2001) 593-600, DOI: 10.1117/1.1377306  
3. Tsekenis SA, Ramaswamy KG, Tait N, Hardalupas Y, Taylor A and McCann H, Chemical species tomographic imaging of the vapour fuel distribution in a compression-ignition engine, Int. J. Engine Research  19 (2018) 718-731 DOI: 10.1177/1468087417730214 
4. Chang Liu, Zhang Cao, Yuzhen Lin, Lijun Xu, Hugh McCann, On-line cross-sectional monitoring of a swirling flame using TDLAS tomography, IEEE Trans. Inst. and Meas. 67 (2018) 1338-1348 DOI: 10.1109/TIM.2018.2799098
5. Fisher E, Tsekenis S, Yang Y, Chighine A, Liu C, Polydorides N, Wright P, Kliment J, Ozanyan K, Benoy T, Humphries G, Lengden M, Johnstone W, and McCann H, A Custom, High-Channel-Count Data Acquisition System for Chemical Species Tomography of Aero-Jet Engine Exhaust Plumes, IEEE Trans. Inst. and Meas. 69 (2020) 549-558 DOI: 10.1109/TIM.2019.2895932
6. Lengden M, Stewart G, Johnstone W, Upadhyay A, Wilson D, Polydorides N, McCann H, Liu C, Enemali G,  Tsekenis A, Wright P, Kliment J, Archilla V, Velasco J, Sánchez-Valdepeñas J, Beltran M, Polo V, Armstrong I, and Mauchline I,  Recent Progress in the Development of a Chemical Species Tomographic Imaging System to Measure Carbon Dioxide Emissions from Large-Scale Commercial Aero-Engines, OSA Optical Sensors and Sensing Congress 2020, paper JM2F.3, DOI: 10.1364/AIS.2020.JM2F.3
7. Tsekenis S-A, Polydorides N, Fisher E, Chighine A, Wilson D, Humphries G, Lengden M, Benoy T, Johnstone W, Kliment J, Wright P, Feng Y, Nilsson LJ, Victor DA Prat, Jia J, and McCann H, Chemical Species Tomography of Carbon Dioxide, Proc. 8th World Congr. Industrial Process Tomography, ISBN 978-0-853-16349-7, Paper B17, Iguassu Falls, Brazil, September 2016
8. Polydorides N, Tsekenis SA, Fisher E, Chighine A, McCann H, Dimiccoli L, Wright P, Lengden M, Benoy T, Wilson D, Humphries G, and Johnstone W, Constrained models for optical absorption tomography, Appl. Opt. 57 (2018) B1-B9, DOI: 10.1364/AO.57.0000B1           
9. Godwin Enemali, Rui Zhang, Hugh McCann and Chang Liu,  Cost-Effective Quasi-Parallel Sensing Instrumentation for Industrial Chemical Species Tomography, IEEE Trans. Ind. Electronics (accepted), DOI: 10.1109/TIE.2021.3063963     
10. Yong Bao, Rui Zhang, Godwin Enemali, Zhang Cao, Bin Zhou, Hugh McCann and Chang Liu, Relative Entropy Regularised TDLAS Tomography for Robust Temperature Imaging, IEEE Trans. Inst. and Meas. 70   (2021) 1-9 DOI: 10.1109/TIM.2020.3037950