A high-speed Chemical Species Tomography system for
application in a multi-cylinder automotive engine
Paul Wright1, John L. Davidson1, Sergio Garcia-Castillo1, Charles GarciaStewart1, Stephen Pegrum2, Steve Colbourne2, Tim Litt3, Stuart Murray3,
Krikor B. Ozanyan1 , and Hugh McCann1
1
School of Electrical & Electronic Engineering, University of Manchester, UK
Roush Technologies Ltd., Brentwood, UK
3
AOS Technology Ltd., Melton Mowbray, UK
2
[email protected]
Abstract. High-speed Chemical Species Tomography (CST) of gaseous hydrocarbons in
gasoline fuel has been applied to a multi-cylinder automotive engine. A measurement grid
consisting of 27 optical paths has been implemented in one cylinder. A second-generation lownoise opto-electronic system has been developed, with laser intensity modulation at
frequencies up to 600 kHz. Robust optical access to the combustion chamber has been
achieved, yielding adequate performance on many beams after more than 2 hours of fired
engine operation. The measured data are discussed and preliminary examples of tomographic
images are presented. The concept is readily extendable to further species of interest. This
research has demonstrated that CST offers considerable promise to penetrate the fundamentals
of many dynamic chemical reaction processes.
1. Introduction
The spatial distributions of reactants are first-order determinants of the behaviour and performance of
many types of chemical reactor, including the internal combustion engine. Numerous imaging
techniques have been developed (refer to [1] for a wider review) to aid the optimization of both reactor
design and operating conditions but few are able to address purely gaseous subjects. Planar LaserInduced Fluorescence (PLIF) has been widely used in the study of fuel-air mixture preparation within
engines for around 20 years but offers very limited temporal resolution and requires the use of nonstandard fuels within a heavily modified engine. CST based upon near-IR absorption can potentially
address all of these shortcomings, albeit with much reduced spatial resolution, making it an attractive
complement to PLIF for in-cylinder application. Just as significant is CST’s ability to be implemented
on minimally modified multi-cylinder engines, allowing its use at all stages of the development
process, from initial flow bench studies to final design validation.
In a previous paper [2], we have discussed the practicality of applying near-IR absorption
tomography to multi-cylinder spark-injected (SI) engines. Here, we present the completed system as
implemented on a Ford 2.0 litre 4-cylinder port fuel-injected engine, with a preliminary analysis of
data obtained and tomographic images produced. A related paper within this conference [3] describes
laboratory tests using the system, with particular emphasis on image reconstruction techniques.
2. Hardware
The tomography system discussed here was developed for the DTI-Link project IMAGER and is a 32channel hard-field instrument, optimized for in-cylinder application. It comprises an OPtical Access
Layer (OPAL), which defines the measurement beam layout within the engine, and a rack-mounted
optoelectronic measurement system. The engine and OPAL were mounted on a dynamometer and
were connected, by fibre optic links, to the measurement system, located outside of the test cell.
2.1. Measurement system
The system implements a dual wavelength ratiometric transmission measurement. It employs two
fibre-coupled laser sources, a 1700 nm device for hydrocarbon (HC) sensing and a 1651 nm unit as a
non-specifically absorbed reference. Custom laser drive electronics provide temperature and current
control. Each source has a maximum output (in fibre) of 3 mW and can be directly modulated at up to
1 MHz. A 4-channel direct digital synthesis (DDS) module provides modulation signals, typically 300
kHz and 500 kHz, to the laser drivers and to the lock-in amplifiers used for signal recovery. The DDS
approach achieves high spectral purity (55 dB SFDR) and allows software control of the frequency,
amplitude and phase of the modulation/reference signals. An integrated WDM fibre coupler and 1 ×
32 splitter (Sifam Fibre Optics, UK) connects the sources to the launch channels of the OPAL, via
single mode fibre links. All fibre joints within the system either employ APC connectors or are fusion
spliced, to aid in the suppression of optical noise. The design of the OPAL itself is discussed in the
following section. Multimode fibres couple the collected light from each receive channel of the OPAL
to a bank of 32 receivers, based on extended InGaAs photodiodes (G8421, Hamamatsu Photonics).
The receiver design employed adds little excess noise (typically 3 dB) to the shot noise of a full scale
transmission signal, although its relative contribution will inevitably increase as fouling of the OPAL
reduces the collected signal power [4]. Each receiver feeds a pair of hybrid lock-ins, which recover
and sample, with 12-bit precision, the demodulated transmission signal at each wavelength.
Centralised control of the ADC timing allows essentially simultaneous capture of all 64 measurements
within each frame, at up to 105 frames per second. The use of such a high sampling rate has significant
advantages in this application: 1) transient phenomena can be accurately measured; 2) post-processing
can trade a reduction in effective sample rate for increased precision; 3) the anti-alias filter cutoff can
be located well above the (software) filtering used to set the lock-in detection bandwidth, ensuring
accurately matched filtering of the data to be ratioed. A high-speed, PLD-based, digital data controller
manages the transfer of data from each lock-in to a PC, via a DAQ card (PCI-6534, National
Instruments), as well as providing an interface for a rotary encoder, connected to the engine
crankshaft.
2.2. OPAL
Collection of transmission data requires accurate and highly stable positioning of the launch and
receive optics within the combustion chamber; acceptable beam alignment must be maintained during
rapid variations of both pressure (0 to 30 bar) and temperature (-30 to +90 Celsius). This led the
authors to adopt a relatively thick (≈5 mm) and rigid OPAL structure (Figure 2), very different from
the ‘optical gaskets’ used in flame front studies [5]. This structural independence allows the beam grid
to be developed, and indeed operated, wholly outside of the engine. The launch and acceptance angles
are controlled by collimation optics embedded within the structure of the OPAL, which helps to ensure
that the hard-field assumption remains valid, even in the presence of scattering. The OPAL replaces
the uppermost portion of the engine block, so as to maintain the correct compression ratio, and
includes all of the various oil and coolant galleries present in the standard engine. The optic fibres are
located in protection channels within the OPAL, exiting via two ports on its perimeter into protective
sleeves. This arrangement has proved robust; the current OPAL has endured several hours of both
fired and motored (engine rotated under dynamometer control, without combustion) operation, at loads
and speeds from unloaded idling up to 8 bar brake mean effective pressure (BMEP) at 4000 rpm.
Within this range, the sealing integrity and thermal performance of the engine are not noticeably
impaired, in keeping with our general aim of minimal modification of engine form and function.
Figure 1 The OPAL during manufacture and in conceptual form. Fibre protection channels, coolant
galleries and a fibre entry/exit port (upper right) are all clearly visible.
3. Data collection and analysis
An extensive program of both fired and motored testing was carried out over many load and speed
combinations. This was greatly assisted by the use of an advanced dynamometer facility that allowed
both steady-state and transient studies to be performed. Complementary measurements, including
cylinder pressure and exhaust gas analysis, were also made.
In common with other in-cylinder measurement techniques, the raw data require considerable
preprocessing prior to interpretation. Following time-domain filtering, the data are re-expressed as a
function of crank angle before intra-cycle referencing, in which all measurements from a particular
engine cycle are normalized to some reference location within that same cycle. This is usually
followed by further referencing, for example comparison with data obtained during, otherwise
equivalent, motored operation. This final step has proved useful in removing various artefacts from the
measurement data that are unrelated to the HC distribution, for instance those arising from thermal
expansion of the engine block. Determination of the optimum combination of referencing strategies
for a given dataset is the subject of ongoing research.
1.05
Ratiometric data value
Ratiometric data value
1.7
1
0.95
0.9
0.85
0.8
1.2
200
250
300
Crank angle (degrees)
350
200
250
300
350
Crank angle (degrees)
Figure 2 Example of data after filtering, transformation to crank angle form and intra-cycle referencing.
The data used are taken from 16 successive cycles of fired operation. The left hand plot shows all cycles
overlaid on the same axes, whereas the right hand presents successive cycles offset to allow the
repeatability to be examined. TDC corresponds to 360 degrees in these plots.
Figure 2 shows ratiometric data (T1700/T1651), collected from 16 successive cycles of fired operation.
Intra-cycle referencing has been applied to force to unity the average value in the crank angle range
200º to 220º. A high degree of repeatability and low noise is evident across all cycles. A decrease in
repeatability is seen as top-dead-centre (TDC) is approached but this is absent from the corresponding
motored data, suggesting genuine variability in the HC distribution late in the compression stroke.
4. Tomographic reconstruction
From a tomographic perspective, this is an unusual problem. The physical constraints imposed by
operation within an engine not only limit the number of beams but also preclude any regularity in their
distribution. An essential element of the IMAGER project was the development of reconstruction
techniques suitable for application to such a severely underdetermined problem. The images presented
here have all been produced using the resulting median filtered Landweber technique [3]. Figure 3
shows a sequence of 16 images, produced from data collected on an engine running at 2000 rpm, and 2
bar BMEP load. The images cover the period of the compression stroke between 66º and 21º before
Figure 3 Sequence of 16 images reconstructed from data taken during fired engine operation (2000 rpm,
2 bar BMEP load). The data were taken between 66º and 21º before TDC. Image interval 3º CA / 0.25 ms.
TDC, at intervals of 3º of crank angle (CA), equating to temporal resolution of 0.25 ms. The data were
pre-processed by low-pass filtering at 17.5 kHz, followed by conversion from temporal to angular
form and averaging of 3ºCA blocks. An intra-cycle referencing strategy was then applied, using a
datum taken early in the compression stroke, when the HC concentration is low. Subsequent images
within a given cycle are therefore relative to the conditions present during this reference period. The
measurements were then referenced to a corresponding motored dataset to remove background effects
not associated with the fuel vapour. No averaging across cycles was performed. These images were
generated using data from an OPAL with a significant build-up of combustion residues and other
contaminants. A qualification test was therefore applied to reject data from beams which exhibited low
signal quality, for instance poor SNR. The images presented were reconstructed using ratiometric
transmission data from just 12 of the 27 available measurement paths. This approach helps to prevent
the resulting image being impaired by the inclusion of misleading data and is likely to be of great
practical value as it increases the operational period possible before cleaning of the OPAL becomes
necessary.
A smooth progression from frame to frame is apparent in Figure 3. It should be emphasized that
this is due to the highly repeatable nature of the measurement, rather than being the result of the
applied temporal smoothing. HC specific absorption is initially observed only at the left (inlet
manifold) side of the cylinder but as the piston rises a more widespread, and uniform, charge
distribution evolves. The final four frames all suggest a relatively homogeneous HC rich region
extending in a broad band across the measurement plane. In this engine, the spark plug is coaxial with
the combustion chamber, with the point of ignition located 9.5 mm above the centerline of the OPAL.
The ignition timing is controlled by the engine management module but 18º BTDC, i.e. immediately
following this image sequence, would be typical for this load condition. In this case, therefore, charge
preparation appears to be fairly good. Additional work is necessary to establish the extent of any
artefacts arising from preferential attenuation of the 1700 nm by means other than HC vapour within
these images; possible mechanisms of this type include thin-film effects from lubricant or fuel films
covering the OPAL.
Figure 4 Sequence of 4 images corresponding to those in the leftmost column of Figure 3, from data
collected 10 engine cycles (0.6 seconds) later. Image interval 12º CA / 1 ms.
Figure 4 shows four images generated, in an identical fashion to those in Figure 3, from data
collected ten engine cycles later. The frames selected are at 12ºCA intervals and correspond to those in
the leftmost column of Figure 3. Absolute overlay is not expected as inter-cycle variability is
characteristic of internal combustion engines. Indeed, the variance observed in the measured pressurevolume diagram is a commonly used combustion diagnostic. Consistent trends are however evident,
with three of the four pairs of frames being in close agreement. Although the images in Figures 3 and
4 should be considered provisional, they do show that the data and methods now available yield results
that are both repeatable and physically plausible.
5. Conclusion
It is not the authors’ intention to present the images in this paper as a completed work but rather to
show that very significant progress has been made towards in-cylinder HC mapping by CST. Robust,
sustained optical interrogation of the combustion chamber has been achieved, resulting in a set of data
with unprecedented temporal resolution. The ability of the OPAL to be cleaned and reused repeatedly
has been established. The optomechanical performance of the OPAL is encouraging, with many
channels performing well. Stability of alignment will always be a concern in this application as
modern engine blocks are known to flex significantly in operation, due to their lightweight
construction. Experience gained during production of the current OPAL should enable improvements
in this respect in future versions and some progress has been made in the use of referencing to
compensate for moderate variations in alignment. The techniques developed for OPAL manufacture
are likely to be of use in other situations requiring compact optical access to harsh environments.
Fouling of the optics is also inevitable but experience to date is very positive with substantial periods
of operation proving possible. Future advances in laser technology at these wavelengths should
increase the available optical power, which would help in this regard.
Considerable effort is still required in the area of data analysis if the available measurement data
are to be used to best effect. At the time of writing, only a small fraction of the experimental data has
been examined so a great deal remains to be learned about the generality of current approaches.
Ultimately, the aim must be an algorithm requiring little or no user input, with decisions about the
most appropriate processing strategy being made either based upon the test conditions or on the
characteristics of the data themselves, but the immediate goal is a set of tools to support user-led data
exploration, processing and reconstruction.
Image artefacts are to be expected in such an underdetermined problem but we have shown that
useful images can be obtained from a very small number of measurement paths, using a median
filtered Landweber reconstruction. Moreover, we have demonstrated the implementation of sufficient
paths within a minimally-modified 2.0 litre vehicle engine.
Many of the difficulties associated with in-cylinder implementation of CST have been overcome.
Assuming that remaining issues in data processing can be successfully addressed, an opportunity
clearly exists to provide engine designers with previously unobtainable data on the dynamics of the
mixture preparation process.
Acknowledgements
The authors wish to acknowledge the support of the UK Engineering and Physical Sciences
Research Council and the Department of Trade and Industry under the IMAGER project. Particular
thanks also go to Steve Gratze and Steve Woodard at DTI, and to the many staff at all three
institutions who have contributed to this work.
References
[1] Scott D and McCann H, Process Imaging For Automatic Control, CRC Press, 2005
[2] Wright P, Ozanyan KB, Garcia-Stewart CA, Carey SJ, Hindle FP, Hurr W, Pegrum S,
Colbourne S, Turner P, Crossley SD, Litt T, Murray SC, and McCann H, Toward InCylinder Absorption Tomography in a Production Engine, Appl. Opt. 44 (2005) 6578-6592
[3] Davidson JL, Garcia-Stewart CA, Ozanyan KB, Wright P, Pegrum S, and McCann H, Image
reconstruction for Chemical Species Tomography with an irregular and sparse beam array, to
appear in Proceedings of Photon06, Institute of Physics, 2006
[4] Wright P, Ozanyan KB, Carey SJ, and McCann H, Design of High-Performance Photodiode
Receivers for Optical Tomography, IEEE Sensors J. 5 (2005) 281-288
[5] Philipp H, Plimon A, Fernitz A, Hirsch A, Fraidl G and Winklhofer E, A tomographic camera
system for combustion diagnostics in SI engines, SAE Technical Paper No. 950681, Society
of Automotive Engineers Inc, 1995