Comparison between Pressure- and ion-Current-Based Closed-Loop Combustion Control Performance.
Combustion feedback is essential in the latest-generation SI engines, to achieve both maximum performance and efficiency and reliability. Knock intensity estimation function has recently become a must for Engine Control Units (ECUs), since external disturbances can modify the engine knock tendency during on-board operation. The effects on the knock intensity of some of the engine boundary conditions can be mapped or modeled, and compensated by the open-loop spark advance controller, but the remaining uncertainties (fuel octane number, engine ageing...) cannot be fully included in the base spark advance calibration, but they are, at least partly, entrusted to the on-board knock measurement and control system. Several sensors can be installed on the production engine to estimate knock intensity, such as accelerometers, ion sensing, microphones, and of course in-cylinder pressure, which can be considered as a benchmark for other systems, characterized by higher cost and lower durability .
Knock intensity control can be insufficient to realize maximum efficiency operation, since in knock-free operation, where the knock control is supposed not to react and the spark advance control is purely in open-loop, external environmental disturbances like air humidity , engine ageing, and cylinder-to-cylinder differences, which cannot be taken into account by the open-loop SA controller, can result in sub-optimal operation.
In knock-limited conditions, a protective controller (i.e., providing only negative corrections to the spark advance) is not pushing toward the maximum brake torque spark advance when operating with high octane fuel. Conversely, an aggressive strategy (i.e., calculating and applying also positive corrections) would require, apart from great confidence on knock measurement that is however needed, some guarantee (feedback) that the controller is not over-advancing the spark angle, for the same uncertainties explained above relating spark advance to combustion phasing. It is then clear that a feedback about the efficiency of each combustion event, in addition to its knock intensity, can be extremely useful and make the spark advance control potentially ideal. Moreover, the combustion control calibration effort could be reduced, since many of the corrections related to the accountable disturbances could be simplified or ignored, leaving them to the on-board closed-loop control. The well-known bell-shaped torque-CA50MFB (Crankshaft Angle at which 50% of Fuel Mass is burnt) characteristic, makes the CA50MFB and, equivalently, the pressure peak position reliable indicators of the distance from the maximum efficiency condition for the given operation . Pressure signal is of course perfectly suited for this purpose, but such signal is very rarely used in production applications because of the high cost and low durability of the sensor . Apart from in-cylinder pressure, there are two other signals that can be used to evaluate combustion phasing: ionization current and engine block vibrations. Ionization current sensing is a well-known technology that allows measuring a signal strictly related to free ions concentration, in-cylinder temperature, and therefore the combustion process . Substantially, it is a good surrogate of the pressure signal, even if with a lower SNR (i.e., signal-to-noise ratio). Accelerometers are used to measure engine block or cylinder head vibrations, which can be partly produced and transmitted to these sensors by the combustion process. This makes it possible to estimate knock intensity, since this kind of combustion is well characterized in the frequency domain; moreover, it has been demonstrated how this signal can be processed to estimate pressure-peak position . Anyway, the robustness of this last approach has to be further investigated and will not be considered in this work.
In this article, ionization current signal is processed to be used as input to a closed-loop combustion control strategy, as it contains all the required features listed above, and it can therefore provide the required information throughout the whole engine operating field.
The objective of this work is to demonstrate the capability of the ionization current signal to be used as the main input to a closed-loop combustion controller, both in knock-free and knock-limited conditions. Pressure signal is considered as a benchmark, to evaluate the ion-based control performance, and the control logic is the same for both controllers, to make them comparable. In a first stage, the closed-loop combustion controller has been developed and calibrated in a Model-in-the-Loop (MiL) environment, considering both pressure and ionization current signals. Then, the controller has been implemented in a Rapid Control Prototyping (RCP) system and validated with the real engine. Experimental tests were performed at the engine test bench, without further optimizing the controller calibrations, since the objective of this work was mainly demonstrating the capabilities of the ionization current to replace the pressure signal to fully control the combustion process.
Ionization Current Signal
During combustion, molecules get ionized by heat, making the charge conductive. By applying an electric field, it is possible to measure such conductivity and hence the ion concentration. The great advantage is that gasoline engines already have electrodes inside the combustion chamber, so no modification to the chamber is required. In production engines, the sensing system is generally integrated with the coil and it measures the current while applying a voltage difference between the two electrodes, immediately after the spark discharge event .
Ion Sensing Circuit
A schematic of the ion current sensing circuit is reported in Figure 1. On the left side, the arrows represent the current flow during the spark event, immediately after the IGBT opening. The inductive voltage generated in the secondary coil is used to generate the spark between the electrodes, but simultaneously it is used to charge the capacitor shown on the right. Once the spark ends (right side of the figure), the capacitor drives the ionization current through the secondary coil and the gap between the electrodes, with an intensity that is proportional to the gas conductivity and to the capacitor potential. The current intensity can be evaluated through the sensing resistance shown in the bottom right side of the schematic [6, 7].
Ionization is caused by the heat released during the combustion process, through two different mechanisms. One takes place in the flame front, where the oxygen reacts with the carbon and the hydrogen, rapidly passing through several intermediate ionized stages , and finally producing water and C[O.sub.2]. Since the flame front is supposed to be thin and moving across the combustion chamber, ion generation is localized and can be sensed through the spark plug electrodes only during the early stage of the combustion. This mechanism is usually called "chemi-ionization" [8, 9], "chemical ionization" , "flame ionization"  or "flame front" , and it corresponds to the first peak on the ion signal, after the ignition, as it can be seen in Figure 2.
The second mechanism of ions formation is related to massive thermal dissociation, concerning the whole charge, which is promoted by high temperature and it corresponds to the second peak of the signal shown in Figure 2. Temperature, pressure, and volume are closely related (e.g., the ideal gas law can be applied), so that a strong correlation is expected between Pressure Peak Position (PPP) and ion current thermal peak position [2, 3]. The angle corresponding to in-cylinder peak temperature, to which it should correspond the maximum ion concentration, is expected between the TDC and the peak pressure angle; but the sensing dynamics, the local temperature measurement, and probably other aspects very complex to consider result experimentally in a "thermal" or "post-flame" ionization peak extremely close to the peak pressure angle. Figure 2 shows how reliably the thermal peak can be used to estimate the pressure peak angle, and this is generally true for all engine operating conditions. This is the feature of the ion signal that will be considered in this work as an indicator of the combustion phasing, to be used as input to the SA closed-loop controller.
Depending on the operating condition, the thermal peak can be partially merged with the first peak, resulting in a harder recognition of the corresponding angle. Eriksson in  proposes several methods to systematically identify the thermal peak position. At high load conditions, which is the case of the presented study, the thermal peak is usually well defined and easily identifiable , so particularly smart algorithms are not required in this case. In specific, a built-in functionality of the indicating system, which calculates the maximum and the maximum position within a defined angular window, has been used.
In  the authors propose and test a closed-loop control based on the thermal peak position of the ionization current signal to enhance engine efficiency and emissions in cold start conditions. In  the closed-loop control strategy involves both the combustion phasing and the knock intensity estimation, but results are tested and described just for low load condition and during a slow coolant temperature transient. In  the authors quickly show how the closed-loop controller, at the test-bed, is able to control the thermal peak position to a step-changed target value.
Ion signal can be used to detect knock too, as widely described in the literature [4, 6, 14, 15]. The high-frequency pressure oscillations induced by knocking combustion are reflected on in-cylinder temperature and therefore on ion concentration, causing a corresponding oscillation of the ion current signal, as can be seen in Figure 3.
As for the pressure signal, oscillations are to be found during the expansion stroke, generally from the peak pressure position onwards, and their spectrum is related to the chamber resonance frequency range. The analogy between the two signals allows adopting similar algorithms, which are mainly based on high-pass filtering and windowing stages.
Several attempts at using this potential of the signal to control knock intensity can be found in the literature: in  a SA controller which evaluates knock intensity through ionization signal processing is proposed and tested, but just showing the controller behavior for one cylinder in a virtual environment; Guoming et al. demonstrate the effectiveness of a stochastic knock controller by using ionization current feedback.
Other Ion Signal Features
Ion current signal can be effectively used to detect misfire, since in absence of combustion the signal is null [17, 18]. The same signal can also be used to detect preignition, since an early heat release results in a flame-ionization before the spark . Further, some attempts have been made to estimate in-cylinder air-to-fuel ratio from the current signal , since the thermal ionization peak amplitude shows a non-linear correlation with this parameter.
There are other characteristics of the signal, in the earlier angles displayed in Figures 2 and 3, which can vary depending on the ignition and sensing circuit. They are shortly described here for the sake of clarity. In general, the ion current signal is available during the whole engine cycle, unless being biased by the ignition process. The first spike, during the compression stroke, is given by the start of the coil energizing phase, and the second spike corresponds to the spark discharge angle (the angular distance between such spikes is the dwell angle; see Figure 2). The interval between the spark discharge and the first rising edge is the spark duration, during which the current flows through the diode shown in Figure 1 in parallel to the voltage divider. As the spark ends, the remaining energy in the coil is responsible for the oscillating response that can be seen before the chemical phase.
The goal of this work is to resume and demonstrate the possibilities for closed-loop control of the ionization current signal both in knock-free and knock-limited conditions, by proposing a simple and effective control strategy which is able to satisfy the seeking of the MBT spark advance, while keeping the knock intensity at or under the desired level. Moreover, the proposed control strategy is used to compare the ion- and the pressure-based controls in terms of CA50MFB and knock targets tracking ability, so that a quantitative evaluation can be performed, allowing an estimation of the ionization current signal potential.
Experimental tests have been conducted on a V-8 3.8 liters GDI turbocharged high-performance engine (Table 1), equipped with ion sensing technology as standard production equipment. Several ion-based strategies are implemented on the production engine (misfire diagnosis, knock, and preigni-tion detection), but not the combustion phasing closed-loop control. The test bench is equipped with a Borghi & Saveri dynamometer, an AVL 733S fuel balance and Alma-Automotive OBI (On-Board Indicating), and charge amplifiers. Every cylinder is equipped with a Kistler 6045A piezoelectric pressure sensor in the test bench set-up, and pressure and ion signals of every cylinder have been acquired and sampled at 200 kHz.
In the first stage, both pressure and ion signals have been recorded during steady-state spark sweeps, to generate the database required by MiL activity (described below). 4500 rpm and high load (2200 mbar of intake absolute pressure) is the operating condition chosen for this phase of the project.
In the production layout, the ECU is receiving via CAN the ion-based indexes and the corresponding SA corrections calculated by the dedicated module (see left side of Figure 4).
For real-time implementation, the ion module has been replaced by a rapid control prototype, so that the protective knock-based corrections calculated by the ion module are ignored and replaced by the strategy implemented in the RCP. At the same time, the RCP is connected via CAN to the test bench combustion analyzer, which provides standard pressure-based indexes (CA50MFB and knock intensity, the latter measured via MAPO -Maximum Amplitude Pressure Oscillation ) and customized ion-based indexes (AThP and IntIon) to feed the control strategy. The RCP is implemented in Miracle2, a National Instruments hardware-based multi-purpose platform by Alma-Automotive. The LabVIEW code running on the platform is integrated by a compiled Simulink model, which contains the control strategy described in the next paragraphs.
Spark Advance Controller Development
The controller task is to manage each cylinder spark advance to achieve a pre-defined combustion phase, while guaranteeing that a pre-fixed knocking intensity level is not largely nor frequently overcome. The possible outcome, in terms of engine performance and costs optimization, is remarkable: combustion efficiency could in fact be optimized under all engine operating conditions, cylinder by cylinder, without requiring significant calibration efforts, and in a fully auto-adaptive way. Such results have a very strong impact on the overall engine development process, and on the performance the engine can guarantee during its entire life:
* Combustion efficiency optimization is achieved by controlling the combustion phasing, considering the well-known bell-shaped Brake Specific Fuel Consumption (BSFC) vs. CA50MFB curves . Closed-loop control can therefore be implemented by setting as target the optimal CA50MFB value, previously identified as a function of speed and load. Then, information about combustion phasing (i.e., CA50MFB) is extracted from the in-cylinder pressure signal, by evaluating the so-called normalized heat release curve .
* During combustion phasing optimization activity, the controller varies the SA angle to reach the target value, and its authority should be limited to avoid excessive knocking levels. In-cylinder pressure signal should then be real-time processed to provide both combustion phase and knocking intensity information. The pressure-based index used in this work is called MAPO . It represents the maximum of the absolute value of the high-pass filtered in-cylinder pressure signal (a 5 kHz high-pass filter has been used). Knocking is a stochastic event , and very high percentile MAPO values, around 98-99%, are normally used to determine the knocking intensity of a sequence of combustion events, each characterized by its MAPO value, once they have been collected in a First In First Out buffer.
* With the proposed control system, the optimization mentioned above could be performed cylinder by cylinder: even the ideal open-loop controller, "perfectly" calibrated, would inevitably achieve an overall higher fuel consumption.
* One of the most interesting outcomes of the proposed solution is the great reduction of calibration costs and time. The SA open-loop controller could be fully eliminated, together with the associated calibration time and costs.
* Finally, the closed-loop controller is inherently auto-adaptive, both to engine-to-engine variations and to combustion variations during engine life (fuel quality, ambient conditions, ageing effects).
In-Cylinder Pressure Controller Layout
The controller structure was developed by considering different sub-functions: targets definition, closed-loop controllers, hierarchy definition, and SA actuation. As shown in Figure 5, the controller compares the target CA50MFB and the maximum admissible MAPO 99th percentile with the measured ones, and two parallel error calculations are performed. The CA50MFB controller filters the measured values through a moving average (to avoid reacting to intrinsic combustion variability) and it is bidirectional, in the sense that it can request both negative and positive corrections to reach the target value (the minimum SA variation is 0.75[degrees]CA). The knock control strategy, instead, can only require negative SA corrections, if the knocking level is higher than the threshold. In that case, the CA50MFB error is frozen, and the knock controller error prevails. In this way, the SA actuation variations requested by the combustion phase controller are executed only if the knocking level is below the actual threshold, while the knock controller applies SA reductions if such threshold is overcome. The errors of the two strategies, with respect to their targets, are summed and the resulting error is considered to feed the PI controller.
Weights are available to allow the controller to react differently to the two errors. In addition, a fast protective action, highlighted in blue, performs a permanent correction ("DSA_PROTECTION" in the diagram) to the integral value of the PI every time a heavy knock cycle (which Mapo value overcomes "MAPO_THR") is detected.
Pressure-Based Model-in-the-Loop Results
The controller was then tested, further developed and optimized in a self-built Model-In-the-Loop environment. The engine model, and the associated combustion variability, was reproduced by randomly selecting in-cylinder pressure cycles from pre-recorded engine running conditions (described in section "Experimental Setup"), cylinder-by-cylinder, and for the specific SA value that the controller would output, as was done in previous works . In particular, for the investigated operating point, 500 cycles have been acquired for 15 consecutive values of SA, corresponding both to excessive knock intensity and sub-optimal knock-free operation.
The engine model (database) is then made by two 3-D matrices, one for the CA50MFB and one for the MAPO values. The applied SA (calculated by the controller), the random number generated every iteration, needed to randomly select a cycle, and the cylinder number define the cell of the matrices to be selected.
Considering cylinder n. 1, Figure 6 shows a steady-state test where the combustion phasing controller is activated starting from an open-loop condition, while the engine is running at 4500 rpm and 23 bar of IMEP (Indicated Mean Effective Pressure). In the higher part of the top plot, the dashed line represents the CA50MFB target value and instantaneous and averaged CA50MFB values are also shown. The moving average value, represented by the thick solid line, is the input to the controller. The lower part of the upper plot shows the SA correction applied by the controller with respect to the open-loop value. The lower plot shows instantaneous MAPO values and its 99th percentile. The dashed line is the MAPO 99th percentile threshold.
Finally, the x-axis represents elapsed engine cycles in both plots. In the first part of the test, the controller target is set to 15 degrees ATDC, corresponding to a light knock condition: the target is reached with a 3 degrees SA correction (limiting the SA oscillation to one step once it has been reached), and the MAPO 99th percentile is still under the threshold, even if it has significantly increased. The CA50MFB target is then reduced to 12 degrees ATDC at cycle n. 1000. The controller reacts by further advancing SA, thus further increasing knock intensity. Once the knock threshold has been overcome (cycle n. 1200), the SA is reduced by the action of the knock controller, and from then on, both controllers cooperate to keep the CA50MFB as close as possible to the target, while limiting the 99th MAPO percentile below the threshold.
Ion Current Controller Development
The basis on which this work is founded is that ion current system may not only replace accelerometer-based knock controllers, but it could allow a closed-loop SA control able to maximize engine efficiency also under knock-free operating conditions. In fact, both CA50MFB- and MAPO-related information can be extracted from the ion current signal.
Ion Indexes Calculation In this initial phase, ion-based indexes have a relatively simple definition, also to be compatible with real-time calculation limitations.
The knock index, called IntIon, is calculated as follows: the signal is high-pass filtered and then the mean value of its absolute value is evaluated within a predefined angular window (Equation 1).
IntIon = [[[summation].sup.Win_end.sub.Win_start]|[ION.sub.high--pass]|/samples] Eq. (1)
In particular, the cut-off frequency of the filter has been set to 15 kHz, and the signal has been windowed between 25 and 55[degrees]CA ATDC.
The angle corresponding to the thermal peak of the ion signal has been calculated by identifying the up-down zero-crossing of the first derivative of the low-pass filtered ion signal (cut-off frequency set at 2 kHz), windowed in the range 15-70[degrees]CA ATDC. Figure 7 shows on the left side the correlation between CA50MFB and ATHP (Angular THermal Peak position), which is significantly high (87%), and almost linear.
The right side of Figure 7 shows the correlation between MAPO and IntIon ion-based knocking index, for the considered engine operating condition (4500 rpm and 23 bar of IMEP). In this case, the correlation level is lower (62%), but still sufficient to correctly close the control loop.
For a better assessment of the ion signal potential, the controller was purposely kept identical to the pressure-based one, and its inputs are previously converted from ion-based to pressure-based ones. In other words, a"conversion" stage is added to the controller shown in Figure 5, to convert ATHP into CA50MFB, and IntIon into MAPO. Such conversion has been performed by inverting the fitting functions reported in Figure 7, to calculate the corresponding values of CA50MFB and MAPO as a function of ATHP and IntIon, respectively.
Ion-Based Model-in-the-Loop Results A database similar to the one used for developing the pressure-based controller has been generated from the same experimental dataset for ATHP and IntIon indexes development, and the same type of tests have been performed.
Figure 8 shows an example of how the ion-current-based virtual controller behaves during a test like the one shown in Figure 6. Also in this case, the phase target is decreased from 22 to 19 ATHP degrees (i.e., from 15 to 12 CA50MFB degrees) at cycle n. 1000, and the further SA increase applied to achieve such target induces excessive knocking levels. The intervention of the knock controller enables a condition where the SA is forced to oscillate to keep the engine as close as possible to the target combustion phase, avoiding excessive knock.
As it can be observed, the ion-based controller performance is very similar to the pressure-based one. To quantify and compare their behavior, standard deviation and average error have been considered as possible metrics. Generally, under knock-free conditions, the two controllers achieve the same particularly high accuracy in terms of CA50MFB and ATHP (mean error equal to 0.02-0.03 CA degrees), and the combustion stability once the loop is closed is almost unaffected (standard deviation of about 1[degrees]CA in both cases). Under knock-limited operation, the pressure-based system allows reaching the threshold level very accurately, while the ion-based one is slightly less robust, due to greater false positives occurrence.
The final step of this phase of the project consisted in testing the developed controllers in real time, on the real engine. A self-developed, National Instruments based, Rapid Control Prototyping (RCP) system was used to verify the controller performance in the test cell. As described in Figure 4, the RCP system receives, in real time from the combustion analysis system, pressure- or ion-current-based indexes (combustion phase and knock intensity), and it applies a SA correction to the one calculated by the ECU in open loop. The following analysis is extended to all the engine cylinders, to demonstrate the ability of the controller to reach the same CA50MFB target (or the same knocking limit) by acting individually, and differently, on the SA angles of the various cylinders. For clarity, the results shown in the article are limited to one of the two banks of the V-8 engine described above.
In-Cylinder Pressure Controller Experimental Results
As an example of the in-cylinder pressure-based controller performance under knock-free operation, Figure 9 shows an experimental test during which a SA step was externally imposed, to analyze the controller ability to reject external disturbances.
In the first part of the test, up to engine cycle n. 1300, the controller is operating under steady-state conditions, with a CA50MFB target of 20 CA degrees ATDC. The top plot shows both the target and the achieved CA50MFB for the four cylinders (evaluated as a moving average-MOVA- of the instantaneous CA50MFB, and corresponding to the controller input), while the lower plot reports the individual, and different, DSA corrections applied to the cylinders (between -3.0 and -1.5 CA degrees). As it can be seen, the 4 considered cylinders require slightly different mean SA values to reach the target combustion phase. At cycle n. 1300, an external, 1.5 CA degrees wide, SA step disturbance is imposed to all the cylinders, as shown in Figure 9. The controller reacts by increasing the negative SA correction on all engine cylinders, and once the transient is over (in about 200-250 engine cycles), the CA50MFB target is reached again for all the cylinders. As it can be seen, the controller performance is very similar to the one observed in the virtual environment, confirming the high accuracy in terms of CA50MFB. Also in this case, the combustion stability is essentially unaffected with respect to open loop, constant SA operation.
The performance of the in-cylinder pressure closed-loop SA controller under knock-limited operation may be analyzed by looking at Figure 10. For the sake of clarity only cylinder 1 is investigated. Also in this case, the figure shows both the ability of the controller to guarantee that the threshold knocking level is not overcome under steady-state conditions and its robustness in terms of disturbance rejection. The test is performed by setting a CA50MFB target equal to 12 CA degrees ATDC, which corresponds, for the given engine operating conditions, to excessive knocking intensity. Then a positive SA disturbance step is externally applied, to abruptly increase the knocking level and to verify the controller ability to reduce the SA to continue respecting the threshold knocking level.
The top plot shows the instantaneous MAPO values (thin line), the corresponding MAPO 99th percentile (thick line), and the MAPO 99th percentile threshold (thick dashed line). The lower part of the top plot presents both the externally applied SA advance step (thick dashed line), and the SA correction imposed by the closed-loop controller. Finally, the bottom plot shows the CA50MFB moving average value, or MOVA, for cylinder n.1, which is an input to the controller, and the CA50MFB target (equal to 12 CA degrees). In the first part of the test, up to cycle n. 1300, it can clearly be observed how the controller continuously tries to reach the target CA50MFB by reducing the SA correction, but then the SA is forced back to smaller values (i.e., the SA correction assumes again greater negative values) since the MAPO percentile overcomes the threshold. Also, when the SA disturbance is applied, the controller reacts by requesting greater negative SA corrections, to limit the knocking intensity by compensating for the disturbance. At the end of the test the controller correction is in fact equal to around -3 CA degrees, 1.5 CA degrees smaller than at the beginning of the test.
Ion Current Controller Experimental Results
Finally, several experimental tests were conducted to evaluate the controller performance based on ion current rather than in-cylinder pressure measurements. Figure 11 reports an exemplary behavior of the controller operating under knocklimited operation, focusing the attention on cylinder n. 2.The CA50MFB target has been set at 12 CA degrees ATDC, a combustion phase that corresponds to excessive knock intensity. The top plot shows, in the upper part, the controller internal variables, such as the ion-current-based knock index (IntIon), the 99th percentile value of the same index (IntIonPerc), and the threshold corresponding to maximum knocking intensity (IntIonThr).
The lower part of the upper plot shows SA variations, both internally calculated by the controller (DSA) and externally imposed (SA disturbance step). The lower plot shows the controller performance, both in terms of CA50MFB moving average (CA50MOVA) and MAPO (MAPO, MAPOPerc), with respect to the corresponding target (CA50 Target) and threshold (MapoPercThr).
It can be clearly noticed how also in this case the maximum knock intensity is not frequently overcome, and at the same time the CA50MFB target is constantly tracked by reducing the SA correction. When the disturbance step is imposed (around cycle n. 1300), the knocking intensity increases, and the controller reacts by further reducing the SA, thus restoring a borderline knock condition, as desired.
Comparison between Pressure- and Ion-Based Control
To numerically compare the two controls, mean values and standard deviations of the controlled variables have been considered, to evaluate accuracy and precision, respectively.
In Table 1, the comparison between the pressure-based and the ion-current-based control performance is reported. In the upper part of the table, knock-free condition is considered, with two different target values of CA50MFB. The open loop condition is realized with a constant spark advance angle that realizes the CA50MFB closest to the target, identified for the specific engine and operating condition before the test. This condition is needed to set an "ideal" reference value for the standard deviation of CA50MFB.
The closed-loop control, both implemented with pressure and with ionization current signals, maintains the target with good accuracy and with a small increase on CA50MFB variability (i.e., standard deviation).
The error between the target of 20[degrees]CA MFB50 and the mean value realized with the ionization-current-based control (19.41[degrees]CA) is to be attributable to the identified MFB50 regression model (see Figure 7, left plot), which is assumed to be linear, while a higher polynomial degree would produce smaller regression errors. In knock-limited operation, the performance of the controller is evaluated in terms of mean value (and secondly standard deviation) of the MAPO 99th percentile, which is the targeted variable.
Because of the stochastic nature of the knock phenomenon, it is impossible to control a quasi-static value of the knock intensity (i.e., MAPO 99th percentile).
Moreover, the controller architecture needs the threshold to be crossed in both directions, and since the positive distance (with respect to the threshold) is generally higher than the negative one, the mean knock intensity is slightly higher than the targeted value. That is the reason why the pressure-based mean-controlled knock intensity is higher than the target (5.2 bar instead of 4 bar, 8.75 bar instead of 8 bar). The same applies to the ionization-current-based controller if the variable MapoPercEQ is considered. As explained before, the two control architectures are the same, but in the ion-based version an ion-to-pressure indexes conversion has been introduced upstream of the controller. Therefore, the ion-based control is running on the equivalent pressure indexes.
When controlling knock intensity with ionization current signal, the equivalent pressure knock index values (MapoPercEQ) and the measured pressure knock index values (MapoPerc) should match on average, unless a regression error is committed, or the regression model is not sufficiently robust. The lower standard deviation obtained for the knock intensity is related to the lower mean value of the index itself.
Conclusions and Future Work
The aim of this study is to confirm the robustness and the reliability of the ion signal to describe combustion features, both concerning combustion phasing and knock intensity, and to demonstrate how such information can be used to perform a fully closed-loop, real-time Spark Advance controller, both in knock-free and knock-limited operation, which is the element of novelty of this article.
Pressure- and ion-current-based combustion closed-loop controls are compared in the article, by implementing both solutions in real-time. Both controllers are characterized by an aggressive strategy, which pursues the optimal combustion angular phase, and by a protective action governed by measured knocking levels.
The study demonstrates the feasibility of combustion phase ion-based real-time closed-loop control, achieving very similar performance to the pressure-based control. The concept has been applied to a single operating condition, but it is extendible to the whole engine operating domain, thus allowing significant calibration costs and time reduction, and permanent fuel consumption optimization.
The proposed control system allows to close the loop on SA actuation (and therefore combustion phasing and efficiency) for any engine operating condition. The controller is intrinsically adaptive, and the benefits are several and significant: cylinder-to-cylinder disuniformity may in fact be compensated, as well as aging effects, part-to-part variation effects, environmental conditions, and, last but not least, fuel anti-knocking properties variation. Further, SA calibration efforts, and costs, may be significantly reduced.
The proposed approach is being implemented in the realtime production controller on-board the vehicle. A real-time, physics-based model of knock-induced damage is currently being validated, and future work will consist of coupling it to the proposed controller, to further optimize combustion efficiency by targeting the maximum knock level if needed.
Ing. Nahuel Rojo, Phd
University of Bologna Viale Risorgimento 2, Bologna, Italy
ATDC - After Top Dead Center
AThP - Angle of Thermal Peak position of the ionization current signal
BSFC - Brake Specific Fuel Consumption
CA50MOVA - CA50MFB moving average
CA50MFB - Crank Angle at which the 50% of Mass Fraction is Burnt
CAN - Controller Area Network
DSA - Delta (variation of) Spark Advance
GDI - Gasoline Direct Injection
IGBT - Insulated Gate Bipolar Transistor
IMEP - Indicated Mean Effective Pressure
IntIon - Customized Integral knock index calculated on the Ionization current signal
IntIonPerc - IntIon 99th percentile achieved by the controller
MAPO - Maximum Pressure of Pressure Oscillation
MAPOPerc - MAPO 99th percentile achieved by the controller
MapoPercEQ - Equivalent MAPOPerc
MiL - Model-in-the-Loop
MOVA - Moving average
PPP - Pressure Peak Position
RCP - Rapid Control Prototyping
SA - Spark Advance
SNR - Signal-to-Noise Ratio
TDC - Top Dead Center
[1.] Storm, X., Salminen, H., Virrankoski, R., and Niemi, S., "Analysis of Cylinder Pressure Measurement Accuracy for Internal Combustion Engine Control," SAE Technical Paper 2017-01-1067, 2017, doi:10.4271/2017-01-1067.
[2.] Eriksson, L., "Methods for Ionization Current Interpretation to Be Used in Ignition Control, "Master Degree Project in Vehicle Systems, Technical University of Linkoping, Sweden.
[3.] Heywood, J.B., Internal Combustion Engine Fundamentals (New York: McGraw Hill, 1988).
[4.] Cavina, N., Moro, D., Poggio, L., and Zecchetti, D., "Individual Cylinder Combustion Control Based on RealTime Processing of Ion Current Signals," SAE 2007 Transactions - Journal of Engines 116(Section 3), 2007, doi:10.4271/2007-01-1510.
[5.] Businaro, A., Cavina, N., Corti, E. et al., "Accelerometer Based Methodology for Combustion Parameters Estimation," Energy Procedia 81(2015):950-959, 2015, doi:10.1016/j.egypro.2015.12.152.
[6.] Malaczynski, G., Roth, G., and Johnson, D., "Ion-Sense-Based Real-Time Combustion Sensing for Closed Loop Engine Control," SAE Int. J. Engines 6(1):2013, 2013, doi:10.4271/2013-01-0354.
[7.] Cao, Y. and Li, L., "A Novel Closed Loop Control based on Ionization Current in Combustion Cycle at Cold Start in a GDI Engine," SAE Technical Paper 2012-01-1339, 2012, 2012, doi:10.4271/2012-01-1339.
[8.] Glavmo, M., Spadafora, P., and Bosch, R., "Closed Loop Start of Combustion Control Utilizing Ionization Sensing in a Diesel Engine," SAE Technical Paper1999-01-0549, 1999, doi:10.4271/1999-01-0549.
[9.] Henein, N., Bryzik, W., Abdel-Rehim, A., and Gupta, A., "Characteristics of Ion Current Signals in Compression Ignition and Spark Ignition Engines," SAE Int. J. Engines 3(1):260-281, 2010, doi:10.4271/2010-01-0567.
[10.] Andersson, I., "A Comparison of Combustion Temperature Models for Ionization Current Modeling in an SI Engine," SAE Technical Paper 2004-01-1465, 2004, doi:10.4271/2004-01-1465.
[11.] Haskara, I., Zhu, G., and Winkelman, J., "IC Engine Retard Ignition Timing Limit Detection and Control using InCylinder Ionization Signal," SAE Technical Paper 2004-01-2977, 2004, doi:10.4271/2004-01-2977.
[12.] Zhu, G., Daniels, C., and Winkelman, J., "MBT Timing Detection and Its Closed-Loop Control Using In-Cylinder Ionization Signal," SAE Technical Paper 2004-01-2976, 2004, doi:10.4271/2004-01-2976.
[13.] Eriksson, L., Nielsen, L., and Glavenius, M., "Closed Loop Ignition Control by Ionization Current Interpretation," SAE 1997 Transactions, Journal of Engines 106(Section 3):1216-1223, 1997, doi:10.4271/970854.
[14.] Giglio, V., Police, G., Rispoli, N. et al., "Experimental Investigation on the Use of Ion Current on SI Engines for Knock Detection," SAE Technical Paper 2009-01-2745, 2009, doi:10.4271/2009-01-2745.
[15.] Kumar, D., Ramesh, A., Babu, M., and Manivannan, P., "An Ionization Current Based Cylinder Gas Pressure Estimation for Knock Detection and Control in a Single Cylinder SI Engine," SAE Technical Paper 2009-32-0118, 2009, doi:10.4271/2009-32-0118.
[16.] Zhu, G., Haskara, I., and Winkelman, J., "Stochastic Limit Control and Its Application to Knock Limit Control Using Ionization Feedback," SAE Technical Paper 2005-01-0018, 2005, doi:10.4271/2005-01-0018.
[17.] Panousakis, D., Gazis, A., Paterson, J., Chen, W. et al., "Ion Current Signal Interpretation via Artificial Neural Networks for Gasoline HCCI Control," SAE Technical Paper 2006-01-1088, 2006, doi:10.4271/2006-01-1088.
[18.] Cavina, N., Poggio, L., and Sartoni, G., "Misfire and Partial Burn Detection Based on Ion Current Measurement," SAE Int. J. Engines 4(2):2451-2460, 2011, doi:10.4271/2011-24-0142.
[19.] Cavina, N., Rojo, N., Poggio, L., Calogero, L. et al., "Investigation on Pre-Ignition Combustion Events and Development of Diagnostic Solutions Based on Ion Current Signals," SAE Int. J. Engines 10(4):2017, 2017, doi:10.4271/2017-01-0784.
[20.] Upadhyay, D. and Rizzoni, G., "AFR Control on a Single Cylinder Engine Using the Ionization Current," SAE Technical Paper980203, 1998, doi:10.4271/980203.
[21.] Corti, E. and Forte, C., "Statistical Analysis of Indicating Parameters for Knock Detection Purposes," SAE Technical Paper 2009-01-0237, 2009, doi:10.4271/2009-01-0237.
[22.] Spelina, J., Peyton Jones, J., and Frey, J., "Recent Advances in Knock Analysis, Simulation, and Control," SAE International Journal of Engines 7(2):947-955, 2014, doi:10.4271/2014-01-1349.
[23.] Cavina, N., Po, G., and Poggio, L., "Ion Current Based Spark Advance Management for Maximum Torque Production and Knock Control," in 8th Biennial ASME Conference on Engineering Systems Design and Analysis, Torino, Italy, July 4-7, 2006, doi:10.1115/ESDA2006-95558.
Nicolo Cavina and Nahuel Rojo, University of Bologna, Italy
Andrea Businaro, Alma Automotive srl, Italy
Ruggero Cevolani, Ferrari Auto spa, Italy
Received: 07 Nov 2018
Revised: 31 Jan 2019
Accepted: 04 Mar 2019
e-Available: 08 Apr 2019
Cavina, N., Rojo, N., Businaro, A., and Cevolani, R., "Comparison between Pressure- and Ion-Current-Based Closed-Loop Combustion Control Performance," SAE Int. J. Engines 12(2):219-230, 2019, doi:10.4271/03-12-02-0016.
TABLE 1 Engine specifications. Stroke 82 mm Bore 86.5 mm Compression ratio 9.4:1 Displacement 3855 cc, 8 cylinders TABLE 2 Comparison between pressure-based and ion-current-based control Control CA50MFB CA50MFB mean std Knock-free Open - 20.42 1.41 loop Pressure MFB50 20 1.48 ION tgt: 20 19.42 1.48 (knock tgt 8) Open - 15.56 1.22 loop Pressure MFB50 15.02 1.39 ION tgt: 15 15.03 1.39 (knock tgt: 8) MapoPerc MapoPerc MapoPercEQ mean std mean Knock- Pressure Knock 5.21 1.64 - limited ION tgt: 4 3.93 1.52 5.72 (MFB50 tgt: 12) Pressure Knock 8.75 3.04 - ION tgt: 8 6.83 2.6 9.48 (MFB50 tgt: 12)
|Printer friendly Cite/link Email Feedback|
|Author:||Cavina, Nicolo; Rojo, Nahuel; Businaro, Andrea; Cevolani, Ruggero|
|Publication:||SAE International Journal of Engines|
|Article Type:||Technical report|
|Date:||Apr 1, 2019|
|Previous Article:||A Comparison of EGR Correction Factor Models Based on SI Engine Data.|
|Next Article:||Limitations of Two-Stage Turbocharging at High Flight Altitudes.|