Real-Time Combustion Parameters Estimation for HCCI-Diesel Engine Based on Knock Sensor Measurement J. Chauvin, O. Grondin, E. Nguyen, and F. Guillemin Institut Fran¸cais du P´ etrole - 1 et 4, avenue de Bois-Pr´ eau 92852 Rueil-Malmaison - France Abstract: Future internal combustion engine technologies require an accurate combustion monitoring and control. This can be performed through high frequency recordings of cylinder pressure. However, this solution is limited by the sensor cost and reliability. Another method consist in reconstructing combustion related variables from indirect measurements. In this paper, we propose a combustion indicator estimation method from the vibration trace of the engine block recorded with a standard knock sensor. The relevance of such a method is demonstrated through experimental results on an HCCI engine application. Keywords: Diesel engine, knock sensor, combustion analysis, cylinder pressure, combustion diagnosis and control. 1. INTRODUCTION Increasing legal requirements in engines regarding exhaust gas emissions, fuel economy, on-board diagnosis, together with a steady demand on performance, pushes the industry towards innovative solutions with reduced time-to-market. In addition with the introduction of new combustion con- cepts such as HCCI (Homogeneous Charge Compression Ignition) or CAI (Controlled Auto Ignition), the engine combustion, control, and diagnosis are becoming increas- ingly complex. In order to perform closed loop combustion control, infor- mation about the combustion process can be investigated with in-cylinder pressure transducer. The cylinder pressure is a relevant feedback variable and internal combustion engine management systems relying on cylinder pressure signal became of a particular interest (see Leonhardt et al. [1999] for example). The combustion parameter for closed loop control are computed from the heat release analysis (Krieger and Borman [1966], Gatowski et al. [1984]). This solution has been investigated in many papers but is still not cost effective and commercial car implementations are thus limited. Other techniques relying on indirect mea- surement such as knock sensors, accelerometers, engine speed, torque sensor or ionization current measurements are possible to estimate combustion indicator for engine control. In this paper, we focus on how to extract some information on combustion using knock sensors. The use of knock sensors is motivated by the cost and the number of sensor used to characterize the combustion in each cylinder (for example two sensors may be used for a 4 cylinder engine). The aim is to extract parameters which can be linked to ⋆ J. Chauvin, O. Grondin, E. Nguyen, and F. Guillemin are with the engine control team at IFP. Corresponding author: Jonathan Chauvin. jonathan.chauvin@ifp.fr the combustion phasing and to the combustion energy. These parameters will allow us to control the combustion thanks to the injection phasing, the mass of fuel injected and the EGR (Exhaust Gas Recirculation). The accelerometer has multiple applications for IC en- gine. The first application was the detection of knock and the evaluation of its energy for SI engine, (see Boubal and Oksman [1998], Cerda et al. [2002], Zhekova and Guillemain [2004], ?]). Other studies focused on the re- construction of the cylinder pressure signal. The signal processing tools used for that purpose were : deconvolu- tion methods in Wagner et al. [1999], spectrum analysis in Gao and Randall [1999], cyclostationarity properties in Antoni et al. [2002] and methods using neural net- work models in Johnsson [2006]. Other studies tried to identify the different source signatures on vibration signal, see El Badaoui et al. [2005]. However, all of these methods cannot be implemented in real time and thus closed loop combustion control cannot be performed. Our aim is to estimate combustion indicators cylinder to cylinder on a cycle to cycle basis. For that purpose, on-line algorithms are required. Jargenstedt [2000] uses the analytic signal (Hilbert transform) to extract an envelope of the combus- tion. This method needs an accurate band pass filtering to extract the combustion signature and is very sensitive to contributions from other sources (due to low signal to noise ratio). The time frequency can be implemented by a recursive algorithm with the sliding discrete Fourier transform Ker et al. [2007]. It needs to choose an accurate time window length and the frequency band to look at. This frequency band may be too large and let the contri- bution of non combustion sources pollute the signal. This frequency band depends on the time window length. It is possible to model combustion from the vibration signal with an autoregressive moving average filter (ARMA fil- ter). Souder et al. [2004] uses ARMA filter to estimate the start of combustion. However, the proposed model Proceedings of the 17th World Congress The International Federation of Automatic Control Seoul, Korea, July 6-11, 2008 978-1-1234-7890-2/08/$20.00 © 2008 IFAC 8501 10.3182/20080706-5-KR-1001.1789