Estimation of Ethanol Content in Flex-fuel Vehicles Using an Exhaust Gas Oxygen Sensor: Model, Tuning and Sensitivity

Throughout the history of the automobile there have been periods of intense interest in using ethanol as an alternative fuel to petroleum-based gasoline and diesel derivatives. Currently available flexible fuel vehicles (FFVs) can operate on a blend of gasoline and ethanol in any concentration of up to 85% ethanol. In all these FFVs, the engine management system relies on the estimation of the ethanol content in the fuel blend, which typically depends on the estimated changes in stoichiometry through an Exhaust Gas Oxygen (EGO) sensor. Since the output of the EGO sensor is used for the air-to-fuel ratio (AFR) regulation and the ethanol content estimation, several tuning and sensitivity problems arise. In this paper, we develop a simple phenomenological model of the AFR control process and a simple ethanol estimation law which can be representative of the currently practiced system in FFVs. Tuning difficulties and interactions of the two learning loops are then elucidated using classical control techniques. The sensitivity of the ethanol content estimation with respect to sensor and modeling errors is also demonstrated via simulations. The results point to an urgent need for model-based analysis and design of the AFR controller, the ethanol adaptation law and the fault detection issues in FFVs. Tuning and sensitivity issues are demonstrated via simulations and limitations are also discussed. NOMENCLATURE AFRs Stoichiometric air-to-fuel ratio ÂFRs Estimated stoichiometric air-to-fuel ratio e Volume fraction of ethanol in gasoline-ethanol blend ê Estimated volume fraction of ethanol in gasoline-ethanol blend ∗Address all correspondence to [email protected]. em Mass fraction of ethanol in gasoline-ethanol blend MAF Mass air flow measured by a MAF sensor Tm Manifold temperature measured by a manifold temperature sensor Wcyl Air flow rate into the cylinder Ŵcyl Estimated air flow rate into the cylinder Wf Fuel flow rate into the cylinder Wf b Feedback fuel command Wf f Feedforward fuel command Wf f 1 Feedforward fuel command not compensated by the fuel puddle dynamics X Aquino parameter denoting the wall-impacting portion of the injected fuel αtr Triggering signal for switching on ethanol adaptation εAFRs Estimation error in stoichiometric air-to-fuel ratio θ Throttle angle λ Ratio of actual AFR to stoichiometric AFR, λ is measured by an EGO sensor ρeth Density of ethanol ρgsl Density of gasoline τ Aquino parameter denoting the vaporization time constant of the fuel puddle τd Transport and induction-to-power delay from the cylinder input AFR to the exhaust AFR τs Time constant of the exhaust oxygen sensor lag INTRODUCTION Petroleum-based fossil fuels are the dominant energy source for transportation. Recently, however, ethanol is being increasingly used as a fuel additive and is emerging as an alternative to carbon-neutral transportation. The advantage of ethanol, among others, is that it is a renewable fuel produced from biomass such as barley, corn, wheat, sugar cane, trees and grasses. Therefore, to lessen dependence on fossil-based fuels, federal mandates, such as the Energy Policy Act of 2005 (EPACT 2005), require that 7.5 billion gallons of bio-fuel be produced in 2012. Ethanol can be blended with conventional gasolines in varying percentages. The blend is denoted by the EXX nomenclature, where XX represents the volumetric percentage of ethanol in the blend. The United States commonly uses E85 as an alternative to the normal E0 or gasoline fuel. In Brazil, however, the fuel blend also contains water and E100 refers to a blend of 93% ethanol and 7% of water [1]. Such fuel blends mixed with additional water are not considered in this paper. Vehicles that can operate on any blend of ethanol are called Flex Fuel Vehicles. These vehicles are designed to run on gasoline or a blend up to E85 in the U.S. and are currently being offered by many manufacturers. The characteristics of ethanol differ from those of gasoTable 1. PROPERTIES OF ETHANOL COMPARED WITH GASOLINE. Property Gasoline Ethanol Research Octane Number (RON) 92 111 Density (kg/m3) 747 789 Heat of combustion (MJ/kg) 42.4 26.8 Stoichiometric air-to-fuel ratio 14.6 9.0 Boiling point (◦C) – 78.5 Latent heat of evaporation (kJ/kg) 420 845 line, as shown in Table. 1. Various effects of ethanol fuel on a spark ignition engine are well reported in [2]. Often ethanol fuel is associated with driveability and startability problems in cold and hot weather [3, 4] and at high altitude [5]. Existing FFVs achieve lower range (miles driven per tank) when operating on high ethanol content fuel due to its lower combustion heating value as compared to gasoline. However, as shown in Table. 1, ethanol has a higher octane ratio and therefore, a higher compression ratio and higher combustion efficiency can be obtained without knocking problems. Another advantage is that, the high vaporization heat can be used for charge cooling [6], thus improving further the knock resistance and potentially fuel economy. Given the effect of fuel variation, FFVs should embed engine calibration maps in their controllers and management systems to account for this variation. To accomplish this, the first task in a flex-fuel strategy is to estimate reliably the ethanol percentage. Although, this estimation is possible with the addition of a di-electric or electrochemical sensor in the fueling system, the reliability of these sensors has not yet been proven. Further, apart from the cost and reliability issues associated with such sensors, on-board diagnostic (OBD) requirements would require a redundant method for assessing the ethanol percent in order to diagnose the ethanol content sensor faults or degradation. Currently the ethanol content estimation depends on the AFR measurement through an exhaust gas oxygen (EGO) sensor immediately after refueling is detected. This trigger is used to avoid misclassifying ethanol content variations with actuator drifts or component aging. The ethanol detection period also needs to be as short as possible to reduce the probability of another (EGO or mass air flow (MAF) sensor, and/or injector) fault. Finally it is necessary to thoroughly understand and develop models for how uncertainty, sensor and actuator drifts propagate through the detection process and affect the ethanol content estimation. Basic discussion regarding the robustness of the estimation using the exhaust oxygen sensor feedback is provided in [7]. In this paper, a simple stoichiometric AFR estimation law using the exhaust oxygen sensor is proposed, analyzed and discussed in light of AFR control, which yields the estimated ethanol percentage in the fuel. Characteristics of the algorithm including sensitivity are quantified via simulations. ETHANOL CONTENT ESTIMATION Air-to-fuel ratio control around the stoichiometric ratio of a fuel blend is important to meet stringent emission requirements for spark ignition (SI) engines. For a given air charge, the stoichiometric fuel is typically achieved by a combination of feedforward and feedback control on the fuel injection. The feedback controller is based on the measured ratio (λ) of the actual air-tofuel ratio (AFR) to the stoichiometric ratio (AFRs) through an exhaust gas oxygen sensor. The λ ratio is compared to λdes = 1 and the error is used by a PI to adjust the feedback fuel command. Due to the long delays in the feedback loop, most engine controllers employ a feedforward fuel command which is primarily derived from the estimated cylinder air charge divided by the assumed stoichiometric ratio of the assumed fuel blend. Furthermore, the feedforward is usually designed to eliminate the transient effects of fuel puddle dynamics in port fuel injected (PFI) engines. Since the puddle dynamics is dependent on the ethanol content, estimation of ethanol content may also be used for the transient fuel compensation (TFC). A model of fuel puddle dynamics with alternative fuels is discussed in [8] where a fuel blend is modeled as a certain combination of organic compounds that mimic the distillation behavior of an actual fuel. When the assumed stoichiometric ratio is correct, and there are no errors in the air charge and fuel puddle dynamics estimation, and no drifts or faults in the injector, the feedforward fuel command is then perfect and the feedback fuel compensation should be zero. An estimation algorithm can utilize a nonzero feedback fuel command to adapt and improve the feedforward fuel compensator so that the feedback converges back to its nominal zero value. Ideally, the adaptation will address the core problem in the feedforward path. In the case of an FFV, the adaptation of the assumed stoichiometric ratio or the adaptation for a miscalibrated Manifold Breathing Dynamics