Method for operating a gaseous fuel engine

Abstract

Method for operating a gaseous fuel internal combustion engine 10 comprising calculating a global correction for adjusting a fuel command 20. The global correction is learned periodically by implement

Description

Technical field The present invention generally relates to methods for operating gaseous fuel engines. More specifically to a method for determining a fuel quantity to inject using a lambda sensor. Background Art The present invention relates to methods and systems for determining the lambda (A) value, or air-fuel equivalence ratio, specifically for gaseous fuel engines. In internal combustion engines, accurate control of the air-fuel ratio is crucial for optimizing combustion efficiency, reducing emissions, and maintaining engine durability. Lambda (A) represents the ratio of the actual air-to-fuel ratio to the stoichiometric air-fuel ratio, where A = 1 indicates an ideal balance of air and fuel, ensuring complete combustion. For liquid-fuel engines, typical lambda values range from 0.9 (slightly rich) to 1.1 (slightly lean), depending on performance demands and emission controls. In conventional liquid-fuel engines (gasoline, diesel...), the lambda value may be determined using an oxygen sensor located in the exhaust, which measures the concentration of residual oxygen in the exhaust gases. This oxygen concentration provides an indirect measurement of the combustion efficiency and, consequently, the air-fuel ratio. The lambda value, derived from this oxygen reading, then informs the engine control unit (ECU) to adjust the amount of fuel to inject, thereby forming a closed-loop control strategy and ensuring that the combustion process remains within optimal parameters. However, in gaseous fuel engines, such as those powered by natural gas, hydrocarbons, hydrogen, or a combination thereof, conventional lambda determination methods face inherent limitations. Gaseous fuels exhibit distinct combustion characteristics that may differ significantly from those of liquid fuels. Traditional oxygen sensors, designed for liquid-fuel engine operations, often fail to provide reliable lambda readings in nominal gaseous fuel environments, leading to inaccurate air-fuel ratio assessments. This results in suboptimal engine performance, increased emissions, and potential engine degradation overtime. Technical problem It is an object of the present invention to provide a method for operating a gaseous fuel engine which overcomes the aforementioned drawbacks, and overcomes the limitations of conventional oxygen sensors and methodologies. General Description of the Invention This object is achieved by a method for operating a gaseous fuel internal combustion engine as claimed in claim 1. A method for operating a gaseous fuel internal combustion engine is provided, wherein the engine comprising a plurality of fuel injectors each configured to deliver gaseous fuel in a corresponding cylinder of the engine, and a lambda sensor positioned to respond to exhaust gases generated by the cylinders. The method implements at least one fuel injection strategy wherein: - the fuel injectors are operated to discharge into engine cylinders fuel quantities corresponding to a predetermined fuel command (Qc); - the fuel command (Qc) is computed from a base fuel amount (Qb), which is determined from a desired lambda value, and adjusted by a stored global correction (Cg); According to the invention, the global correction (Cg) is learned periodically by implementing a global correction learning routine comprising the steps of: - defining a test demand fuel quantity based on a test lambda value comprised between 1.0 and 1.6; - performing injection events by operating the plurality of injectors to deliver the test demand fuel quantity to the cylinders; - determining a corresponding, actual lambda value from the lambda sensor; - determining a new global correction based on the actual lambda value and the test lambda value; - updating the stored global correction by storing the new global correction in a memory of a control unit. The inventors have found that during lean operation of the engine, the accuracy of modern oxygen sensor is insufficient to ensure acceptable engine performance. To solve this issue, the inventive method implements an intrusive global correction learning routine, whereby injections are performed based on lambda value which is considerably lower than the lambda value during normal operation of the engine. The global correction determined during this learning routine is then stored and used during normal operation of the engine to correct the fuel demand, resulting in improved engine performance. As indicated, the global correction learning routine is implemented periodically, for the purpose of learning. The global correction is thus learned at predetermined time points, and the updated / stored value is used until the value is updated by a subsequent implementation of the global correction learning routine. The term periodically (or periodic) distinguishes from conventional closed-loop learning strategies that are performed in parallel or asynchronously to the normal engine operation. In embodiments, the desired lambda value is greater than 2.0, and is preferably comprised between 2.3 and 2.6. In embodiments, the global correction is a global correcting factor determined from a ratio between the actual lambda value and the test lambda value, and the fuel command is determined from a product between the base fuel amount, and the global correcting factor. In embodiments, the steps of defining a test demand fuel quantity based on a test lambda value, performing injection cycles by operating the plurality of injectors to deliver the test demand fuel quantity to the cylinders, determining a corresponding, actual lambda value from the lambda sensor, and determining a new global correction based on the actual lambda value and the test lambda value are repeated during a single global correction learning routine, and the test demand fuel quantity is adjusted by a previous global correction based on a new global correction from a previous iteration, such that the global correction learning routine is performed in closed-loop. Hence, whereas the global correction learning routine is implemented periodically, and hence the global correction is updated periodically, when the learning routine is implemented the learning itself may be advantageously done in closed loop. In embodiments, the step of updating the stored global correction by storing the new global correction is performed only if the new global correction is stable. In embodiments, the new global correction is considered stable if the difference between consecutive determined global corrections is lower than a predetermined stability threshold for a predetermined stability duration. The stability threshold may be expressed as an absolute value or percentage of the difference with respect to a previous value. The duration may be represented by a percentage or an absolute value. The predetermined stability duration may be in the order of a few seconds, e.g. 3 to 7 seconds. In embodiments, the global correction learning routine has a maximum duration, and the new global correction is discarded if no stable new global correction was determined within this maximum duration. The trigger for the periodic learning may be accumulated kilometers, engine operating hours, or engine cycles. In embodiments, the global correction learning routine is performed periodically i.e. with a predetermined time interval between two consecutive implementations of the learning routine. For example, the time interval between two consecutive implementations of the learning routine can be greater than 20, 50, 80 100, 150 or 200 hours of engine operation. In embodiments, the global correction learning routine is begun only when the learning conditions are met. The learning conditions may be based on one or more inputs or conditions selected from: sensor status, runtime, engine load, engine temperature, engine speed (RPM), engine idling. Since the learning procedure is intrusive, it advantageous to apply it when there is no particular demand from the driver / operator or the system. Accordingly, one learning condition may be that the engine is in idle condition. In embodiments, the global correction learning routine is aborted if an input from a user is detected. It may be noted that it may be desirable to perform a first implementation of the global correction learning routine in the context of the "initial start-up" or "first fire” of the engine. In this context, the present routine may be configured to allow the operator to select whether to correct the sensor accuracy or whether to directly correct the fueling. Alternatively, this selection may be preset in the system. In embodiments, where the global correction learning routine is run as first implementation, the new global correction may be stored as a lambda sensor correction, and the step of determining the oxygen concentration in the exhaust gas is performed by measuring the oxygen concentration in the exhaust gas and applying the lambda sensor correction to the measured oxygen concentration. This first correction is thus used to correct the lambda sensor instead of the fuel quantity commanded by the fuel injectors. This is useful when the accuracy of the injectors has been verified shortly prior to the first global correction learning routine. Alternatively, the value determined at first implementation of the global correction learning routine can be used to directly update the global correction Cg. In this case it may be desirable to store new global correction independently from the global correction. The evolution of the global correction relative to the first global correction can then be used for monitoring and diagnostic of the injection system over its lifetime. The lambda sensor may generally be a sensor that is responsive to the oxygen concentration in the exhaust gas. Any appropriate sensor technology may be used. The lambda sensor typically generates a sensor signal that varies with the oxygen concentration. So-called wide-range oxygen sensors can be used. In embodiments, the oxygen concentration in the exhaust gas may be measured by determining presence of oxygen by means of a so-called switching sensor. In such case the routine is operated at Atest =1.0. In embodiments, the fuel command is computed from the base fuel amount, the stored global correction, and an Individual Cylinder Fuel Control correcting factor determined from a signal of the lambda sensor. These and other features of the inventions are recited in the appended dependent claims. According to another aspect, the invention relates to a control unit comprising a processor configured to perform the method according to the present disclosure. According to still another aspect, the invention relates to a gaseous fuel multi-cylinder internal combustion engine comprising fuel injectors arranged to fuel cylinders and a lambda sensor positioned to respond to exhaust gases generated by the cylinders, and a control unit configured to perform the method according to the present disclosure. Brief Description of the Drawings A preferred embodiment of the invention will now be described, by way of example, with reference to the accompanying drawings in which: Fig. 1 is a functional diagram implementing an embodiment of the inventive method; Fig. 2 is a plot of oxygen concentration (%) vs. lambda number; Fig. 3 is plot of the tolerance (%) in the lambda measure vs. lambda number; Fig. 4 is a flowchart representing the learning operation of the inventive method; Fig. 5 is a flowchart representing possible outcomes for the first learning operation. Description of Preferred Embodiments An implementation of the present method will now be explained with reference to Figs.1 and 4. The functions and equipment shown in Fig.1 are generally known and will only be briefly explained. The present invention relates to fuel control strategies in internal combustion engines operating with gaseous fuel. The fuel delivery system typically comprises a fuel tank, e.g. containing pressurized gaseous hydrogen, a fuel rail and a plurality of fuel injectors for selective fuel injection into the engine. The fuel injectors may be coupled directly to the fuel rail via so-called sockets, or indirectly via tubes. Conventionally, one fuel injector is provided per cylinder, either in direct injection (DI) configuration (fuel is introduced directly into the cylinder / combustion chamber) or in PFI configuration (the injector is arranged to discharge fuel upstream of the intake valve(s)). As is known in the art, the engine comprises an engine block with a plurality of cylinders with associated reciprocating pistons mechanically coupled to a crankshaft. At least one fuel injector is provided per cylinder -according to the mentioned configuration PFI or DI- to inject fuel to be combusted in the respective combustion chamber and generate torque. Each cylinder comprises at least one intake valve for admitting fresh air and at least one exhaust valve for discharging combustion gases. Introduction of fuel in a given cylinder, i.e. cylinder fueling, is performed during an injection event, by applying a drive signal to the fuel injector to activate an electromechanical actuator, e.g. a solenoid actuator, to cause the injector to open during a predetermined time period. Much simplified, injection control strategies use mappings (known as calibrated flow curves) that relate the fuel quantity to the injector actuation time that is referred to as pulse width, PW. To perform an injection event, a drive pulse is applied during a time period PW to discharge a corresponding fuel amount. Conventionally, injection control strategies are programmed in the Engine Control Unit, ECU, that receives various signals indicating the state of the engine from various sensors, and is, inter alia, configured to determine a fuel quantity to be injected and a corresponding timing of injection. More specifically, the ECU is configured to determine a desired fuel quantity to be injected to achieve a given demand (e.g. of torque or engine speed), and subsequently determines the injector control signal (injector actuation PW) corresponding to the desired fuel quantity. It may be noted that most of the functionalities and techniques (and their implementation) used in the inventive method are known in the art and will not be explained in detail. In general, the method according to the present disclosure can be implemented by hardware and / or software. In practice, it may conveniently be implemented by a control unit comprising a processor and a memory, such as e.g. the Engine Control Unit. In such case the memory may contain instructions which, when executed by the ECU / processor, cause the latter to carry out the present method. Turning now to Fig.1, engine components are represented in box 10 and comprise: - injectors 12 with associated drive unit, configured to selectively inject / discharge predetermined fuel quantities into the associated cylinders. The drive unit is conventionally designed to apply the control signals to the injector to discharge a fuel quantity corresponding to the respective fuel command for each cylinder combustion cycle; - an engine 14 with cylinders defining the combustion chamber; - an exhaust system 16 with an exhaust manifold and exhaust pipe, which collects and directs exhaust gases from the engine cylinders out of the vehicle, aiding in efficient engine performance and emission control; - a lambda sensor 18 responsive to the air-fuel ratio in exhaust gasses in the exhaust system 16. Specifically, the engine comprises a single lambda sensor 18 arranged in the exhaust system. The lambda sensor 18 may conventionally be an oxygen sensor, namely a wide-range oxygen sensor (also wide range or IIEGO sensor) where the sensor signal (i.e. signal representing the oxygen concentration) corresponds to the pumping current. Module 18 may thus typically includes a mapping that relates the oxygen concentration with the lambda value. Reference sign 20 indicates a fuel adjusting function, which outputs the fuel command Qc to the injectors drive unit 12. The drive unit 12 operates each injector by applying control signals thereto in such a way as to discharge the fuel amount Qc through one or more injection events for a given combustion cycle. In case cylinder fueling is done in one injection event, then Qc is converted into a drive signal having a duration PW using the above-mentioned flow-curves. The fuel command Qc is computed by fuel adjusting function 20 based on a base fuel amount Qb, on a global correction (here a global correcting factor) Cg and preferably on an ICFC correcting factor Ci. The fuel determination can thus be written as: Qc = Qb x Cg x Ci [Eq.1] The base fuel amount Qb may typically originate from a conventional fuel determination structure, not shown here. Typically, a driver torque function or a speed control function receives demands from various components, for example direct demand from the driver (accelerator pedal) or indirect demand via cruise control, demands from the transmission system, from driving dynamics, from the gearbox or demands related to specific components (e.g. accessory torque). The driver demand function coordinates these various demands and generates a global demand Td. This global demand Td may be limited / capped by a maximum demand Tmax that may depend on various factors. This function outputs a gross indicated demand Togross. A desired fuel mass Qd (or fuel demand) is then determined to meet the request Togross, typically by calculation based on IMEP (Indicated Mean Effective Pressure), cylinder volume and combustion efficiency coefficients. This is typically based on calibrated mappings. In parallel, a lambda set point is determined, referred to as desired lambda Ad, in function of the current operating point (engine speed, load...). As is known, the Lambda value determines the mass ratio of air and fuel in the combustion chamber, in regards to the stoichiometric air-fuel ratio. The lambda setpoint may be determined to optimize combustion efficiency, combustion stability and pollutant emissions (NOx). A desired air mass Md (representing the air mass desired in the cylinder) is computed based on the desired fuel mass Qd and taking into account the desired lambda Ad. The throttle and turbocharger gate positions are adjusted on the basis of the desired air Md. It may be noted that since the air loop is typically much slower than the fuel loop, the calculation of the fuel mass to be injected is done on the basis of lambda desired Ad as well as, advantageously, on the basis of the air mass Mf (or fresh air flow) actually entering the cylinder (hence not on the Md). Air mass Mf can be estimated based on the intake manifold pressure and temperature and volumetric efficiency. This logic privileges the respect of the air / fuel ratio and is advantageous during transitory regimes where the inducted air mass can significantly vary from the computed air mass Md. Finally, the base fuel amount Qb is determined as the fuel quantity to be injected in the next scheduled injection event / combustion. The base fuel amount Qb is advantageously determined from the lambda setpoint (desired lambda Ad) and from the fresh airflow Mf. The ICFC correcting factor Ci is determined in module 22 based on the signal of the lambda sensor 18. Module 22 may be configured to implement any Individual Cylinder Fuel Control, ICFC, i.e. a strategy designed to eliminate or reduce the imbalance in air-fuel ratio (lambda). Such functions are known in the art and will not be described herein in detail. US 6,382,198 discloses a known ICFC strategy, but this is only an example and this or other strategies may be implemented. Although ICFC correction is advantageous, it could be dispensed with. Conventionally, in liquid fuel engine, the global correcting factor Cg is a factor determined from an average closed loop lambda control module based on the sensor signal. The lambda sensor signal is read asynchronously, e.g. every 10 ms. The global correcting factor Cg is determined by a proportional integral (PI) controller function based on the deviation from the lambda setpoint. This global correcting factor Cg determination is implemented in module 24. <lnvention> However, the inventors have found that this continuous, closed-loop approach for the determination of the global correction is not suited to internal combustion engine powered by gaseous fuel, in particular by hydrogen fuel. Indeed, these engines nominally operate at a lambda value comprised between 2.3 and 2.6. As shown on figure 2, as the lambda value varies between 1 and 2, the corresponding proportion of oxygen in the exhaust gas vary between 0 and 10. However, as the lambda value varies between 2 and 3, the corresponding proportion of oxygen in the exhaust gas vary between 10 and roughly 13.7. The rate of change of the proportion of oxygen in the exhaust gas thus decreases as lambda increase, such that the proportion of oxygen slowly converges towards 21%. Hence at higher lambda values / oxygen proportions, inaccuracies in the oxygen measurement in the exhaust gas result in significantly increased inaccuracies when determining the corresponding lambda value. As shown on figure 3, the tolerance of the sensor for lambda = 1.6 is about 2%, but for lambda = 2.3, i.e. during normal operation of a hydrogen engine, this tolerance triples to 6%. The inventors have concluded that such inaccuracy is too large to reliably ensure the determination of an appropriate global correction. To solve this problem, the inventive method for operating a gaseous fuel internal combustion engine implements a global correction learning routine with an intrusive learning operation, a flowchart of which is show on figure 4. The learning routine may be programmed in module 24, which may advantageously include a PI controller function. In an initial step SO, a periodic timer TM1 is set to wait a set duration. When the timer ends or when an initial adaptation request is received at S1, a learning request is sent to the control unit at S2. For example, the TM1 timer may be set such that a learning request is sent at predetermined time intervals of e.g. between 20 and 200 operating hours. The control unit then checks if the learning (or enabling) conditions are met at S3. Learning conditions may be based on or more inputs or conditions selected from: sensor status, runtime, engine load, engine temperature, engine speed (RPM), engine idling. In embodiments, learning conditions may be that the engine is idling and warm (fully warmed up - around 90°C). If the learning conditions are not met, the control unit postpones the start of the intrusive learning operation until the learning conditions are met. Once the leaning conditions are met, the intrusive learning operation is started at S4. If, at any time during the intrusive learning operation, the learning conditions are no longer met (e.g. if a request from an operator is received), the intrusive learning operation may be aborted and another learning request is sent at S2. Alternatively, inputs from the user may be suppressed if the global correction learning routine is ongoing. The intrusive learning operation comprises the following steps, which are here repeated and performed in closed-loop: A test demand fuel quantity Qtest is first defined based on a test lambda value Atest comprised between 1.0 and 1.6. This Qtest is adjusted by a previous global correction factor CG_Previous. During the first iteration within the intrusive learning operation, Cg previous may be set to the current value of the global correction factor Cg. An Injection cycle is then performed by operating the injectors to deliver the test demand fuel quantity Qtest to the cylinders. A corresponding, actual lambda value Aactuai is determined from the lambda sensor, i.e. by measuring the concentration of oxygen in the exhaust gas. A new global correcting factor CG_new is then determined based on the previous global correction factor CG_Previous, and the ratio between the actual lambda value and the test lambda value, i.e. CG new = ^*CG previous. [Eq. 2] ^test This new global correcting factor CG_new is used as previous global correction factor CG_Previous for the next iteration of the intrusive learning operation, such that the intrusive learning operation is performed in closed-loop. The stability of the new global correcting factor CG_new is assessed at S5. The new global correcting factor CG_new is considered stable if the difference between consecutive new global correction is lower than a stability threshold for a predetermined stability duration (e.g. timer TM2). For example, the new global correction CG_new may be considered stable if the difference between consecutive new global corrections is lower than 0.5% of the previous value for 5 seconds. The global correction learning routine may have a maximum duration, and the new global correction may be considered unstable if no stable new global correction was determined within this maximum duration (e.g. until lapse of a timer TM3 initiated at the start of the learning routine). If the value is not considered valid and stable, the new global correction CG_new is discarded and another learning request is sent at S2 to perform another intrusive learning operation at a later time. If the value is considered valid and stable, the global correcting factor Cg is updated and set to the new global correcting factor CG_new. The updated Cg is used for the determination of the fuel command as previously described, using the relationship Qc = Qb x Cg x Ci [Eq. 1], clnitial start-up> It may be noted that it may be desirable to perform a first implementation of the global correction learning routine in the context of the "initial start-up" or "first fire” of the engine, i.e. the first time that the engine is started after assembly. This may typically be carried out in the manufacturing facility. Accordingly, the first timer TM1 may be initially set to 0 s, or a dozen minutes, for example, such that the global correction learning routine is started as soon as possible. In this context also, the present routine is advantageously programmed to allow the operator to select whether to correct the sensor accuracy or whether to directly correct the fueling. Alternatively, this selection may be preset in the system. This will now be explained with respect to Fig. 5, which shows a flowchart representing possible outcomes for the first implementation of the learning operation (i.e. learned just after engine first fire). So, during the first implementation of global correction learning routine (represented by C1), the method is configured to give the operator the option to decide whether to save the value of the first global correction CG_new as a lambda sensor correction or as a correction to the fuel injectors / quantity. This may be done through the use of a variable, the value of which is evaluated in C2. For example, at C2 it is checked whether variable Scor equals 0 or 1. Scor =1 indicates that lambda sensor correction is desired. This is done at step C3. The value of the first global correction is saved as a lambda sensor correction. That is, the lambda sensor is configured such that the actual lambda value is computed based on the ratio of the measured O2 concentration and this lambda sensor correction. This may be implemented as follows, where C02 is the actual oxygen concentration, Ccorr is a lambda sensor correction and Cmeas is the reading from the sensor signal: C02 = [Eq. 3] CCorr In step C3 the lambda sensor correction is assigned the learned value of the global correction: Ccorr = Cg__new. Since this first global correction is not directly used to correct the injected fuel quantity, Cg is set to 1. This is useful when the accuracy of the injectors has been verified shortly prior to the first run of the global correction learning routine, or when one is confident that they operate nominally. Otherwise, Scor = 0, which indicates that fueling correction is desired. Here the value of the first determined global correction may be saved as a correction to be applied to the fuel quantity (step C3’). In that case, the lambda sensor is configured such that the actual lambda value is computed directly from the measured O2 concentration (the lambda sensor correction Ccorr is set to 1). This first global correction is then directly used to correct the injected fuel quantity, i.e. Cg is set to this first determined value of the global correction CG_new. It may be useful to store this first global correction independently from the global correction. The evolution of the global correction relative to the first global correction can then be used for monitoring and diagnostic of the injection system over its lifetime. One possible implementation of Fig.5 may use the relationship: Cg CQ ^irSf. GQ_peri0CtiC [Eq. 4] Cg is the factor used in Eq.1. However Cg is implemented as a product of two correction factors: CG_first and CG_periodic. CG_Periodic is updated by CG_new when the global correction routine converges (this after the first implementation at engine first fire). CG_first is updated in the context of the first determination, in step C3’, i.e. where fueling correction is to be applied. In such case, CG_first=CG_new and Ccorr = 1. In case step C3 is implemented, then Ccorr =CG_new and Cc-first = 1. It remains to be noted that the above implementations have been explained considering that the lambda sensor is a proportional oxygen sensor. The inventive method may also be implemented using a binary switching sensor, qualitatively indicating whether oxygen is present in the exhaust gas. In such case, the sensor output signal switches or toggles between first and second states corresponding to lean and rich conditions of the sensed exhaust gas, relative to a stoichiometric air / fuel ratio. In case such binary switching sensor is used, then the learning routine and specifically step S4 is performed with the test lambda value Atest set to 1.0.

Claims

1. Method for operating a gaseous fuel internal combustion engine, the engine comprising a plurality of fuel injectors each configured to deliver gaseous fuel in a corresponding cylinder of the engine, and a lambda sensor positioned to respond to exhaust gases generated by the cylinders, the method implementing at least one fuel injection strategy wherein:- the fuel injectors are operated to discharge into engine cylinders fuel quantities corresponding to a predetermined fuel command (Qc);- the fuel command (Qc) is computed from a base fuel amount (Qb), which is determined from a desired lambda value (Adesired), and adjusted by a stored global correction (Cg);characterized in that the global correction (Cg) is learned periodically by implementing a global correction learning routine comprising the steps of:- defining a test demand fuel quantity (Qtest) based on a test lambda value (Atest) comprised between 1.0 and 1.6;- performing injection events by operating the plurality of injectors to deliver the test demand fuel quantity (Qtest) to the cylinders;- determining a corresponding, actual lambda value (Aactuai) from the lambda sensor;- determining a new global correction (CG_new) based on the actual lambda value (Aactuai) and the test lambda value (Atest);- updating the stored global correction (Cg) by storing the new global correction (CG_new) in a memory of a control unit.

2. Method according to any of the preceding claims, wherein the desired lambda value (Adesired) is greater than 2.0, and is preferably comprised between 2.3 and 2.6.

3. Method according to any of the preceding claims, whereby the global correction is a global correcting factor (Cg) determined from a ratio between the actual lambda value (Aactuai) and the test lambda value (Atest), andwherein the fuel command (Qc) is determined from a product between the base fuel amount (Qb), and the global correcting factor (Cg).

4. Method according to any of the preceding claims, whereby the steps of defining a test demand fuel quantity (Qtest) based on a test lambda value (Atest), performing injection cycles by operating the plurality of injectors to deliver the test demand fuel quantity (Qtest) to the cylinders, determining a corresponding, actual lambda value (Aactuai) from the lambda sensor, and determining a new global correction (CG_new) based on the actual lambda value (Aactuai) and the test lambda value (Atest) are repeated during a single global correction learning routine, andwhereby the test demand fuel quantity (Qtest) is adjusted by a previous global correction (CG_Previous) based on a new global correction (CG_new) from a previous iteration, such that the global correction learning routine is performed in closed-loop.

5. Method according to the previous claim, whereby the step of updating the stored global correction (Cg) by storing the new global correction (CG_new) is performed only if the new global correction is stable.

6. Method according to the previous claim, whereby the new global correction (CG_new) is considered stable if the difference between consecutive determined global corrections is lower than a stability threshold for a stability duration.

7. Method according to claims 5 or 6, whereby the global correction learning routine has a maximum duration, and whereby the new global correction (CG_new) is discarded if no stable new global correction was determined within this maximum duration.

8. Method according to any of the preceding claims, whereby the global correction learning routine is performed periodically, preferably with a time interval betweentwo consecutive implementations of the learning routine greater than 20 operating hours.

9. Method according to any of the preceding claims, whereby the global correction learning routine is begun when at least one predetermined learning condition is met, which includes the engine being in idle condition.

10. Method according to any of the preceding claims, whereby the global correction learning routine is aborted if an input from a user is detected.

11. Method according to any of the preceding claims, whereby if the global correction learning routine is a first implementation of the global correction learning routine, the new global correction (CG_new) is stored as a lambda sensor correction, and;whereby the lambda sensor is configured to determine the oxygen concentration in the exhaust gas, wherein the measured oxygen concentration in the exhaust gas is corrected by applying the new global correction (CG_new).

12. Method according to any of the preceding claims, whereby if the global correction learning routine is a first implementation of the global correction learning routine, the new global correction (CG_new) is stored independently from the global correction (Cg).

13. Method according to any of the preceding claims, whereby the oxygen concentration in the exhaust gas is measured by determining presence of oxygen by means of a switching sensor and the global correction learning routine is performed with the test lambda value equal to 1.0.

14. Method according to any of the preceding claims, whereby the fuel command (Qc) is computed from the base fuel amount (Qb), the stored global correction (Cg), and an Individual Cylinder Fuel Control correcting factor (Ci) determined from a signal of the lambda sensor.

15. A control unit comprising a processor configured to perform the method according to any of the preceding claims.

16. A gaseous fuel multi-cylinder internal combustion engine comprising fuel injectors arranged to fuel cylinders and a lambda sensor positioned to respond to exhaust gases generated by the cylinders, and a control unit according to claim 16.

Images