SE504197C2 - Method for determining position for max. pressure during combustion in internal combustion engine - Google Patents

Abstract

The ionising sensor output signals are sampled under a measurement window activated after combustion has been initiated. A primarily applied parameterised function is curve-adjusted to the sampled output signals from the ionising sensors. The primarily applied function acquires parameter values which generate at least two local maxima (PF,PP). The maxima are within the pref. interval of 17 crankshaft degrees. The primarily applied function is optimised with regard to the relative positions and amplitudes for the generated local max. points and with regard to the min. deviation of the primarily applied function from the sampled output signals from the ionising sensor. The position for the latter local max. points (PP) calculated in crankshaft degrees and relative to top dead centre is used for determining the position for max. pressure during combustion.

Description

504 197 2 20 30 dominant that it is more a rule than an exception that the zero crossing is missing where the solution shown in US, A, 44l7556 can not be applied.

SE, A, 9402686-1 shows another variant in which the center of gravity of the integrated value of the ion current signal is used to determine the position of maximum combustion pressure. This solution is more stable than the solution shown in US, A, 4417556, mainly in the case of operating cases / combustion where no zero crossing of the derivatives of the ion current signal takes place during the post-ionization phase. A disadvantage of the center of gravity determination of the ion current signal is that the result is affected by how and when the integration starts, which integration must be initiated so that the integrated value does not include signal values which are attributable to the ionization phase. Under certain operating conditions / combustion, the transition from the ionization phase to the epithelialization phase is not unambiguously determined by the nature of the ion current signal, which means that the integration must be initiated adaptively depending on the appearance of the ion current signal and possibly supplemented with a back-interpolated ion current signal from integration.

THE INVENTION The object of the invention is to be able to determine in a safer way and during a larger operating range the position of maximum pressure during combustion in an internal combustion engine, without the use of a pressure sensor and only based on the output signal from a sensor which detects the degree of ionization in the combustion chamber.

Another object is that from an output signal from an ionization sensor arranged in the combustion chamber lcurma detect the position of maximum pressure during combustion even in operating cases where the output signal of the ionization sensor does not give a local maximum during the post-ionization phase, which can occur during combustions where such a long duration that a local maximum in the output signal of the ionization sensor does not occur at the crankshaft angle positions where the combustion accident has its greatest value. This type of combustion can occur sporadically during certain operating cases but also regularly during certain operating cases, at low loads and speeds and when the engine is regulated extremely lean, so-called lean-burn control, or when large amounts of exhaust gases are recycled to the combustion chamber, so-called EGR control . Another object is to be able to obtain from the output signal of the ionization sensor a more stable determination of the position of maximum pressure during combustion, which determination is not as sensitive to disturbance for sporadic and short-term interference pulses on the output signal of the ionization sensor. Another object is that, based on the position of the maximum current during combustion determined from the ion current signal, it is possible to enable a regulation of the ignition time point in a feedback manner so that an optimization of the work taken can be obtained.

BRIEF DESCRIPTION OF THE INVENTION The method according to the invention is characterized by the core design part of claim 1.

The distinctive features and advantages of the inventive method Other invention appear from the characterizing parts of the other claims and the following description of exemplary embodiments.

The description of dehydration examples is made with reference to urer gures specified in the following fi yeast drawing.

LIST OF FIGURES Figure 1, shows the appearance of the ion current signal Um which can be obtained with an ionization sensor arranged in the dry combustion chamber, where the ignition phase SjoN, the ionization phase Fm; and the post-ionization phase PIQN are indicated. Figure 2 shows 6 combustion arcs with detected ion current signal l and two Gaussian curves G1 and G; which curve is adapted to the ionization phase of the ion current signal and the post-ionization phase; Figure 3, shows 6 dry bursts in succession with detected ion current signal I and where the two curve-fitting Gaussian functions have been added F; Figure 4 shows the second curve-fitting Gaussian ion G2 relative to the pressure P in the dry-burning chamber as it is registered with a pressure sensor arranged in the dry-burning chamber.

DESCRIPTION OF PREPARATION / PEL 1 Figure 1 schematically shows the appearance of the ion current signal Um; obtained with an ionization sensor arranged in the combustion chamber. The ionization sensor can suitably be included in a measuring system corresponding to that shown in SE, A, 9500189-7, so further descriptions of the ionization sensor itself are not included. The relative level Um is indicated on the Y-axis and is here measured in volts, and can be between the output signal O volts to just over 2 volts. The X-axis indicates crankshaft degrees ° VC, where 0 ° indicates the upper dead center when the piston is in its uppermost position.

During the SION phase, a spark is generated which occurs within a crankshaft angular range 25-15 crankshaft degrees before the dead center, with the dry ignition being relevant during the prevailing operating condition, mainly depending on load and speed. The generation of the spark generates a first pulse, caused by the overshoot in the spark plug gap during the so-called break down phase, but this high measuring pulse can be filtered out, and its measured value is not used in the preferred embodiment. As a matter of fact, the acquisition of measurement signals is controlled by a microcomputer, in such a way that the microcomputer only reads the ion current signal at certain determined motor positions or times, i.e. in certain measuring windows. These measuring windows are preferably activated depending on the ignition time, so that these measuring windows are opened long enough after the flashover phase SIGN has ended.

After the flashover phase, a ame ionization phase (Flame ionization phase), in Fig. 1 called FIQN, starts, during which phase the measuring voltage is affected by the establishment of the burning core of the air-fuel mixture in or at the spark plug gap.

After the flame ionisation phase, the post ionization phase (Post ionization phase), in Figure 1 called PION, starts, during which phase the measuring voltage is affected by the combustion in the combustion chamber as the number of ionizing particles increases with temperature and combustion pressure. Typically, during the ionization phase Pm, the ion current signal UION reaches a maximum value, indicated by PP in Figure 1, when the combustion pressure has developed to its highest value and the front has reached the walls of the combustion chamber, which results in an increase in pressure. The transition between the ionization phase and the post-ionization phase as well as the peak values within each phase can be detected, for example, by a derivative circuit, or alternatively a derivative in the control unit's software. The first zero crossing of the derivative dUIONIdVC detects the peak value PF, the second zero crossing detects the transition from the jon ionization phase to the post-ionization phase and the third zero crossing detects the peak value PP.

The ideal curve for the ion current signal thus consists of three different phases; the ignition phase SION, the flame ionization phase FION and the post-ionization phase PION, which phases form the basis of the inventive algorithm. Since both the jon ionization phase and the post-ionization phase have a characteristic shape to the appearance of the ion current signal, this characteristic shape can be used to obtain an ideal curve adapted to the ion current signal. As shown in Figure 4, much develops in the combustion chamber, the curve P, as a bell-shaped curve. of the ionization signal corresponding to the pressure in a corresponding manner should mimic a bell shape.

A function that fits well with this clock form is a Gaussftmlction. The Gaussian function can be mathematically expressed as; f (e) = e,. e '° = <° fl = f This Gaussian function can therefore be used to obtain a curve-fitting ideal curve for the post-ionization phase of the ion current signal.

For a curve-matched ideal curve to the ionization phase of the ion current signal, a Gaussian function is also used here, because the ion current signal during the ionization phase has a high peak value and in the start phase and end phase of the ionization can be assumed to have a zero value. 20 25 30 5 åso4 197 There may be better functions for an approximate ideal curve for each phase than the above-mentioned Gaussian functions. The choice of Gaussian functions is made partly because the Gaussian function is relatively simple.

The curve fitting of the selected functions to the ion current signal is suitably done by some form of optimization model. A standard optimization model can minimize the square of the error signal. This type of optimization is called the least squares method, and can be interpreted as minimizing the energy in the error signal. The optimization then becomes a minimization of a specific condition, which is a well-established problem. However, this does not mean that the problem is easily solvable, on the contrary, a relatively insoluble problem arises when the functions are not linear or convex.The optimization then does not converge unambiguously due to the functions not being convex An example of how this optimization can be solved and how a proposed algorithm for curve fitting can be implemented is described in The following section: Implementation of an algorithm The Jonström signal consists of three phases where the last two phases coincide with the phases during combustion where the interesting information is located.Based on this assumption, the algorithm tries to adapt a parameterized function to these last two phases, which parameterized function in this case consists of two Gaussian shaped functions.

The function to be adapted to the ion current signal is, as previously suggested, a Gaussian function, which is mathematically described as; fte> = al. f ° = <° r = f, where the choice of parameters is, ot; , ot; can be used to change the shape of the parameterized function in such a way that; 0: 1, changes the height or amplitude of the function, ot; , changes the width of the function, and ot; , changes the position of the function center.

The ion current signal shall be adapted to a sum of two signals, where the sum is two added Gaussian functions as above. If the sampled ion current signal is denoted by; In (9) and the two Gaussian fluorescences are named f1 (6 l and f2 (9), then the relation I (9) = f1 (9) + f; ; f1 (6) = ot1e '° = [° "“ * f f2 (6) = ßle-ßzh-Öiy _ 504 197 6 20 30 When optimizing the curve fit, the least squares method was used. The method can be interpreted as minimizing the energy in the error. The fault function, err (6), which is to be minimized can then be expressed according to; me) = 1 (e) - f, (e) -f2 (6) [mo] The minimization takes place over the current Since the ion current signal is sampled, the ion current signal is time-discrete values at points 6 l, where i indicates the order number of the sampling. Using the least squares method, the thirst function will thus be minimized; f '= _zs (err ( 9l)) z ie l- Thus, with the assigned functions, the thirst function 1 'shall be minimized with respect to the six parameters al, al, a3, ßl, ßl and ß3; z 2 2 rlapaaaaßaßaßl) = § (1 (@. «) - «1e“ = l ° f “= l - ßlfßff-ßß)) [sin Some intuitive conditions can be assigned to the loss function.A first assumption is that al . <_ ß, [6.12] must be valid for the first Gaussian function fl (6) to be adapted to the fl amfront phase which occurs before the post-ionization phase. This condition is not necessary to test during optimization or curve fitting. Instead, this condition can be tested after the optimization is completed, and if the condition is not met after the optimization, the functions change. The next condition is that; al) 0 ßz> 0 which results in the thus paramulated function obtaining a cloclform.

[613]

[614] l. Analytical solution of two parameters Solving this optimization is a complicated one because the functions are non-linear and non-convex.

However, two parameters can be solved analytically, namely the parameters al, ß l. This reduces the number of parameters during optimization to "only" four. The basis for this parameter reduction is given below. The partial derivative of the loss function F with respect to parameters a1 and ßl is given by; al__ _ 'Gzfaf fl af _ -ß: [9. "5:] 2 _' fl zæf fl sf aal - Ziezslpßäl) ale l ßle) e, samt ar _c__a: _ _: _ll_jz í = _ziezsí (l (ei) _ale: (51 s ) _ßle Btw. Bs)) _e BÅÜ. B) 20 30 F 504 197 8 I 'ö f' For a stationary point, condition ä; = 0 oc -a-ß- = 0 must be met. In a system with l 1 two equations, the parameters a1 and B1 can be solved, a reversal then gives that: a1z (e4 = t @, a, f) * + ßl 2e _ =, <@, l¶,> = e-ß, <@ i- ß,> 2 = 2, (eí) ea = 2 res, res, - tes, ß12 (e-ß, (@, - ß,> *) ° m! 2e «== <ß.« =, 12e- ß, 1 = XML) e-ß = (@. ~ -ß,> 2 resi tes, tes, - which equations are denoted according to; otlA + ß, B = D, ß1B + ot1C = E The parameters otl and ßl can thus be solved out according to; a _ CD - BE AE - BD 1 'CA - 122 CA - 32 The last step is to verify whether the stationary point is a minimum point or not, one, and ßl = [615] local minimum is characterized by a positive second Second order derivatives of the error deviation error F with respect to al and [51 can be drawn as follows; Ä L? = Zšftfltt fl liï> 0 62 r _ __ == a B1 2: zš- (e Bah. 51))> 0 2. Numerical solution of four parameters For each value set of the parameters otz, ot3, ß2 and ß3; thus, it is possible to trigger the parameters a1 and [31 _ The optimization algorithm thus only needs to focus on four variables instead of six. The selected optimization algorithm is an algorithm based on the gradient. If the problem is convex, the algorithm will converge towards an optimum. In this case, the problem is not convex, which means that the algorithm can converge towards a local minimum, instead of the global minimum. However, it is to some extent possible to avoid the function converging towards an incorrect local minimum. If the algorithm starts from a good approach that is close to the global mini-point, then the algorithm will also converge towards this global mini-point. The closer the approach is to the global mini-point, the faster the algorithm will converge towards the global mini-point. It is thus important that the algorithm starts from a good approach.

The first Gaussian function corresponds to the frontal phase and the maximum point generated during this phase in ion current signal extraction. Therefore, ot 3 can be assigned the value of this maximipunlct. The second Gaussian function corresponds to the post-ionization phase and the maximum point generated during this phase in the ion current signal. Since the maximum point during the post-ionization phase does not always exist, it is also not practically possible to search for this maximum point. Extensive measurements of ion current signals show, however, that the maximum point during the post-ionization phase is normally in the range of 14-25 crankshaft angles, with an average value of 17 crankshaft degrees. This means that a good approach is to let the parameter [33 assume the value ot3 + 17 crankshaft angular degrees.

The parameters (z 2 and ß 2 are assigned values which are typical of normal ion current curves, namely ot 2 = 0.01 and ß 2 = 0.003. This is a heuristic approach that always provides a permitted initial solution for the algotite.

Since the algorithm is based on the gradient, it is required that the partial derivative of the error deviation function F is calculated as below; E. =) _a1e-G: (9; ~ d; \ § _ ølerßzüifßsšä pie-Gzwrdalš / a3) 2: _ = 2> 1 (1- «1e ~ == <° ~ = fl f -) ßle-ß = <° «~ ß fl f ar 'azæf fl sf' ßzæfßaf 'az- (ei fl sf í = ° 4z (l (ef)" aie _ ß1e hie azæf-as) .___ I¿ = __a1e-d: (9, ~~ d;) \ _ øle-ßzßfßalš) p $ e-5: (9; “Bs) ë ßáeí _ BIS) This means that the gradient Q for the error deviation function F can be drawn as; fa I- N (1 QJQJ Pm. [6.16] QJQJ" l QJQJ -15 \ a Bs) Some parameters must be set start values before the algorithm starts. These values are: now (maximum number of iterations), e (a small set value for the error deviation function), LBD (a lower limit value for the error deviation function) and UBD (an upper limit value for the error deviation function).

If É denotes the gradientš in the iterative step n, and X n denotes the vector fr? , with the values of (X2 a,, ß2 and ß3 isteget n. 7 The pseudocode for the optimization algorithm becomes “504 197 FOR every cycle DO BEGIN (* Find aninitial guess, X1 *) 5 Search maximum for the ionization current, Gm a; = 0m , ß3 = 0m + 17 °, a2 = 0.01, ß2 = 0.002 (* Initialize the optimization algorithm *) UBD = °°, LBD = O, stop = false, n = 1 REPEAT 10 Compute a! and ß! from [6.15 ] Compute å, and F from [6.1 6] and [6.11] m 1 "(UBD THEN BEGIN Save the current solution as best UBD = 1" 15 END (* Test if stop *) IF (EM = O) or (Re ) or (rànm) THEN stop = true (* compute the step length, t, and the new XW *) 1 "- LBD f = 2 n = in arr-fa, å, 1 + 1-6 20 (* test the constraints [6.13] and [614] *) IF (a: in Xn fl) (0 THEN Set (12 to the previous value in 1 ,, IF (ßz in XW) (0 THEN Set ß: to the previous value in X ” n = n +1 UNTIL stop = true 25 (* test the constraint [612] *) IF 013) ß 3 THEN BEGIN (* Exchange the Gaussian functions *) aiæßi »azæßz» aaæßs END 504 197 'o 20 30 This algorithm stops after maximum now rose, dependent on the test condition. The algorithm can stop before that if either the gradient or the loss function reaches a sufficiently low value. In most cases, the algorithm re-runs the maximum number of iteration steps assigned.

Figure 2 shows the two Gaussian curves G2 and G2, respectively, obtained by the optimization algorithms at six consecutive combustions in a combustion chamber. The Gaussian functions G2 and G2, respectively, are optimized against the sampled ion current signal I so that the sum of the functions G2 and G2, respectively, gives a composite function which, as a result of the optimization algorithms, mimics the ion current signal I. Figure 3 shows the composite function F, which thus constitutes the sum of the two Gausses. the functions G1 and G2_ together with the ion current signal I. The figure shows that the proposed algorithm gives a good curve fitting of two Gaussian functions on the ion current signal In Figure 4 the second Gaussian function G2 is plotted against a signal P from a pressure sensor arranged in the combustion chamber. The peak value of the Gaussian G2 G2 device corresponds well with the position for maximum pressure As can be seen from the guras, the algorithm is capable of detecting the position for maximum pressure even in operating cases / combustions where the ion beam signal has no local maximum point during the post-ionization phase, corresponding to a zero review of first derivative differences may exist between the position of maximum pressure and the position of the peak value of the second Gaussian function G2, for example in combustion cycles 2 and 3, where the ion flow signal during the flame ionisation phase FION is very irregular. In combustion cycle 3, during the jon amionization phase FION, two local peaks appear on the ion current signal. An improved curve fit and better agreement between the maximum pressure position and the position of the second Gaussian function G2 peak value can be obtained by detecting whether there are two local dead center values, ie before ö. 0 crankshaft degrees on the x-axis in fi gur 2, and then set a third Gaussian function -u fjte) = elen fl n fl which is set a representative value with respect to parameters 51, which corresponds to the amplitude of the second peak value below FION, 0-2, which corresponds to the width of the second peak value, 52, which corresponds to the position of the second peak value below FION.

The ion current signal must then be adapted to a sum of three added Gaussian functions in a corresponding manner. If the sampled ion current signal is denoted by; In (9) and the three Gaussian functions are named f1 (6), f: (9) and f3 (6), the relation 20 'is obtained f 504 197 I (9) = f, (9) + f2 (6) + f3 (9). Such a selectively activated curve fitting results in the second Gaussian function obtaining a better match with the pressure curve. especially in Nest Burns where the ionization signal is irregular during the flame ionization phase FION and during this phase exhibits more than one local maximum point.

If more than two peak values exist before the upper dead center, a corresponding number of Gaussian functions can be adapted to the frfront phase in order to further improve the conformity of the second, or the last, Gaussian function with the pressure curve.

The proposed method of using a model where a curve fitting is performed on the ion tube signal can of course use other types of parameuized functions than the Gaussian functions proposed here, especially with respect to those parts of the composite function which are to represent the fl front phase FION. The most important thing is that the parameterized function can give at least four injection points, which corresponds to at least the 6th degree polynomial. The proposed optimization algorithm is relatively time-consuming, which makes it difficult to implement in control systems which in real time must have time to regulate the ignition time between the burnings. The optimization algorithm shown has only been chosen to demonstrate the solubility of the problem, and has not been optimized per se. There can therefore be more time-efficient algorithms that can be selected and which, together with faster processors, can have time for real-time control. The first Gaussian G1 can also be used to obtain information from the combustion. In the same way as stated in SE, A, 9500189-7, the slope of the rising fl anchor on the first Gaussian function can be used to detect the current mixing ratio air / fuel, ie lambda value at each individual combustion.

Claims (7)

504 197 20 30 35 ii PATENT CLAIMS A method for determining the position of maximum pressure during a combustion in a dry combustion engine, which dry combustion engine comprises a sensor arranged in the dry combustion chamber which detects the degree of ionization in the combustion chamber characterized in that the output signal of the ionization sensor is sampled and activated. a primarily applied parameterized function is curve fitted to the sampled output signal from the ionization sensor, which primarily applied function is assigned parameter values which generate at least two local maxima (PFPP) which maxima are in the range 10-24 crankshaft degrees apart, preferably 17 crankshaft degrees apart, - after which it the function employed is optimized with respect to relative positions and field plots for the generated local maximum points and with respect to minimal deviation of the primarily employed function from the sampled output signal from the ionization sensor, and where the position for The local maximum point (PP) of the optimized function calculated in crankshaft degrees and relatively upper dead center is used to determine the position of maximum pressure during combustion. Method according to claim 1, characterized in that the sampled output signal from the ionization sensor is curve-fitted in a first part of a first Gaussian function (G1). On the sampled output signal (I), which first Gaussian function (G1) is laid out depending on the start, and that the sampled output signal from the ionization sensor is curve-fitted in a second part of a second Gaussian function (G2) on the sampled output signal, which second Gaussian function (G2) is laid out later than the first Gaussian function (G1) with a relative position predetermined relative to the first Gaussian function, the two Gaussian functions are added and optimized with respect to relative position and field amplitude for minimal deviation of the added Gaussian functions from the plotted output signal. and where the second Gaussian function (G2) after optimization and minimization of the added Gaussian functions (G1 +62) relative to the sampled output signal (I) was used to determine the position of maximum pressure during combustion Method according to Claim 2, characterized in that the first Gaussian function (G1) Pfimer is applied so that its amplitude maximum coincides with the position (PF) for maximum output signal (UION) from the ionization sensor. Method according to claim 3, characterized in that the second Gaussian function (G2) is primarily applied so that its maximum amplitude occurs within a crankshaft angle range 10-24 degrees after the position for maximum output signal from the ionization sensor, preferably 17 degrees after the position of maximum output signal fi- from the ionization sensor. Method according to claim 4, characterized in that the measuring pattern opens already a few crankshaft degrees before the upper dead center, preferably 20-10 degrees before the upper dead center depending on the current ignition time and end of the flashover phase (SIGN), and if the sampled output signal from the ionization sensor undergoes a local minimum before upper dead center, a third Gaussian function (G3) is applied which third Gaussian function is primarily applied so that its amplitude maxima coincide with the position of the first local maximum point of the ionization sensor output signal (I) before upper dead center (0 VC) during the amionization phase (FIQN). since the first Gaussian function is laid out with its amplitude maximum at the next local maximum point of the ionization signal during the ionization phase (FION), preferably before the upper dead center. Method according to Claim 2, 4 or 5, characterized in that the position of the amplitude axes of the second Gaussian function (G2) after optimization and minimization of the added Gaussian functions (G1 + G; + G3) relative to the sampled output signal (I) is used for determination of the maximum pressure mode during combustion. Method according to claim 6, characterized in that the ignition time is regulated depending on the detected position for maximum pressure in such a way that the position for maximum pressure, which position depends on the ignition time, is regulated against being within a dry-determined interval depending on at least load and speed.