DE10336598A1 - Vehicle combustion engine diagnosis procedure records measurements in compressed form using threshold to select linear or logarithmic function - Google Patents

Abstract

The vehicle combustion engine diagnosis procedure records measurement values and events and reduces the data by conversion to a format with fewer bits using linear or logarithmic conversion above and below a threshold with either function call or compression table (Gij) and storage in four time intervals.

Description

The The invention relates to a diagnostic method for detecting the state of Internal combustion engines by means of compressed load collectives. The load collectives are recorded in compressed form on-board and stored. The evaluation of the load collective takes place off-board.

Stress collectives are eg from the DE 100 60 694 A1 known. Thereafter, load collectives on the one hand accumulated times (sampling time intervals) of a particular operating state, for example, the time total of the operating conditions that were above a predetermined speed or whose operating temperature was above a predetermined coolant temperature. On the other hand, events are cumulated, eg the number of cold starts or the number of kilometers driven.

Farther are from the above German patent application, the classification sizes and Composite classifications known. Here are two measured variables in fixed intervals sampled and the value pair corresponding cell of a frequency matrix incremented. This corresponds to the so-called two-parametric instantaneous value counting or also residence time classification. An example of this is the Classification engine load over Turning number, which is regarded as standard classification for engine load collectives becomes.

The Detention and storage of operating states is here via the Operating time of a motor vehicle from a service stop to the next service stop, on which the stored load collective for further evaluation can be read out, very memory intensive. In the case of scarce EEPROM memory resources Therefore, we are looking for suitable data compression methods.

Known data compression methods, such as from the DE 100 16 153 A1 known, are only of limited use for on-board load collectives. Known data compression methods use redundancies in the form of recurring characters or strings (eg of text or image information), which replace them with relatively short codewords as placeholders. By its nature, this principle can only be applied to already existing digital information packets that have to be analyzed and compressed with the data compression algorithm. The compression algorithms are very expensive and are not suitable for real-time compression of operating data. The elaborate algorithms alone require storage space and computing power that are not available in simple control units of a motor vehicle. The required extreme simplicity for onboard applications is not given in known data compression methods.

Task according to the invention It is, therefore, a possibility specify the data compression used for on-board applications of Motor diagnostics can be used.

The solution succeeds with the features of claim 1. Further advantageous embodiments are in the dependent claims and included in the description.

With the invention mainly the following advantages are achieved:
A very simple and very fast data compression algorithm is presented. This allows a wide range of applications for the onboard detection of load collectives, in particular for the purposes of engine diagnostics of internal combustion engines.

The inventive method takes in the field big binary numbers a possible Error due to shortening the memory word lengths in purchase. However, it is ensured that the maximum occurring relative error remains controlled and limited. In the area smaller binary numbers works the data compression according to the invention, except for the inevitable quantization error, without data loss and without mistakes. This ensures that in load collectives Relatively rare events, however, for the determination of wear a very big one Can influence e.g. one-time overspeeding the engine in cold condition, not lost by the data compression walk.

in the The following is the process of the invention explained in detail by means of graphical representation. Showing:

1 An exemplary approach to off-board assessment of stress collectives,

2 A schematic representation of the compression function,

3 A schematic representation of the influence of normalization on the compression function,

4 The influence of quantization in decompression,

5 A schematic representation of the entire compression function,

6 The relative error in 16- to 8-bit compression,

7 The relative error in 16- to 6-bit compression,

8th A schematic representation for the determination of table points,

9 Various load collectives suitable for engine diagnostics.

Generally serve load collectives of obtaining basic information about the Vehicle life history, i. both the driving and environmental conditions as well as the specific driving style. They represent the sum of the burden, so to speak an engine and can as a tool to assess engine oil consumption, engine wear or component wear (e.g. Cooling system Belt) are used.

As a general rule Load collectives are very memory intensive, as many engine operating variables, the into several collective blocks be split, save. To the planned series use To cope with it, a compression method became economical Use and reasonable limitation of the load collective needed, permanent read / write memory, RAM or in particular EEPROM, prepared.

in this connection In particular, the interests of the load collective were considered. However, the procedure can be everywhere be used where many amplitude values for which a defined, low error can be accepted, saving space must be stored.

The The present invention is primarily concerned with close inspection the proposed compression and storage method. The content The load collective itself will be touched only briefly.

realization of stress collectives

Stress collectives provide on the one hand
accumulated times, eg sampling time intervals, of a specific operating state, eg the time total of the operating states "high speed" or "high temperature of the cooling water",
to become another
Cumulative events,
eg the number of cold starts or the number of kilometers driven.

Farther is for the detection of some operating conditions the correlation between two motor operating variables of Interest, for what Advantageously, composite classifications can be used. at In a composite classification, two measured variables are sampled at fixed time intervals and the cell of a frequency matrix corresponding to the value pair incremented. This corresponds to the so-called two-parametric instantaneous value counting or also residence time classification. An example of this is the Classification engine load over Speed, which is considered the standard classification for engine load spectra can be.

Furthermore There are numerous other one or two parametric counting and Classifying method, e.g. the rainflow count, the predominantly used for recording the fatigue strength in materials mechanics becomes.

load collectives for the engine diagnostics

The load groups for the engine diagnosis are in 9 shown. Due to their general structure, they can be transferred to any motor after appropriate adaptation of limit and threshold values.

For this were following "base sizes" and "classification sizes" defined.

The classification frequencies of two engine operating variables in the form of a two-parametric instantaneous value count are recorded in the classification variables:
Cooling water temperature T_KW over crankshaft speed N_KW,
Engine load LAST over crankshaft speed N_KW,
Motor load LOAD above travel speed V. (see 9 )

The load collectives are stored onboard separately after 4 time ranges, preferably in an EEPROM memory:
Present (about 15 000 km driving distance)
Past 1 (about 15 000 km)
Past 2 (about 15 000 km)
Total past (whole engine life) - basic sizes only

The Data from the present, past 1 and past 2 become according to the principle of a FIFO shift register at longer intervals, approx. 300 h engine operating time, stored as contiguous blocks. By suitable Choice of the grid are combined by the two past blocks covered a maintenance interval (30 000 km).

in the Traps of scarce EEPROM memory resources can block the past 1 and past 2 compressed in the manner described here because their memory contents remain frozen.

The Collectives of the total past (base sizes only) are running in shorter periods updated, e.g. at every engine stop. They stay - like that Present-block - expediently uncompressed, so that a constant, linear incrementing possible is. However, for the time range of the total past only basic quantities stored, so that from the outset does not accumulate too much data.

It was used as the basic sampling time for the time variables 1 sec selected. The included in this rhythm readings are now classified and then in a slightly coarser grid (1 ... 20 sec, depending on the measured variable) in the stored actual collective counters.

The derived from the so-defined load collectives in principle Statements are briefly summarized below.

In The load collectives are therefore the real driving and environmental conditions quasi stored as a vehicle and engine CV in condensed form.

The The above load collectives can be identified in the following diagnosis Use applications:

a) load collectives as a basis for comparison and evaluation for determined wear or consumption or defects occurred.

Examples:

of faulty and measured increased Engine oil consumption in connection with an accumulation of load / speed points with low target oil consumption can be a damage close the engine.

of faulty and measured increased Fuel consumption combined with frequent short distances (many Cold starts) and rarely warm-running engine (classification N_KW / T_KW) lets the Conclude that the Engine is ok, but a driving defect is likely is.

The trend increased Cooling water temperature (Classification N_KW / T_KW or frequency T_KW> Max) with predominantly low driving resistance (Classification LAST / V_TACHO) and little stop / go operation (gang collective) lets the Conclude that a damage of the cooling system comes into consideration.

b) Increasing the informative value of damage statistics.

Requirement:

The Collective collectives are included in the damage statistics and these can be specific, observed damage to the engine / components assign.

From this may arise important notes for constructive improvements for the areas development / construction.

The Stress collectives can also as a basis for suitable test bench programs serve the "damaging" load cases are targeting.

Example: frequent Coincidence of the failure of a certain aggregate or component associated with a typical "stop / Go collective "(Gang collective or classification N_KW / T_KW) gives an indication of damaging influences especially at low speeds / idle, it can be targeted Studies are made, e.g. harmful vibrations or resonances discover.

description the compression method

Scheme of Off-board Bewertuna

in the The following is a general example of the composite classifications Procedure of off-board assessment of stress collectives described.

in this connection the load collectives are evaluated with maps by: each individual classifying field is weighted.

An overall evaluation is thus obtained from a weighted summation of the individual frequencies zij and relativization to the total frequency zg = Σ zij (= sum of all entries):

In this B is the overall rating, Gij are the rating weights, zij the meter readings of Cells i, j.

• – ein Fahrprofil, z.B. Kurz/Langstreckenfahrt, Steigung/Ebene, etc.
• – der spezifische Verschleiß eines Bauteiles oder des Motors
• – der spezifische Verbrauch, z.B. des Motoröles oder des Kraftstoffes

abgeleitet wird.The general term "evaluation" means that in the specific case of the collective:

• - a driving profile, eg short / long distance, gradient / level, etc.
• - the specific wear of a component or the engine
• - specific consumption, eg engine oil or fuel

is derived.

1 shows an exemplary procedure of the off-board assessment of a load collective described in the previous section.

Preliminary

It be first assumed that the Data compression with the obvious, simple "shortening of the word length with laps ".

Example 16 bits → 8 bits: Cut off LSBytes and rounds

• 384 (dec) = 1 80 (hex) becomes 2, error 33.3%,
• from 2688 (dec.) = A 80 (hex.) becomes 11, error 4.8%.

From this It can be seen that the so resulting errors are very large can, this being the larger, the smaller the numbers are.

The Stress collectives zij will now be different according to (1) Weights Gij rated. This is the resulting total error then especially high, especially if the small (especially faulty) Occupation numbers zij be assessed with relatively large weights.

moreover very small occupation numbers are completely suppressed (im Example: Numbers <128).

• • selten auftretende "Überdreher" (Drehzahl Kurbelwelle), ev. kombiniert mit kaltem Motor
• • selten auftretende Überhitzungen des Motors (Kühlwassertemperatur)
• • nur vereinzelt auftretende, extreme Belastungszustände mit hoher Drehzahl und
• • gleichzeitig hoher Last.

However, small population numbers (frequencies) are important for engine diagnostics in that certain, rare events can cause high wear / consumption, as the following examples show:

• • seldom occurring "overspeed" (crankshaft speed), possibly combined with cold engine
• • seldom overheating of the engine (cooling water temperature)
• • only occasional, extreme load conditions with high speed and
• • at the same time high load.

from that High valuation weights result especially for the rare events.

Therefore separates the described, simple compression method "shortening of the word length with rounds "for load collectives out!!

conditions to a suitable method

• – unabhängig vom absoluten Wert der Größe ein relativer Fehler entsteht, der stets kleiner oder gleich einer definierten Schranke C ist. Der durch Kompression entstehende, relative Fehler soll also gleichmäßig auf der Werteskala verteilt sein
• – sehr kleine Werte der Belastungskollektive 100 % fehlerfrei erhalten bleiben.

Due to the described conditions, the following requirements are imposed on a suitable compression method:
In general, a k-bit wide (large) memory word is to be converted into an m-bit wide (smaller) memory word, so that:

• - Irrespective of the absolute value of the variable, a relative error arises that is always less than or equal to a defined barrier C. The relative error resulting from compression should thus be distributed evenly on the value scale
• - very small values of the load collective remain 100% error free.

Approach to a compression function

Searched is an initial one continuous function f (x) which maps a "k-bit number" x into an "m-bit number" f (x) such that the relative Error by rounding up or down the f (x) values to integers arises, for all the values to be mapped x are equal.

In 2 this becomes clearer:

approach

Let C be a small, constant factor, then let a change of the input quantity x = x ₁ by the amount Δx Δx = C x 1 (2) always cause a change of 0.5 at the "output", regardless of the size of x ₁ .: 0.5 stands for the rounding error on integers that occurs when quantizing f (x). f (x 1 + Δx) - f (x 1 ) = 0.5 (3)

Now, the following approximation is used at the point x = x ₁ :

und daraus nach beidseitiger Integration:

From (2), (3), (4) follows:

and from this after bilateral integration:

x = Eingangsgröße (zu komprimierende Werte)
k = Wortbreite der Eingangsgröße x
m = Wortbreite der (komprimierten) Ausgangsgröße, m < k
C = Fehlerfaktor, angepaßt an m und k (vgl. hierzu Kap. 3.5)The boundary condition is: f (x Max ) = y Max So f (2 k - 1) = 2 m - 1 so that finally results in the desired compression function.

x = input value (values to be compressed)
k = word width of the input variable x
m = word width of the (compressed) output variable, m <k
C = error factor, adapted to m and k (see chapter 3.5)

y = komprimierte Werte = Eingangswerte für die Dekompression (Wortbreite m)From (5) follows the corresponding inverse function for decompression:

y = compressed values = input values for the decompression (word width m)

Criteria for C, linearization

in the next Step are the criteria to find, as the function f (x) according to (5) usefully over the still open constant C is parameterized.

in the Sense of a possible small compression error, C should be as small as possible.

However, this is limited by the fact that, if the values of C are too small, small numbers x can no longer be compressed, as in 3 (Case 1) is shown graphically.

Case 1, too small C (C = 0.01):

The "passage point" through the abscissa is big Numerical values, so that underlying, small Values that - like mentioned - of importance are not compressible. The example shows the passage point at x = 399.6, i. Numbers between 0 and 399 are completely suppressed.

Case 2, too big C (C = 0.03):

Much of the y-value is given away unnecessarily. So in the example above ( 5 Case 2), the numbers x = 0 ... 22 are mapped into the numerical range y = 0 ... 122, ie here there is no compression but an expansion for small x-values. Only from x0 = 154 onwards is compressed, since from now on only f (x) <x.

As the second example shows, there is then a "compression point" x ₀ with: f (x 0 ) = x 0 (7) and f (x) <x for all x> x ₀ ,
if the error factor C is sufficiently large.

On the assumption that this point exists, the one in chapter 3.3. Requested property that small x values must not be suppressed, realized by using the identity function for all numbers x below x ₀ : fidet (X) = X (8)

With (5) and (7), therefore, the "point criterion" for x ₀ and C is:

Illustratively, Eq. (9) represents the intersection of the first bisecting line with the logarithmic function (5). Accordingly, if C ia is sufficiently large, there are 2 real solutions for x ₀ (intersection) which, in the limiting case, fuse into one solution (touch).

there the desired compression point corresponds to the greater value of both, since only from this point a reduction of the x values, ie compression, can occur.

Specifying C and x ₀ requires another condition. It results from the suitable definition of the slope at the interface x ₀ . The following section is used for this purpose.

Influence of quantization in decompression

So far By the construction of the function f (x), only the quantization effect became considered during compression.

For the suitable Parameterization (C) as well as for however, the overall assessment must be Similarly, the quantization of decompression be considered.

According to the derivation in Chap. 3.4, f (x) is constructed so that there is an increase of 0.5 with an increase in the x values of Δx = C x 1 ((2)) which corresponds to the "permissible", absolute error.

4 shows the influence of quantization during decompression (eg: rounding up from Δx = 1.5 to 2).

Out 4 follows that by quantizing the x-values at the points of the function where Δx = 0.5, 1.5, 2.5, ..., etc., an additional rounding error of 0.5 is formed, so that is the total abs.error = C · x 1 + 0.5 can occur. This represents the worst case, since the coincidence of maximum rounding of f (x) and of x (by 0.5 each) is purely coincidental.

The relative error at these points is Eq. (2):

Out (10) it follows that the "allowable" error C is around the Factor 2, 1.33, 1.2, 1.143, etc ... locally (or near) the places with Δx = 0.5, 1.5, 2.5, ... can increase.

At on the other hand, where Δx is integer, the relative error according to (2) is always maximum C.

These Results are summarized in the following worst-case table.

Tabelle 1: Worst-case-Werte des relativen Fehlers nach Kompression und Dekompression

Table 1: Worst-case values of relative error after compression and decompression

At this point will be attached to the comments the last section, in which a suitable second "connection criterion" for the interface x0 was searched.

definiert.For this purpose, taking into account the above table:

Are defined.

In order to becomes the part of the function f (x) in which the high error of 2 C can occur, hidden and by the identity function almost covered. The possible increase the relative error by a factor of 1.33 in the further course of the Function is consciously accepted with the definition (11).

(9) and (11) together:

(12) represents the searched (nonlinear) constraint equation for x ₀ , which must be solved iteratively, where k and m are given.

Of the associated Error factor C follows afterwards from (11), whereby the parameterization of f (x) is fixed.

The Results of this calculation are based on a few examples below Section together, where word lengths of 16 bits or 32 bits for the uncompressed raw values were accepted.

Results

The evaluation of the Glg. (12) and (11) show:

5 shows a schematic representation of the total compression function (k = 16, m = 8).

und die Dekompressionsfunktion (zur Offboard-Auswertung):

In summary, the compression function (for onboard editing) is:

and the decompression function (for offboard evaluation):

The next two 6 and 7 show the real relative error for the first 256 values of 16 bit → 8 bit and 16 bit → 6 bit compression.

For this the raw values (x-values) were first calculated according to Eq. (5) compressed and then reconstructed via the decompression equation (6), where the numbers were rounded to whole values both times.

Mostly, the band of the allowable error of ± C is well maintained. Exceptions are the places where - as explained - the quantization in the decompression noticeable and where a slight overshoot is noted. The resulting maximum values are well below the worst-case prediction according to the table, 7 (1.33 C, 1.2 C, etc.).

at the 16 bit → 6 Bit compression (C = 0.0833) is additional in the further course sporadically a slight exceeding of the permissible error recognizable by the fact that Eq. (4) (replacement of functional change represented by the differential quotient) is only an approximation. The maximum overrun due this effect is 4.8%, so that the relative error at these points max. 1,048 0.0833 = 0.0873.

Using conversion function call

For the analytical calculation in the microprocessor, Eq. (5) from Section 3.7 appropriately reshaped so that as little as possible arithmetic operations at runtime are required. You get like this:

The corresponding program piece in the language C is thus.

in this connection to take into account is that the required math and floating point libraries are. Furthermore, a must certain computing time, due to the 1n function call, to be accepted.

• • Rechenzeit ca 965 μs, entsprechend 2411 Rechenzyklen = 4822 NOP-Zyklen Speicherplatz (ROM-Programmspeicher) 0.74 kByte, bei Bedarf eventuell zusätzlich für die Floating-Point-Bibliothek 2.23 kByte

For the example 80C166 / 20 MHz in 1994 at the company Tasking Software GmbH, Leonberg, a rough memory and computation time estimate was carried out with the following results:

• • Calculation time approx. 965 μs, corresponding to 2411 calculation cycles = 4822 NOP cycles Memory (ROM-Pro gram memory) 0.74 kByte, if necessary additionally for the floating-point library 2.23 kByte

conversion with a table

• – keine Floating-Point-Bibliothek, keine 1n-Funktion erforderlich
• – Rechenzeit geringer

An alternative to analytic transformation is conversion to a table, which offers the following advantages:

• - no floating-point library, no 1n function required
• - Calculation time lower

These However, benefits are higher program overhead compared.

In order to the table that is in the program memory does not become too large, Table conversion is usually only up to a maximum of m = 8 (ie 16 bits → 8 Bit) makes sense.

Tabelle 2: Kompressionstabelle 16 Bit → 8 Bit The following is an example of the table required for 16 bit → 8 bit compression:

Table 2: Compression table 16 bits → 8 bits

Generally is to any value × 16 = [0, 65535] the table value x that is less than or equal to × 16. The corresponding table value n is the searched 8-bit value.

Example for Table 2:

Of the Numeric value x = 4200 is to be converted into 8 bits. According to table applies: 4200> 4153, from which follows n = 178.

Algorithm:

"Continued halving of the search range", ie 8 comparisons are necessary for the 16 bit → 8 bit table. The numbers n = 0, ..., 255 need not be stored explicitly, they can be regarded as a relative address and thus determined "implicitly". The table, 10 , thus occupies a program memory of 512 bytes in this case.

Generally have to the table points are chosen be that the relative errors immediately before and immediately after this equals is great i.e. the table point must be the same distance Δx (Error distance) to the two nearest - integer ones f (x) values corresponding to x values Xn-1 and Xn have:

8th shows a graphical representation for determining the position of the table points Tn. The table points Tn must therefore lie in the middle between Xn-1 and Xn, which in turn are calculated from the inverse function (6): T n = INT {0.5 · (Xn + X n-1 )} (14) with X n = INT {65535 * exp * (2 * C * [n-255])} (15)

Then linearization in the starting area: T n = n for n = 0, 1, 2, .... 56 (16)
T _n = n. Table value of the 16-bit input values
Xn = nearest compression point (x-value) above Tn
Xn-1 = nearest compression point (x-value) below Tn
n = 1, 2, ...., 255 = 8-bit output values
C = error factor = 0.0178
INT = integer function with rounds

Error consideration of compressed class sizes

in the The following is a brief description of how a compression according to Chap. 3.3 affects the rating of a composite classification.

The Occupation numbers of a composite classification are compressed, i.e. they have a relative rounding error independent of the absolute value on.

The rating B is according to Eq. (1) = (A1):

In this B is the overall rating, Gij are the rating weights, zij the meter readings of Cells i, j.

in the following are the Klassierfelder now numbered from 1 to n such that theirs Weights Gi tend to decrease with increasing index i, so that is true Gi ≥ Gk, if i <k, G1 So it's the biggest weight.

The evaluation deviation ΔB due to small changes Δzi of the frequencies zi is (total deviation):

Since, as stated, the maximum error C relatively occurs, are (worst-case): Δ zi = ± C · z i with 1 ≤ i ≤ n (A3)

After determination of the individual differentials from (A1) and subsequent relativization of the expression ckes (A2) on the error-free evaluation (A1) is finally obtained with (A3) after a longer intermediate calculation as the result of the estimation:

from that It can be seen that the relative valuation error for all imaginable weights Gi and occupation numbers are always smaller or equal to the value 2 C, since in (A4) the denominator is always greater than the counter is.

• • es gibt nur zwei Klassierzellen 1 und 2, die mit stark unterschiedlichen Gewichten G1 und G2 bewertet werden (Bsp.: G1 = 100 G2)
• • die Klassenbesetzung z1 ist viel kleiner als z2 (Bsp.: z1 = 0.091 z2)
• • z1 wird aufgerundet, z2 wird abgerundet.

wird der Wert 2C nahezu erreicht. (hier im Beispiel wird 2C zu 81.8 % erreicht)From the more detailed analysis of Eq. (A4) the following, absolutely worst case can be constructed. Under the assumption:

• • there are only two classifying cells 1 and 2, which are evaluated with very different weights G1 and G2 (example: G1 = 100 G2)
• • the class occupation z1 is much smaller than z2 (eg: z1 = 0.091 z2)
• • z1 is rounded up, z2 is rounded off.

the value 2C is almost reached. (here in the example 2C is reached to 81.8%)

Under slight cheaper Assumptions (weight distribution somewhat more balanced, e.g., G1 = 10 G2) the error already halved and reached only for the least favorable occupation numbers the value of about C.

• • mehrere Klassierzellen, Besetzungszahlen auf mehrere Felder (mehr als 2) verteilt
• • nicht absolut extreme Bewertungsunterschiede
• • statistisch unregelmäßig verteilte Rundungsfehler

liegt der Fehler weit unter C.For the conditions that are generally applicable in real life:

• • several classification cells, occupation numbers distributed over several fields (more than 2)
• • not absolutely extreme valuation differences
• • statistically irregular rounding errors

the error is far below C.

• • für alle denkbaren Bewertungs-Kennfelder
• • für alle möglichen Besetzungszahlen (Fahrweise und Fahrumgebung !)

immer unter dem Wert 2C bzw. unter Zugrundelegung von realen Annahmen unter dem Wert C bleibt.In summary, it follows from the fact mentioned in chapter 3.3 that the error resulting from compression is determined by the off-board evaluation

• • for all conceivable rating maps
• • for all possible numbers of occupants (driving style and driving environment!)

always remains below the value 2C or below the value C on the basis of real assumptions.

• • C = 0.365 → bei der 16 Bit – 10 Bit-Kompression
• • C = 1.78 → bei der 16 Bit – 8 Bit-Kompression
• • C = 8.33 → bei der 16 Bit – 6 Bit-Kompression

Assuming different real assumptions, the error is (generally far) below the value C. Numerical examples (derivation in Section 3.5):

• • C = 0.365 → for 16-bit - 10-bit compression
• • C = 1.78 → for 16-bit 8-bit compression
• • C = 8.33 → for 16-bit - 6-bit compression

Claims (14)

Diagnostic method for condition detection of internal combustion engines by means of compressed load collectives, in particular of motor vehicles, in the measured quantities and Recorded events of the operation by means of load collectives and be stored, - the load collective of a Data reduction is performed, in which a first k bit wide Data format with a mapping function into a second m bit wide Data format is converted, - and where the second data format has fewer bits than the first data format. The method of claim 1, wherein the mapping function for data reduction a linear function branch and a logarithmic Function branch has. Method according to claim 1 or 2, wherein the relative Error (Δx / x) through the data reduction over the whole value range of the data to be compressed smaller or smaller is equal to a predetermined upper bound. Method according to Claim 2, in which the linear functional branch the data unchanged leaves. Method according to Claim 2 or 4, in which the linear function branch for the treatment of data, de absolute value is less than or equal to a predetermined upper limit, is used, and in which the logarithmic function branch for the treatment of data whose absolute value is above the predetermined limit, is used. The method of claim 5, wherein the interface the two functional branches by the compression point, in which a data reduction is used is formed. Method according to one of claims 1 to 6, wherein the data compression is calculated by function call. Method according to one of claims 1 to 6, wherein the data compression determined with a compression table. Method according to claim 1, wherein the load collective Be stored on-board and evaluated off-board. Method according to one of claims 1 to 9, wherein the load collective On-board stored separately after at least 2 time ranges become. Method according to one of claims 1 to 10, wherein the load collective stored separately after 4 time ranges. The method of claim 11, wherein the four time ranges be demarcated by the mileage of the engine and in which: - the first Time range The last completed mileage up to 15000 km includes, - of the second time range the last completed mileage between 15000 km and 30,000 km, - the third time range the most completed mileage between 30000 km and 45000 km includes, - of the fourth time range covers the completed total mileage. The method of claim 12, wherein the second and third time ranges are stored compressed and the first time range is stored uncompressed. Method according to one of claims 12 or 13, wherein for the fourth Time range only basic quantities stored become.