DE102005027378B3 - Computer assisted diagnostic system, especially for vehicle, prioritizes test steps in workshop diagnosis - Google Patents

Abstract

The system uses a diagnostic program and generates possible fault quantities by interrogating the system and processing the responses. The test domains which belong to the fault candidates are derived from relevant test steps from a database. A decision function is used to evaluate the relevant test steps, e.g. on the basis of their information content, and to prioritize them.

Description

Workshop Diagnostics is an important element in the life cycle of a vehicle. The vehicle customer expects a fast and inexpensive Remedy in case of a defect. The service technician in the workshop must be supported in modern vehicles in this task. It is only possible with computer support, the large amount to process the resulting information and the test procedure efficiently to design. About that In addition, a diagnostic system must be real-time capable and manageable from a development perspective and be cared for. The diagnostic system according to the invention operates with a decision-making process based on entropy and cost assessments of exams, with the goal of the service technician at any time of a diagnostic session optimal tests to be able to propose. The test domain, the Relations between exams and to be tested Components can work with a rule-based knowledge base, a mathematical physical model of system diagnostics or modeled with Bayes networks become and executable Computer program to be compiled. The focus of this invention lies here with the determination of the error candidates and in the Construction of decision function for test step selection.

An example of a system diagnosis is in the German patent application DE 195 23 483 A1 disclosed. The characteristic of the system diagnosis is the mapping of the system to be diagnosed into at least one physical-mathematical model that can be implemented and processed with computer assistance. In the DE 195 23 483 A1 modeling involves a structural model and an impact model, often referred to as a behavioral model. The structure model depicts the physical relationships of the individual components of the technical system, and the behavioral model maps the functions of the individual components of the system. In a knowledge base, which is essentially a rule table from If / then conditions, which in turn can be mapped to data tuples, the diagnostic knowledge relevant for system diagnostics is stored. With the system diagnostics a fault detection and by recourse to the knowledge base a computer-aided troubleshooting can be carried out.

The System diagnostics has two major disadvantages. The modeling is for larger technical systems, such as. a motor vehicle extremely expensive, if all possible Causes of errors to be controlled by the system. Even more difficult In system diagnostics, ambiguous system states are to be handled when e.g. a recorded error code can have multiple causes in the absence of sufficient error environment data or insufficient information about the current system status, from the system diagnostics not further can be processed. The system diagnostics will then break at this point without diagnostic result from. Another disadvantage of system diagnostics is their principle Unsuitability for the processing of experience of the service technician. As well Customer information is limited to defective functions or to intact functions in the diagnostic process incorporated.

diagnostic systems of the aforementioned type have the further disadvantage that they are very quickly become very complex and the necessary modeling effort, Computing effort and calculation effort for larger technical systems exponential with the number of error possibilities of the individual components of the overall system increases. Also, for the diagnosis all possible exams into a static test step tree be imaged. In reality results in systems with several interdependent components a plethora of possibilities in which order individual partial tests of individual components carried out can be. With 5 components there are already theoretically 5 faculty different Test procedures, the all through a static test tree would have to be pictured. The efficiency of the diagnostic procedures therefore decreases with the number of possible errors drastically.

you therefore has more efficient options sought to build a diagnostic system.

One way to improve the diagnostic process is in the European patent specification EP 1 069 487 B1 described. In parallel with the repair progress, knowledge saved by a service technician at crucial points of the test step tree, known as evidence, can be queried and included in the diagnostic system. This eliminates the need to calculate all possible errors and test options. The diagnostic method can become more efficient the more queries are provided at the appropriate point of the diagnostic procedure on the system side. The input of the evident knowledge takes place via a user interface, which is formed by a display and an input menu.

It is desirable for troubleshooting with future diagnostic systems in garages to pay more attention to the cost-effectiveness of the tests to be performed. A fast and efficient test The procedure is thus a desirable requirement for these future diagnostic systems.

It has therefore, for example, in the European patent specification EP 0410632 B1 suggested that audit costs should already be taken into account when creating the knowledge base for the diagnostic system. It is suggested to proceed in two stages. By dividing the technical system to be diagnosed into a possible first decomposition with a first model-based rule set and carrying out the cost calculation of a check for this first decomposition. Then further possible decompositions of the system are calculated and for each decomposition a cost calculation is carried out for a review of the system. The most economical examination of the system should probably be the one whose decomposition yields the lowest cost. If the most favorable decomposition of the system is fixed, there is, at least in principle, a proposal for the review of the system, if you will, an inspection plan.

The However, the aforementioned system has from the point of view of this present Invention several serious disadvantages:

• - It have to always several possible Decompositions of the technical system are calculated and evaluated. This often does not make a fine-tuning of the technical system possible manageable, since here the decomposition possibilities and in particular the associated costs of the review very quickly very compute-intensive and the danger is that the calculations NP-hard, so longer last as the life waking of a human being is.
• - Succeed it is a preferred disassembly with the lowest testing costs nevertheless, the technician must always model the entire one Check system. It At no time is an attempt made to determine an error to identify or even to narrow down a component.
• - It only the statistical default probabilities of System components. Information about actually present Errors or malfunctions can not processed with the system.
• - The Design of the system assumes that the decomposition always take place like this can and also does that the individual components of the decomposition no influence on the error status of the connected ones Components. In other words, it is assumed that the decompositions regarding the Error condition of the system are equivalent. But this is one Assumption that applies only to a few systems, perhaps telephone networks. For most networked systems, failure causes a component also malfunction in the rest Components of the system. For systems that are not just statistical Consider failure probabilities, but also process detected error messages is the Approach of equivalent decompositions not compatible.
• - It stay completely unclear which systems are at all meaningfully decompose into different decompositions. In almost all, networked systems, there is only one actual networking, in which also the error must be found and it remains unclear to what extent a mathematically possible, deviant decomposition can provide a reliable statement here, which provides a traceable advantage or knowledge gain.

at all possible Dismantling is the fault finding and fault location left to the service technician and the service technician receives Also, no instructions on how to perform the troubleshooting most effective. cost estimates are not statements of effectiveness.

In the after published DE 10 2004 024 262 A1 was considered for the first time, not only to build a fault tree, but parallel to the fault tree and while the error hypotheses are processed, even with the diagnostic progress a dynamic test step tree to adapt to the progress of the diagnostic session. In that sense, that is after published Document a receipt for the diagnostic systems on which the invention is based. Indeed The novelty of the invention to be assessed hereby not be challenged because the characteristics of the license plate of claim 1 not in the post-published Document are included.

In an earlier, subsequently filed application of the applicant with the German Patent and Trademark Office with the official file reference DE 102005015664.9 an independent diagnostic system is described, which is further developed with the invention described herein. The disclosure of this earlier application is expressly incorporated by reference with its full scope in the disclosure of the invention described herein. This earlier application of the applicant mainly contains an interactively operating diagnostic program in which the service technician, within a troubleshooting space of the diagnostic program initially defined, identifies the components identified as possibly defective or Functions, a focus for further, automated troubleshooting can set by the diagnostic program. The focus can be set by restricting to an error code or by restricting to a function. After setting the focus, a limited focus amount is selected that matches the selected focus of possible error candidates. The individual error candidates hereby experience a weighting by offsetting different probabilities for the occurrence of an error code, for the probability of failure of a component or function and, if appropriate, for the presence of an error image.

In an advantageous embodiment of the older one Diagnostic system features the diagnostic system over the extra possibility To process error images. Error pictures are combinations here multiple error codes that are specific to the failure of a particular one Component and so a direct indication of the defective component can deliver. The error pictures can hereby a combination of active and inactive error codes be formed. Here you can the non-active error codes especially valuable hints to not deliver defective components and so the amount of possible Restrict error candidates.

In a further advantageous embodiment of the older diagnostic system can if the review of the Candidates in the focus set revealed that none of the examined Components was broken, a new focus by the service technician and thus a new focus set with weighted error candidates be generated.

In summary let yourself thus note that while there are fault location diagnostic systems, and that there are considerations to cost-optimized test procedures at Diagnostic systems without fault location exist. However, there are so far no diagnostic systems that support the service technician to the necessary checks the point of view of effectiveness perform.

It thus raises the task of the invention for a lot already weighted and identified error candidates, a test step sequence set up with the preselected candidate as possible can be effectively investigated.

The solution succeeds mainly with a diagnostic system, first through system queries and by processing the queried system messages or system states with generates a defect candidate quantity to a diagnostic program. The error candidates in this candidate error quantity are already in terms of error probability prioritized. This prioritization can be used when building a test step tree considered and helps in the respective faulty system state preferably efficient test step tree build. in principle However, the prioritization of the error candidates is not necessary. It can For example, all error candidates of the candidate defect quantity have the same error probabilities. Among the determined Error candidates are over the test domain in which the relations between being performed exams and error candidates to be verified is modeled for the review of Candidate candidates relevant tests determined by computer and by means of a program module for test step prioritization subjected to an evaluation. The evaluation is done with a Decision function, which the individual tests with regard to their information content for the current diagnostic problem in the form of the current candidate error quantity rated.

In an advantageous embodiment contains the decision function the decision function in addition to evaluate the information content of each exam yet a cost evaluation of the individual tests. advantageously, can the two assessments for a rating in terms of efficiency the one to be carried out exam be summarized. The efficiency of the test is then determined by the Quotient of information content and the cost of implementing the single exam.

In a preferred embodiment The decision function and thus the check step prioritization contains the decision function an assessment and selection of the test steps with regard to the still to be expected costs of further diagnosis, the so-called Expected Costs of Diagnosis.

In a further embodiment the test step prioritization according to the invention is the test domain by a Bayes network modeled.

In another embodiment of the test step prioritization according to the invention, the test domain ne modeled with a rule-based knowledge base.

In a further embodiment the test step prioritization according to the invention becomes the test domain with a modeled mathematically physical model of the technical system.

In a further advantageous embodiment of the test step prioritization according to the invention the clamping of the test step tree takes place automatic and dynamic. Automatic means that after the test step prioritization the ones to be carried out exams the service technician of the diagnostic system are displayed, without that the service technician takes the exams still would have to search out. Dynamic clamping basically means two aspects. On the other hand the test step prioritization takes place dynamically with the number of currently suspected error candidates and secondly dynamic with the number of already performed and proposed tests. For this purpose, the diagnostic system has a feedback option with which the service technician the result of the performed exams as evidence in the diagnostic system can enter. After entering a Evidence is updated by the diagnostic system, the error candidate quantity, if an error candidate was found, and the check step prioritization the already used exams will be canceled. Prüfschrittpriorisierung and diagnostic system can so dynamically track the progress of the diagnostic session. If only one single candidate is found, the search ends and a prioritization is unnecessary yourself.

In an advantageous embodiment the test step prioritization according to the invention, Starts the check step prioritization with a prioritized candidate error list generated by a diagnostic system determined during a diagnostic run as erroneously suspected components and prioritized for their error probabilities.

In a less preferred base stage of the inventive check step prioritization, the check step prioritization starts with a unweighted candidate list. This basic stage has the advantage that even commercially available diagnostic systems, such as those in connection with the prior art in 1 can be used for the determination of the error candidates.

in the Details of the invention will be described below with reference to graphical Representations closer explained. Showing:

1 A typical workshop diagnostic system of the prior art;

2 A block diagram of a diagnostic system, as described in the earlier German patent application DE 102005015664.9, and which is further developed here as a preferred diagnostic system with the Prüfschrittpriorisierung;

3 A schematic diagram of the test step prioritization according to the invention;

4 A basic method for the check step prioritization according to the invention;

5 A test step tree of the test step prioritization according to the invention;

In 1 a situation is schematically illustrated, as it is known today in motor vehicle workshops. For the diagnosis of a motor vehicle is a computer-aided diagnostic tester 1 via a standardized diagnostic interface 2 to the communication network 3 for the control units 4 connected in the motor vehicle. Known diagnostic testers are z. For example, the DAS system from DaimlerChrysler or the BMWDIS system. The control units installed in the motor vehicle 4 For example, they are in communication with each other via a data bus. A common data bus in motor vehicles here is the so-called CAN bus (for Control Area Network). Each of the installed control units in the motor vehicle has the ability to self-diagnose in addition to the communication interfaces. In the context of the self-diagnosis of the control units, errors identified in coded form as so-called error codes by the control unit software in specially reserved memory areas, so-called fault memory, are written with the aid of the diagnostic routine in the control units. In the schematic representation of 1 For example, these reserved, non-volatile memory areas are referred to as FS (for error memory). For the communication and for the data exchange between a diagnostic tester and the control units installed in the motor vehicle, a standard has been established, which is known under the name Keyword Protocol 2000 and whose specification and normalization in the ISO standard 14 230-3. With the control commands and data formats agreed upon in the Keyword Protocol 2000, it is possible to use the diagnostic interface to read out the codified contents of the fault memories of the individual control devices with the aid of the diagnostic tester and to transfer them to the computing system of the diagnostic tester. The standard for the keyword protocol 2000 comprises two different application options. On the one hand, the standard stipulates that the communication between the diagnostic tester and the control units is effected via a gateway 5, which is eg. B. the motor vehicle CAN bus to the diagnostic interface 2 binds, takes place or that, as was customary in the past, the fault memory of the control units via the so-called K and L lines and the standardized diagnostic interface 2 can be read out and stored directly in the diagnostic tester. In the schematic representation of 1 is the more modern form of access via a CAN bus and thus represented by a gateway. For the invention is of concern only that there is at least one way to be able to read the error memory of the control units with a diagnostic tester. In the diagnostic tester, the transmitted contents of the control device memory, in particular error codes and status data of the control devices, are further processed in a diagnostic session with an implemented diagnostic program.

The Diagnostic program also includes the option of a VDU workstation as a human-machine interface manually provide further information that is important for a diagnosis enter.

2 shows as a block diagram the most important program modules and realized with these program modules functions of a diagnostic system according to the invention. The individual program modules are integrated into a higher-level sequence control of the entire diagnostic system. This sequence control takes over the call of the individual program modules at the respectively necessary time. The diagnostic system processes error codes FC and inputs by a service technician as input variables. The service technician makes his input from a computer workstation 200 , which is typically equipped with a screen and a computer keyboard, each to the computer system 201 connected to the diagnostic system. About another interface 202 the computer system can be connected to the motor vehicle to be diagnosed. About the OBD socket (On Board Diagnosis), the control units contained in the motor vehicle can be addressed. It can be read out the error memory of the control units, the self-diagnosis routines of the control units can be started and thereby functional tests of the individual control units are started and it can be accessed and read current system state data from the motor vehicle. One possibility of technical realization was related to 1 discussed.

However, the processing of the system data and the course of the diagnostic program deviate significantly in the invention from the known prior art. The most important differences here are the interactive course of the diagnostic program and the associated possibility of forming targeted diagnostic focuses in which one or more focal points, referred to hereinafter as foci, can be set for troubleshooting, thus improving both the diagnostic quality and the duration of the diagnosis can. The program implementation will be described in more detail below:
The diagnostic program implemented on the computer system is characterized among other things by a modular structure. As a result, among other things, the programming and the configuration of the diagnostic system are structured. With a first program module 210 , According to its function called rule table evaluation, retrieved from the motor vehicle data, such as error codes and system status data for each component installed in the motor vehicle, read and processed. Further processing involves checking in a knowledge base 211 saved rule tables. The rule tables contain the diagnostic knowledge relevant for the technical system to be diagnosed. This knowledge is stored, for example, in compressed form in data tuples. The data tuples depict the relationships between the information contained in them. A data tuple is stored for each diagnostic rule. Each data tuple consists of a component identifier Ki, an error code FCi, a symptom sympi as an indication of an affected technical function, and a system status Stat. The rule table evaluation then takes place in such a way that all the stored data tuples are looked up in the entirety, which data tuples contain the error code (s) and which components Ki and functions Sympi are named in the identified data tuples and can therefore be affected by the observed error FCi. The component identifications found in this way are recorded and combined into a first quantity of error candidates and stored. These component identifiers Ki give an indication as to which component or also which function of the technical system can be responsible for the observed error code or for the observed error symptom. The result of this first scouring of the knowledge base is a first set of suspicious components that have been identified based on the identification via the error codes FCi.

In a further processing step or program module 213 becomes the one by the first candidate Crowded troubleshooting room continues to be tense. In a second pass through the knowledge base, the possible sources of error are now enhanced by the relevant functions that may be involved in the motor vehicle. For this purpose, the rule tables are searched again, this time not according to detected error codes, but according to the possibly already affected by the error codes components Ki. The components that are to be affected are the possibly affected Sympi functions. These two quantities do not have to be identical. Because it is possible that an error code refers to a component that is relevant for several functions. The result of this second pass through the knowledge base is a supplemented candidate list 214 , which now also contains possibly faulty functions in addition to the possibly faulty components.

Having arrived at this point, the possibility of interaction now begins in the further course of the diagnostic program. First, on the screen workstation 200 a query 215 performed and displayed whether the already determined error codes or the determined possibly affected functions should be displayed for further processing. In both cases, in a further process step 216 The service technician had the opportunity to set a focus for the further diagnostic procedure. Depending on the selected display, the focus is set by either selecting a displayed error code or a displayed, suspicious function Sympi by means of graphical menu control, and using it for further processing by the diagnostic program. If the focus is set, further data processing is restricted to this focus. This means that not all detected error codes, suspicious candidates or suspicious functions are considered, but only those that fall under the chosen focus.

For the components Ki which are still suspect within the focus, the individual error candidates are identified in a further program module or method step 217 subjected to a weighting.

For the weighting have to the probabilities of error codes FCi, the probability for the Occurrence of sympi and possibly the probability for the existence be calculated from defect images. This requires probabilities which indicate with what security a defective component or a candidate Ki an error code (P (FCi | Kj)), a malfunction (P (Sympi | Kj)) or an error image (P (FBi | Kj)). In addition, will requires the relative error weighting g (Kj) of a component itself. These Information will be for the calculation of the prioritized or weighted candidate list needed. The conditional probabilities are easy to estimate. Most of time they are set to "1." Uncertain symptoms or error codes can but sometimes accept values less than 1.

The Error weights g (Kj) can from experience, e.g. be chosen between one and a hundred and represent a relative failure characteristic. In a below explained advantageous embodiment can also over these error weights g (Kj) take into account the current field occurrence be by putting these weights over Offsetting of error frequencies be adapted.

All the probabilities and error weights are stored in the database when the diagnostic system is defined 218 stored. Appropriately, these can be filed and bedatet together with the component list, the feature or symptom list and the error image list. These lists are created in the construction of a motor vehicle and therefore need only be supplemented by the knowledge of experience with regard to the conditional probabilities and the error weights. Expediently, the data is model-specific. However, cross-product databases can also be created and used. In the case of cross-database databases, however, the possibility of a series-specific selection, eg in the form of an upstream master table, must then be maintained.

From the data explained above, the individual variables are calculated as follows:

The become three calculated sizes to prepare the candidate prioritization, below.

G is a normalization size and will in a preparation step or by means of e.g. as the sum of all Weights g (Kj) determined.

After the probabilities of the error codes P (FCi), the error images P (FBi) and the malfunctions P (Sympi) have been calculated and the focus quantity of the error candidates is determined, the calculation of a prioritized or weighted candidate list can be performed 219 and finally on the screen of the workstation 200 be issued.

The a posteriori error probability or priority Prio (Ki) of a component Ki results from the following product:

In which the following applies: P (FCi | K) = P (FCi) if the error code FCi of one component K independently is analogous to P (Sympk | Ki) = P (Sympk) and P (FBl | Ki) = P (FBl) at independence from the respective component.

These priority is still not standardized and can alternatively be normalized by divided by the sum of all candidate priorities.

The explained above Diagnostic system, the disclosure of which is described here Invention belongs provides a prioritized candidate list, and is the preferred one Diagnostic system with which the test step prioritization carried out can be.

in principle can the test step prioritization according to the invention however, with each diagnostic system, the diagnostic result is an amount of candidate error supplies, performed become.

The final step in the workshop diagnosis is component testing or functional testing, with the goal of complete fault isolation for the quickest possible remedy. Until this time, all available information, such as vehicle diagnoses or customer complaints will be evaluated largely automatically. The last step, the testing on the vehicle, but is usually costly and time-consuming. The aim is therefore to make these vehicle checks cost-optimal by vehicle tests in the form of an optimized test tree or decision tree are proposed to the service technician of his workshop diagnosis system. The focus of the inventive test step prioritization described here lies in the creation of an optimized test sequence. The test step prioritization 220 , assumes an error candidate quantity whose error candidates have already been identified and named in a preliminary stage. This situation is illustrated by the block diagram of 3 , Off error codes 310 and identified malfunction or trouble symptoms 320 From a diagnostic system in a basic version, a non-prioritized quantity of error candidates 330 determined. In preferred embodiments, a diagnostic system becomes a weighted candidate list 340 determined. In any case, the software module 350 the test step prioritization 220 to the identified candidate candidates in a knowledge base or in a database 360 filed, relevant exams, to verify the individual suspects.

Once all the relevant tests have been determined, the individual tests are evaluated, prioritized and placed in a test step sequence by the test step prioritization with regard to information content and / or costs. The prioritization of the test steps can hereby be tracked recursively and dynamically to the progress of the diagnostic session by the already entered tests being entered into the system by the service technician and the used tests being deleted from the test step sequence. The consumption of a proposed check usually leads to a new prioritization of a weighted check step list 370 , which is currently tracked to the status of the diagnostic session.

For a vehicle usually many hundreds of tests are defined and e.g. in a database. Thankfully, during a diagnostic session not all exams but only those relevant to the current problem. After evaluation of all vehicle and customer information is of the Diagnostic system in a first stage a list of suspects Components that creates defect candidate set. Well, from the diagnostics program selects those checks from a test step database, the for the review of suspect Components are relevant. That those exams that have at least one Reference to the suspects Have components. The amount of all stored in the database exams be PS. The relevant tests be ps element of PS. The fundamental key question for the investigation the best exam at the present time is according to the invention: how well does the test ps in the current diagnostic problem, which consists of an error candidate set persists? According to the invention for this purpose a decision-making function that can be processed by a diagnostic program proposed that the relevant tests in terms of their respective Information content for evaluates the particular diagnostic problem or the relevant one exams for the respective diagnostic problem regarding the respective costs for performing the exam rated, or the for the respective diagnostic problem relevant tests in terms of efficiency evaluated, with efficiency by the quotient of information content and costs are incurred.

The information content of a test step can be determined, for example, from its entropy Ent. The efficiency Eff is then the ratio of the information content of a test step ps to its cost: Eff (ps): = Ent (ps) / costs (ps)

wobei

Zi

die möglichen Prüfungsausgänge des Prüfungsschrittes ps, z.B. in Ordnung oder nicht in Ordnung; und

p(ps = zi)

die Wahrscheinlichkeit für das Prüfungsergebnis zi sind.

In this case, the entropy of a test step is understood as the following probability sum:

in which

Zi

the possible test exits of the test step ps, for example in order or not in order; and

p (ps = zi)

the probability of the exam result are zi.

In order to the information content of a test step is greatest when its entropy becomes greatest. The entropy is greatest when a test step many exits, i.e. many possible test results has and if the individual test results one possible have uncertain outcome. This corresponds to the fact that I am with that exam gain the most information, the same time as many Error candidate checked and least of all I know how the test runs out. Conversely, that exam has the least amount of information, which has few outputs and of which I know the result with great certainty already.

In a preferred embodiment of the check step prioritization, the decision function for test step selection is formed by the expected cost of the test steps remaining after the selected check step for the respective remainder of the prospective diagnostic session. This replaces the decision functions of the two previously discussed decision functions in terms of entropy and efficiency with a refined and extended decision function, which is referred to here as Expected Costs of Diagnosis (ECD). This decision function also includes the entropy criterion, which, however, is complemented by a refined and extended cost consideration and cost estimate. The refined and expanded cost estimate calculates the probable costs for the remaining test for each current test step, ie what costs are still to be expected, after which a previous test has already been carried out. The designation of the decision function as ECD (Extended Cost of Diagnosis) is based on the term ECR, the Extended Cost of Repair, which is described in the wis scientific literature is occasionally used.

wobei j ein Laufindex und n die Anzahl der nach dem aktuellen Prüfschritt ps noch verbleibenden Prüfungen ist.The preferred decision function is formed:

where j is a run index and n is the number of tests remaining after the current test step ps.

mit p°= p(Diag = Ki/ps = exit j)For each relevant test ps is summed over their possible test exits. The summand consists of a cost estimate in the square brackets EstCosts, which is multiplied by the probability of occurrence of the relevant test output P (PS = exit j). In the cost estimate, the estimated additional costs are added to the cost value of the current test ps. These costs are estimated as follows:

with p ° = p (Diag = Ki / ps = exit j)

In order to exist the estimated Costs EstCosts from the entropy of the remaining diagnostic problem, multiplied by the average testing cost AvCosts. The latter is the arithmetic mean of the cost of all remaining exams.

Of the each pending, to be selected Test step will then be out of the possible test steps selected, in each case that test step selected which is the expected cost of diagnosis of the remaining one Diagnostic session minimized.

The in the decision functions required probabilities p (ps) and the derived quantities of the relevant ones test steps ps are calculated from the prioritization of the defect candidate list. Among the identified error candidates Ki are the relevant test steps determined.

The test step ps simultaneously checks two candidates K1 and K2, the test step having the result "not OK" (niO) if one of the two candidates proves to be faulty and has the result "OK" (OK) if both Candidates deliver the test result without errors. The probable outcome of the relevant test steps then results from the error candidate probabilities as follows: p (ps = no) = (p (K1 = ok) + p (K2 = niO)) / normalization p (ps = ok) = (p (K1 = 10) + p (K2 = 10)) / normalization = 1 - p (ps = no) wherein the error probabilities p (Ki) for the candidates Ki are taken from the weighted candidate error list.

One another criterion for a test step prioritization are the costs for a test step to be performed. These costs are known as empirical values, e.g. as so-called labor values and are used in the creation of the test step database and the individual test steps assigned. This allows the decision function for the check step prioritization possibly from entropy to efficiency selection or a cost estimate Refined according to the Expected Costs of Diagnosis.

The Decision function for the test step prioritization The diagnostic system is then either a Prüfschrittpriorisierung using the information content of the individual relevant test step, the test step with the largest information content the highest priority gets or it is with the decision function, a prioritization of individual test steps performed on the basis of the efficiency defined above, wherein the test step the highest priority gets the highest Efficiency has, or it is a cost estimate according to the expected Costs of Diagnosis used.

With each of the aforementioned test step prioritizations, it is possible to establish a test step sequence that is automated by computer calculations and computer selection. The service technician can a Prüfschrittreihenfolge proposed and ge on a display of a diagnostic system for display which starts with the highest prioritized check step and follows the further check steps, each with decreasing prioritization.

This can be a computer-based diagnostic system set up, the diagnostic program is extended by the module of a Prüfschrittpriorisierung. The procedure for a basic system with test step prioritization is shown in the block diagram of FIG 4 shown. Starting with the weighted candidate list provided by the diagnostic system 410 be of the inventively completed diagnostic program in a subsequent process step 420 determines the relevant test steps and, for example, evaluates and prioritizes them with the aforementioned evaluation scheme with regard to efficiency, information content or cost estimation. In a subsequent process step 430 For example, the service technician or the installer can select a test step from the displayed test steps, for example by menu control, and perform this on the motor vehicle. In the following process step 440 the result of the test step carried out is entered into the diagnostic system and with the newly gained knowledge a recalculation of the weights of the suspect candidates is performed. The recalculation leads in particular to the deletion of all error candidates already tested and found to be in order from the list of error candidates. The released weights grow in this case the not yet tested error candidate. With the newly compiled error candidate list, the test step prioritization is also recalculated and performed, so that the procedure for the check step prioritization is recursive, with the aim as quickly as possible to obtain an error candidate with a probability of error close to 100%. Clarified in the process step 450 of the block diagram.

One Problem with test step prioritization can the more accurate determination of the information content of a test step be, if one this information content on the currently present Wants to apply a diagnostic session. The above-described equal distribution the exam exits is here only a first approximation, but at closer Viewing may vary with the current diagnostic problem. Especially can already performed exams and the knowledge gained from it the information content and the Value not yet performed exams increase or diminish. Do you want the information content of a test step to adapt to the progress of the diagnostic session you need this a flexible calculation tool that is capable of probabilities deal with it. Here likelihood networks and in particular offer so-called Bayes networks. Bayes networks have been around since the 80's successful in the fields of fault diagnosis, decision support, Reliability analysis, Risk assessment, etc. used. The Bayes networks serve here representation of contexts and sizes that Uncertainties may be subject or only probability are known after. About that In addition, the so-called propagation methods offer the possibility new information, such as Exam results, consistent and thereby calculate the impact on other sizes. The Bayes network now serves to the relations between examination and tested Components and the probabilities needed for the check step prioritizations to calculate more accurately and adapt to the progress of the diagnostic session.

In order to let yourself a method for test tree production set up. Starting with a weighted candidate list a diagnostic node is set up, which has a state for each suspicious component a probability, the weighting of the error candidate corresponds, contains. Well, you can those exams from the test step database selected which are relevant, i. checking the suspect components. The chosen test steps are prioritized and selected with one of the decision functions. One Test step is still available if it has not already been done, or the exam exit not sure yet. For the test tree construction have to then both starting possibilities The examination, Component in order or component out of order, present be. Then it will be for all relevant not yet used exams recursively continued. Such a method is called myoptic or short-sighted, there always only the next Step is optimized. Is the test tree spanned can the Results of the performed test steps as evidence thus entered as secure knowledge in the Bayes network the effect on the remaining components is calculated be and to the left remained components to be spanned again a test tree, the in each case the highest prioritized with the myoptic method described above test steps selects.

The result is finally a test step tree, which is graphically displayed to the service technician. A graphic representation of such a check step tree is in 5 shown. In the diagnostic node, the states of the determined error candidates are in the form of weights. These are the candidates K1, K2, K3, K4, K5. With the decision function, the best test step PS1 is selected for the present diagnostic problem. Depending on the outcome of the exam, the exam result is reflected back. The test result PS1 = NOK (test result Not OK) identifies the error candidate K1. at the test result PS1 = OK (candidate K1 is OK) the next best test is selected with the decision function. This is the test step PS4 for checking the component K4. For the result PS4 = NOK, component K4 is identified as faulty. If the check with test step PS4 reveals that component K4 is OK, PS2 continues with the next best test step, which checks component K2 and identifies it at test output Not OK K2. If the test result is OK, it is continued with the best remaining test step. In the example chosen, this is the test step PS3 for checking the component K3. Here, the test result is not so important, because in any case, the last remaining, suspect component K5 must be examined. For this last review is usually still a test due, but in 5 is no longer shown. It may also be that a restart of the diagnostic system after the performed repairs of the components K1 and K4 shows that thus the error candidates K2, K3, K5 are no longer suspected.

realistic testing problems can around 50 suspects Components include. For Each component has at least one individual test. Furthermore are there still functional examinations, covering more than one component. This is the number of exams generally over the number of components. By evaluating efficiency, the branches of a test step balanced according to their entropy and cost. During one Diagnostic session is growing the error probability and thus the weighting in the candidate error quantity on a few mostly one component. The error probabilities in fact, all of the already tested components will become too Zero and flows as a test result into the diagnostic program so that as the diagnostic session progresses and continuously updating the weighted candidate error quantity by recursive recalculation by the diagnosis program the diagnostic focus getting sharper becomes.

Functional examinations usually check many candidates, some of which are outside the candidate error quantity. These co-examined external candidates must be treated separately when calculating the information content of a test step and the service technician must be given a corresponding note. If the external candidates are tested as faulty by carrying out the check, this can lead to inconsistencies with the candidate defect quantity. An update of the test step tree or of the candidate defect quantity after a positive test with external candidates must therefore not lead to complete relief of the candidates in the candidate defect quantity. This can be prevented, for example, by the fact that in the error probability between external p _external and internal portion p is distinguished _internally , and in the return mirroring of the test result in the diagnostic system only that proportion is zero, which was also tested. The untested portion then remains both in the candidate error quantity and in the test step prioritization. For functional exams, which also check external candidates, the probability is therefore the sum of external and internal probability. p '= p external + p internally

Of the Information content or the efficiency of these composite tests is usually very high, as they examine several components and therefore a big one Provide information. The composite tests are therefore among the most highly prized test steps for one Fault candidates.

Of the completeness half is still on the possibility dependent test steps pointed. Some test steps build on each other and have to be performed in a specific order. Of course, these dependent tests must also as such in the test step database be structured.

Also if with the diagnostic system according to the invention mainly both the error candidate set and the check step prioritization dynamic be tracked to the progress of the diagnostic session and both change the error candidate quantity as well as the test step selection, so can anyway at any time of the diagnostic session, in particular at the customer meeting in the vehicle acceptance, a current cost estimate for each remaining diagnostic session are determined. For this purpose, the error candidate quantity in each case the complete test step tree determined and summed up the costs of the determined test steps. These Cost sum then gives an upper limit to the costs with which at the most possible with the present diagnostic problem will be expected.

The service technician is offered a complete test step tree by the diagnostic system in which the individual test steps with regard to efficiency, information content or expected costs of diagnosis are prioritized. Ultimately, it is up to the service technician to decide in which order he should perform which test step. The service technician does not have to start with the highest prioritized check of the test step tree.

With An optional protocol feature can be used with the diagnostic system the performed Repairs be logged. About this protocol can then exams or test steps be suggested that make sure the built-in components or the repair can be rechecked to to make sure that the repair was successful and no new mistakes were incorporated. Moreover, with the logger function the current status of the vehicle in the repair workshop is continuously logged become. This can be used to determine which components of the motor vehicle are just removed. The dynamic costs hang the diagnostic session This also depends on the current condition of the vehicle, so that the protocol function also for the cost estimate can be used with. The tracking of the shoring condition of the Vehicle in the repair succeeds here over the Mitprotokollieren which test steps with which exam exit already done were.

Claims (14)

Computer-aided diagnosis system, in particular for a motor vehicle, which generates a candidate error quantity with a diagnostic program by means of system queries and by processing the queried system messages or system states, characterized in that the test domain associated with the error candidates is determined from relevant test steps from a test step database and with a decision function the relevant Test steps are evaluated and a test step prioritization is performed. computerized Diagnostic system according to claim 1, characterized in that the Decision function the relevant test steps with regard to their Information content rated. computerized Diagnostic system according to claim 1, characterized in that the Decision function, the test steps to be selected in each case after the remaining after selecting the current test step estimated costs of the remaining test steps. computerized Diagnostic system according to claim 1, characterized in that the Decision function the relevant test steps regarding efficiency rated. computerized Diagnostic system according to one of claims 1 to 4, characterized that the checking domain is through a Bayes network is modeled. computerized Diagnostic system according to one of claims 1 to 4, characterized that the check domain with a rule-based knowledge base is modeled. computerized Diagnostic system according to one of claims 1 to 4, characterized that the check domain with a mathematical-physical model is modeled. computerized Diagnostic system according to one of claims 1 to 7, characterized that with the check step prioritization a test step tree automatically and dynamically clamped and displayed. computerized Diagnostic system according to one of claims 1 to 8, characterized that the error candidate set contains a lot of weighted error candidates is. computerized Diagnostic system according to one of claims 1 to 8, characterized that the error candidate set has a lot of unweighted candidate error is. computerized Diagnostic system according to one of claims 1 to 10, characterized in that with a log function the repair measures carried out be logged and about this Protocol checks for the conducted repairs be proposed. computerized Diagnostic system according to one of claims 1 to 11, characterized in that the test step tree be dynamically adjusted to the progress of the diagnostic session can. Computer-aided diagnosis system according to one of claims 1 to 12, characterized that composite tests that test with a test of several potential candidate candidates are possible. computerized Diagnostic system according to claim 11, characterized in that a Tracking performed will, with which the current condition of the vehicle in the repair is detected.