 |
Claims  |
|
|
We claim:
1. A method for generating a diagnosis of a vehicle's cause of failure of
an emissions test, wherein data is obtained relating to a plurality of
vehicles which had failing emissions tests results and for which a
prediction report was generated and wherein each of said plurality of
vehicles has particular vehicle characteristics and also has passing
emissions test results and a repair report, comprising the steps of:
in a storage media, storing a classifier table composed of a set of rules;
in an emissions testing facility, receiving and storing vehicle
characteristics signals;
in said emissions testing facility, sampling emissions of said vehicle to
create emission test results to generate a series of emission test
signals;
transmitting to a processor said emissions test signals;
transmitting to said processor said vehicle characteristics signal; and
comparing said emission test signals and said vehicle characteristics
signals to said classifier table's set of rules in a manner in which forms
said diagnoses of said vehicle's cause of failure and generating a
prediction report thereof;
receiving said vehicle characteristics signals, said failing emission test
signals and said passing emission test signals of said plurality of
vehicles;
dividing said vehicle characteristics signals and said failing and passing
emission test signals of said plurality of vehicles to generate a training
dataset and a testing dataset; and
correlating said plurality of repair reports with said training dataset to
form new rules; and
from said new rules, forming new classifiers.
2. A method as recited in claim 1 further comprising the step of matching
said prediction report with a failure frequency distribution file to
generate a diagnostic assessment report including failure probabilities.
3. A method as recited in claim 2 further comprising the step of
transmitting said diagnostic assessment report to a host computer where
said diagnostic assessment report is stored.
4. A method as recited in claim 1 further comprising the step of
periodically updating said classifier table.
5. A method as recited in claim 1 wherein prior to said dividing step, said
method further comprising the steps of:
filtering said vehicle characteristics signals and failing and passing
emission test signals of said plurality of vehicles to remove certain
data;
formatting said vehicle characteristics signals and failing and passing
emission test signals of said plurality of vehicles; and
weighting said vehicle characteristics signals and failing and passing
emission test signals of said plurality of vehicles.
6. A method as recited in claim 1 further comprising the steps of:
processing said testing dataset and said classifier table to form first
output;
processing new classifiers to form second output;
comparing first output with said second output to generate a set of updated
classifier rules;
forming an updated classifier table from said updated classifier rules.
7. A system for generating a diagnosis of a vehicle's cause of failure of
an emissions test, said vehicle having vehicle characteristics which form
input to said system in the form of a vehicle characteristic signal,
wherein data is obtained relating to a plurality of vehicles which had
failing emissions test results and for which a prediction report was
generated and wherein each of said plurality of vehicles has passing
emissions test results and a repair report, comprising of:
stored in a storage media, a classifier table composed of a set of rules;
in an emissions testing facility, a receiving component for receiving said
vehicle characteristics signal;
in said emissions testing facility, an emissions sampling apparatus to
sample the emissions of said vehicle to create emission test results and
which generates a series of emission test signals;
a transmitter for transmitting to a processor said emissions test signals;
a transmitter for transmitting to said processor said vehicle
characteristics signal;
a comparator for comparing said emission test signals and said vehicle
characteristics signal to said classifier table's set of rules in a manner
in which forms said diagnosis of said vehicle's cause of failure and
generating a prediction report thereof;
a receiver which receives said vehicle characteristics signals, said
failing emission test signals and said passing emission test signals of
said plurality of vehicles;
a division component which divides said vehicle characteristics signals and
said failing and passing emission test signals of said plurality of
vehicles to generate a training dataset and a testing dataset; and
a correlation component which correlates said plurality of repair reports
with said training dataset to form new rules and therefrom, new
classifiers.
8. A system as recited in claim 7 wherein said system also stores a
frequency distribution file, said system further comprising:
a matching component which matches said prediction report with said failure
frequency distribution file to generate a diagnostic assessment report
including failure probabilities.
9. A system as recited in claim 8 further comprising a transmitter for
transmitting said diagnostic assessment report to a host computer where it
is stored.
10. A system as recited in claim 7 wherein said classifier table is
periodically updated.
11. A system as recited in claim 7 further comprising:
a filter which filters said vehicle characteristics signals and failing and
passing emission test signals of said plurality of vehicles so that said
signals meet selection criteria;
a formatting component which formats said vehicle characteristics signals
and failing and passing emission test signals of said plurality of
vehicles to be acceptable to said correlation component; and
a weight component which weights said vehicle characteristics signals and
failing and passing emission test signals of said plurality of vehicles so
that each of signal is appropriated scaled.
12. A system as recited in claim 7 further comprising:
a processor which processes said testing dataset and said classifier table
to form first output;
a processor which processes new classifiers to form second output; and
a comparator which compares first output with said second output to
generate a set of updated classifier rules to form an updated classifier
table.
13. A method for improving a classifier table which correlates vehicle
emission test failures with diagnostic assessments of the causes of the
failure, said method including utilizing data derived from a plurality of
vehicles which had failing emissions tests results and whose owners
received a prediction report and wherein each of said plurality of
vehicles has particular vehicle characteristics and also has passing
emissions test results and a repair report, said method further comprising
the steps of:
receiving said vehicle characteristics signals, said failing emission test
signals and said passing emission test signals of said plurality of
vehicles;
dividing said vehicle characteristics signals and said failing and passing
emission test signals of said plurality of vehicles to generate a training
dataset and a testing dataset; and
correlating said plurality of repair reports with said training dataset to
form new rules;
from said new rules, forming new classifiers.
14. A method as recited in claim 13 wherein prior to said dividing step,
said method further comprising the steps of:
filtering said vehicle characteristics signals and failing and passing
emission test signals of said plurality of vehicles for errors;
formatting said vehicle characteristics signals and failing and passing
emission test signals of said plurality of vehicles to be acceptable input
for said correlating step; and
weighting said vehicle characteristics signals and failing and passing
emission test signals of said plurality of vehicles so that each signal is
appropriately scaled.
15. A method as recited in claim 13 further comprising the steps of:
processing said testing dataset and said classifier table to form first
output;
processing new classifiers to form second output;
comparing first output with said second output to generate a set of updated
classifier rules;
forming an updated classifier table.
16. A method as recited in claim 13 further comprising steps for generating
a diagnosis of a vehicle's cause of failure of an emissions test:
in a storage media, storing said classifier table composed of a set of
rules;
in an emissions testing facility, receiving and storing vehicle
characteristics signals;
in said emissions testing facility, sampling emissions of said vehicle to
create emission test results to generate a series of emission test
signals;
transmitting to a processor said emissions test signals;
transmitting to said processor said vehicle characteristics signal; and
comparing said emission test signals and said vehicle characteristics
signals to said classifier table's set of rules in a manner in which forms
said diagnosis of said vehicle's cause of failure and generating a
prediction report thereof.
17. A method as recited in claim 16 further comprising the step of matching
said prediction report with a failure frequency distribution file to
generate a diagnostic assessment report including failure probabilities.
18. A method as recited in claim 16 further comprising the step of
transmitting said diagnostic assessment report to a host computer where it
is stored. |
|
 |
|
 |
Claims |
|
|
|
Description |
|
|
FIELD OF THE PRESENT INVENTION
This invention relates to automobile emissions testing and more
particularly to a system and method for predicting the cause of an
automobile's failure of an emissions test.
BACKGROUND OF THE INVENTION
In geographical locations having poor air quality, the United States
federal government has mandated vehicle emission inspection and
maintenance (I/M) programs in an effort to enforce emission limit laws on
automobile owners. The objective of these programs is to identify vehicles
whose emissions controls systems no longer perform acceptably and require
those vehicles to receive the necessary repairs and/or maintenance. The
owner of a car which is within the allowable limits is presented with a
certificate of compliance. However, an owner of a car which is not within
the allowable limits must repair the automobile so that its emissions are
within the allowable levels.
Because of the federal mandate, approximately 34 million vehicles are
tested annually. However, nearly 8.1 million fail the test and must be
repaired. It is estimated that $975 million dollars are spent in parts and
service sales in repairing vehicles to bring them into compliance with
federal emission standards.
A vehicle owner presented with a non-compliance report typically will
engage an automobile repair service provider to bring the vehicle into
compliance. However, because of the number of different types of vehicles
and models, it often difficult for an independent repair service provider
to reliably determine the cause of failure. For example, in one state
inspection program a sample of 10,450 initial inspection failures lead to
4,400 re-inspection failures, such indicating a forty-two (42%) failure to
repair. The retest failure of forty-two percent 42% of 8.1 million failed
vehicles is equivalent to 3.4 million that must be repaired further and
tested a third time or deemed eligible for a waiver if the repair costs of
that particular vehicle exceeded statutory limits.
The cost to vehicle owners for unsuccessful repairs as well as to the air
quality for continued excessive emissions is very high. Moreover, even in
automobiles which are able to pass, oftentimes their reported emission
measurements are close to the limits allowable by law and thus could
benefit from lowering. It would be beneficial for the vehicle's regular
service technician to service the car in a manner which he or she
knowingly could improve emission levels in such a case. Thus, it would be
beneficial if a testing facility were able to provide an analysis of
causes of emissions that are either close to or over legal limits at the
same time the vehicle owner is presented with a emissions test report.
In some states, hot-lines exist for automobile repair service providers to
call for help in diagnosing test failure results. Experts talk with
service providers to brainstorm a solution to the emission problem.
However, with each vehicle, there are many variables to consider,
including multiple emissions category failures. Therefore, it is desirable
to systematically diagnose failure to provide a relatively reliable and
accurate prediction of the type of repair which would bring the vehicle
into compliance with emission laws. Moreover, it would be beneficial to
provide the automobile owner with a prediction prior to bringing the
vehicle to a repair technician for evaluation.
SUMMARY OF THE INVENTION
This invention includes the preparation of a diagnostic report with a
diagnostic assessment for a vehicle owner to use in repairing his or her
vehicle to bring its emissions into compliance with emission standards.
The diagnostic assessment gives the vehicle owner's service technician
probabilistic information about the likely causes of the vehicle's failure
of the emissions test. The diagnosis is derived from operations involving
a classifier table which stores previously derived rules which form the
basis for the prediction of the diagnostic assessment. If a vehicle which
previously failed the emission test finally passes, information relating
to the passing test is used to update the classifier table.
More particularly, a classifier of the classifier table is the data
structure used in the automobile emission testing inspection lane by the
lane diagnostic subsystem, which runs on the lane controller computer, to
provide a diagnosis for a particular vehicle. It allows a quick evaluation
of the likelihood that a vehicle suffers from a given failure based on the
values of characteristics such as its emissions test results and the
vehicle's description.
The classifier predictions are then used to prepare a failure report that
is given to the motorist for use by his or her repair technician. The
diagnosis reached by the system will be uploaded and stored on a central
database server computer for purposes of reporting, correlation with
actual repair, and inclusion in the knowledge base.
In another feature of this invention, the classifiers are continuously
updated in a learning process based on new repair records. The learning
process periodically analyzes the repair data and updates the knowledge
base to include new or revised classifiers. The learning process will
explore, identify and predict failures that correlate with parameter such
as the following: vehicle make and model year; vehicle milage;
on-board-diagnostics (OBD) data; emissions composite values; and emissions
second-by-second values.
The learning process can be described in terms of its inputs, outputs and
functions. The inputs to the learning process utility are suitably
prepared data from the following: vehicle test records; vehicle emissions
repair records; and diagnostic records. The outputs to the learning
process utility are for example: new classifiers; learning process log
entry; administrative report; and a pattern report. The general functions
of the learning process are to describe the data, determine patterns of
significance, and create a classification data structure (classifier) and
mechanisms for applying the classifier in a predictive mode. The
predictive accuracy of the classifier is evaluated periodically using a
dataset representative of current program vehicles. The classifier is
updated as needed to maintain or improve accuracy.
BRIEF DESCRIPTION OF THE DRAWINGS
FIG. 1 depicts hardware elements of the this invention;
FIGS. 2a-2f are graphs of emissions and purge test results;
FIGS. 3a-3d show an example of a prediction report;
FIG. 4a is a diagram of the method of this invention;
FIG. 4b is a legend of the diagram of FIG. 4a;
FIG. 5 is a systematic diagram of the learning process feature of this
invention;
FIG. 6 is an example of a CHAID or similar algorithm tree output; and
FIG. 7 is a systematic diagram of the lane controller with the diagnostic
assessment generation.
DETAILED DESCRIPTION OF THE INVENTION
This description is broken down into three distinct sections. The first
section describes the emissions testing process in general with reference
to the initial diagnostic assessment feature of the this invention. The
second section describes the learning process feature of the this
invention which uses among other things, data emissions testing
information generated from retests of previously failed vehicles to update
the classifiers used to make initial diagnostic assessment. The third
section ties the elements of the first section and the second section
together with reference to the interaction of this invention with the
operations of an inspection lane.
As an introduction, the following is a discussion about the features of an
emissions inspection system. Generally, any of a number vehicle emissions
testing regimes (or procedures) can be used and this invention is not
limited to any one of them. Examples include, two-speed idle, loaded
steady-state, ASM 50-15, ASM 25-25, ASM 2, and I/M 240. I/M 240 will be
used as an example of to illustrate aspects of the data collection feature
invention. As mentioned above, the I/M refers to inspection maintenance.
The 240 of I/M 240 refers to the 240 seconds in which data is collected.
Other emissions test systems are equally applicable. FIG. 1 shows a system
which is used to test for certain emissions during the I/M 240 test. The
emission analysis system 10 uses a tube 22 to collect exhaust from the
tailpipe of automobile 12 to test for HC, CO, CO.sub.2, and NO.sub.x, read
by analyzers 13, 14, 16 and 17 (or what ever emissions collection is
desired). Emission analysis system 10 also includes other typical features
such as flowmeters 18, calibration gases 19, an exhaust pipe to the roof
21.
As is true with all test regimes, the I/M 240 emissions analyzer system is
controlled by a software/hardware combination and is in communication with
the lane controller processor 26. During the I/M 240 driving cycle, the
I/M 240 emissions analyzer system transmits mass emissions data to a
processor at a once-per-second rate. Each grams-per-second reading
time-stamped and transmitted to the processor, which calculates the
resultant second-by-second grams-per-mile results. Each grams-per-second
results also includes a status byte that flags systems failures,
out-of-range conditions, and communications errors so that the processor
can be signalled to take immediate action.
The processor 26 performs all of the described functions and is usually in
communication with a central data base server host processor 28. Usually,
the testing facility runs several test lanes. In other situations, the
test facility operates a single emission analysis system 10. Each lane is
equipped with a processor 26 which supports the execution of controller
software 27 which manages the activities in the lane including the storage
of the emissions data. Then, using the proper weighting factors, it
calculates the total values, which are compared to the appropriate I/M 240
(or other test regimes) exhaust standards to determine pass or fail.
The processor 26 generates a failure report 31 indicating that the vehicle
has exceeded the legal limit of one or more chemical emissions. Turning to
FIG. 2, illustrates the contents of the failure report 31, that is the raw
data generated from the emission test, of FIG. 1 to give the vehicle owner
information about the levels of emissions with respect to the allowable
limits set by law. The failure report is delivered in any format, for
example, written or electronic.
In the system of the this invention, the processor 26, provides a
diagnostic assessment 32. In a first situation, the diagnostic assessment
is provided in the event of a failure of a vehicle to pass emissions test.
At the start of the vehicle inspection where a vehicle is being retested,
vehicle inspection personnel either enters repair data 46 to processor 26
or it is scanned in and up to the host data base server 28 for input using
various other means. Since the vehicle has failed a previous test, that is
this current test is a retest, the performed repair data 46 is surrendered
and is entered at an appropriate time.
A repair data form scanning system that completely automates the task of
reading and evaluation the information collected from vehicle repair
reports is preferred. In a situation where the inspection personnel enters
the data the console display 36 provides prompts and messages to the
inspector and permits entry of responses and data. Preferably, data
entered into the system is thoroughly checked for errors before being
accepted.
Controller software 27 causes the emissions data or similar data shown in
FIGS. 2a-2f to be formatted in a manner so that it can be compared to the
classifiers stored in classifier table 41. Comparator 42 runs an algorithm
so that processor 26 generates diagnostic assessment 32 for an individual
vehicle.
The algorithm to evaluate each vehicle using the classifier table is
preferably computationally economical. The classifier is a set of data
structures--one for each failure to be diagnosed. In one embodiment of
this invention, each failure diagnosed is independent of the others since
there may be multiple failures for a single vehicle.
Each data structure is a series of rules that can be applied to the vehicle
population in the form of "if." Each vehicle has one and only one
applicable rule per data structure. The algorithm then, for each data
structure (or failure), compares the vehicle's parameters with those in
the first rule. If the parameter values don't match, the algorithm goes to
the next rule. As soon as a matching rule is found, the probability that
corresponds with the parameters is provided and the data structure is
exited. Thus, through parsing, the failure analysis feature of this
invention matches emissions result to the repair diagnosis (thus providing
real-time analysis as opposed to batch-calculations).
The classifier table is a data structure used in the knowledge-based system
and is made up of rules that can be applied to make predictions. The rules
represent leaf nodes of a decision tree. Methods of induction of decision
trees from suitable empirical data have been studied by artificial
intelligence and statistical researchers since the late 1970's. The tree
generation is provided by a commercially available program such as
KnowledgeSEEKER(.TM.) by Angoss Software which uses a CHAID or a Chi.sup.2
Automatic Interaction Detection algorithm or by a variant of ID3 which was
devised by J. Quinlan, published in "Machine Learning," 1986. Tree
generation output, which is one element of the update process, will be
discussed in detail below. The preparation of the raw data into input to
CHAID is uniquely determined by an initial analysis in the detailed design
and implementation phase and is described in the second section of this
detailed description of this invention.
The output file is modified to form the classifier as described in detail
below. The rule files are inputs to the classifier formatting module (see
FIGS. 4b and 3d below). In other words, a classifier is a rule file that
has been reformatted and optimized to be useable by the failure analysis
module and the classifier table is a collection of one or more
classifiers.
Once a repair diagnosis has been made, the diagnosis 32 may be ordered by
listing the most likely cause first or by associating a probability with
each one, depending on the source of the items and whether the probability
data is desired. For example, the learning process, which is discussed in
detail below, will identify problems that frequently occur. This allows
the probability to be calculated and included in the knowledge base and
diagnosis.
FIGS. 3a-d combined show an example of a repair strategy report providing
diagnostic assessment 32 (see FIG. 1) as output of the classifier table.
For a 1984 Nissan truck with an idle HC=629, idle CO=8.49 and idle O.sub.2
=9.7 two failure categories (ignition failure probability (FIG. 3a) and
air induction failure probability (FIG. 3c)) are generated using
characteristics of the vehicle and emissions data which satisfy the
classifier table.
Specifically, the air induction failure shown by the combination of FIGS.
3c and 3d are satisfied by rule 10 below. In processing processor 26
matches vehicle make and model year; vehicle mileage; emissions composite
values; and rules of the classifier table. The algorithm processes the
rules in the classifier table 41 to pull out predictors that match vehicle
and test data and associated failure probabilities as shown in FIGS. 3a-d.
These figures were demonstrated using data from a two-speed idle test.
Classifier format modules shown in FIGS. 3b and 3d identify predictors for
the failures probabilities shown in FIGS. 3a and 3c. The graphs show the
probability of a problem in the repair category and how this vehicle
compares with other failed vehicles for the repair categories. The rules
above create the classifiers which form the classifier format modules of
FIGS. 3b and 3d.
In FIG. 3b, the predictor categories of HC and CO at idle correlate with
ignition failure. A value of idle HC which is between 346 & 786 and the
idle CO which is greater than 1.27 indicates a slightly reduce probability
of ignition failure over the average vehicle.
In FIG. 3a the ignition failure probability is shown with regard to all
failing vehicles (47%) and with regard to this vehicle (45%). Similarly,
in FIG. 3d, three predictor categories are shown which present air
induction failure symptoms. In FIG. 3c the air induction failure
probability is shown with regard to all failed vehicles (17%) and with
regard to this vehicle (33%).
The repair categories most likely to be responsible for the failure of the
1984 Nissan are presented in order of descending probability.
Alternatively, predictor percentage ranges may be mapped to English
language descriptions, such as high, moderate and low. From FIG. 3c it can
be seen that there is an elevated likelihood that the 1984 Nissan truck
will have an air induction failure compared to that particular failure
with respect to all failed vehicles which is extremely useful information
for a repair technician in repairing the vehicle.
Below is a listing of potential failure categories and subcategories which
reflect groups of repair actions that exhibit similar symptoms. These are
subject to change in size and content depending on the learning process
performance discuss below. Subcategories are lowest level of information.
The level of information provided as a diagnostic assessment is dependent
upon the correlations which can be drawn during the learning process
discussed below. This is also constrained by the repair actions, the
lowest level of detail given on the vehicle emissions repair report. The
failure categories and the repair actions corresponding to each category
are for example:
fuel.sub.-- delivery
carburetor adjustment
speed adjustment
carburetor
choke
cold start
fuel filter
hoses
injector cleaning
injector(s)
inlet restrictor
pump
regulator
motor/valve/solenoid
tank
air injection
belt
check valve
control
pump
tubes
valves
ignition
cap/rotor
coil
distributor
initial timing
module
plugs
spark advance control
wires
egr
control system
passage/hose
sensor
valve
evaporation
carbon canister
control
filter
hoses
gas cap
purge valve
catalytic converter
converter
heat shield
preheat catalytic converter
air.sub.-- induction
air filter
ducts
sensor
thermostatic air door
throttle bore
oil change HI CO
could put in oil & coolant level
diluted oil
pcv
crankcase ventilation
hose
passage
valve
electronic.sub.-- control
air control
canister purge control
coolant sensor
ECM
EGR control
idle control
MAP sensor/switch
mass air flow sensor
mixture control
pressure sensor
PROM
RPM sensor/switch
spark control
temp sensor/switch
throttle position sensor/switch
vehicle speed sensor
O.sub.2 sensor
O.sub.2 sensor
exhaust
exhaust components
manifolds
vacuum.sub.-- leak
vacuum leak
engine.sub.-- mech
valve
valve timing
While each category listed above includes subcategories, these
subcategories can include subcategories of their own. As the classifier
table becomes more accurate, more subcategories can be addressed by the
rules independently as a category.
Standardization of repair information and consistency is preferable. To
provide consistency, where the repair technician is equipped with
appropriate computer hardware and software, diagnostic assessments are
presented electronically, and via dial-up phone line, for example, by
Internet data delivery. The diagnostic assessments are also provided on a
printed failure report at the inspection lane which the vehicle owner
presents to the repair technician.
As discussed above, the classifier table 41 has been previously built and
stored for access during processing by comparator 42. Accordingly, the
classifier table 41 provides the ability of this invention to "close the
loop" between the repair mechanic and the inspection system by providing
increasingly accurate diagnostic and repair statistics to increase the
success rate of the repair process, bringing more vehicles into compliance
under waiver limits.
Above, mainly the failure diagnostic feature of this invention has been
described. That is, this detailed description up to this point has been
directed to the explanation of the emissions testing process in general
with reference to the initial diagnostic assessments feature of the this
invention. Below, the learning process or update feature of the invention
is described in detail. Accordingly, the following section describes the
learning process feature of this invention which uses emissions testing
information generated from failed tests and passing retests to update the
predetermined criteria used to make initial diagnostic assessments. That
is, failed emissions test data is used, and passed retest data is used
only to validate that repairs performed were successful. The retest
emissions results might include information such as that found in FIGS.
2a-2f.
Looking at the overall process of this invention, including the update
feature is provided by FIGS. 4a and 4b where FIG. 4a shows the system and
FIG. 4b provides a legend for the path configurations. The vehicle 12
visits the inspection station 20 and receives a failure report with
diagnostic assessment 31. The vehicle visits the repair facility 25 and
receives repairs, such as those most likely including those suggested by
the failure report 31 as discussed in detail above. The repair facility 25
generates a repair report 46 and the inspection station 20 retests the
vehicle 12. That retest information 51 is sent to the host 28 along with
vehicle emissions repair reports 52 to be gathered as part of host
databases 53. The learning process 60 performs as described below and
updated classifier data files are transferred 61 to the inspection station
and processor 26.
By capturing information regarding repairs 46 performed on vehicles that
fail emissions inspections, and then retesting the vehicle by emission
analysis system 10, information is provided to processor 26 which is
collected and used to update classifier table during the learning process.
Performed repair data 46 is input to the host 28 so that it corresponds
unambiguously with the vehicle test results record 31 and diagnostic
assessments 32.
As mentioned above, the learning process may be initiated on the host or
other centralized apparatus. Alternatively, the learning process operates
in a client server mode with the learning process connected directly to
the host database tables and a client application running on the PC. In
this configuration, these functions would be implemented in a client
application and the output could be any file formate acceptable to a tree
building algorithm.
The user interface that initiates the learning process preferably requires
the following information from the user: data collection start date;
vehicle test type desired for programs that have multiple test regimes,
i.e. two-speed idle, I/M 240; and the value to be used for excluding
marginal failures, fail.sub.-- margin.
Suitable data are selected, files are assembled and written out to a file
for vehicle records meeting the learning process criteria. There are
several separate types of functions performed including: creating reports
that monitor the effectiveness of the learning process and the diagnostic
assessments issued; filtering vehicle records for learning; assembling a
data record in a temporary table for acceptable vehicles including
formatting and checking failed values; copying the contents of the
temporary table data to an input file for the learning process; creating
additional data files for use in the lane diagnostic subsystem.
Before actually discussing the learning process itself, the preliminary
reports are discussed in that they are generated through the process of
preparing the learning process data for the learning process operation.
Turning to FIG. 5, there is shown a systematic diagram of how the host
diagnostic system performs updates of the classifier table's knowledge
base. The update is an ongoing process of "learning": the statistical
module which receives new data 53 including actual repair data from
retested vehicles; and data from other testing programs in the form of
individual records (as discussed above); filtering for errors and
weighting the data 66 according to its value or ordering its application
so that more credible measures have a greater influence in forming the
diagnosis; formatting and compressing data 67 so that it is in a form
which can be correlated; correlating the actual repairs with the
predictors to create rules 76; compressing and concatenating the rules 69
to provide data structures for individual failures and provide compaction
of the data structure; testing the compacted classifiers to determine
accuracy 79; updating the knowledge base for distribution to all locations
where it resides. The frequency of the updates is adjustable. The
determination of which data to use and how to format it is nontrivial. In
one embodiment, the OBD data is included in the learning process. In a
different embodiment, the vehicle's OBD overrides some or all
probabalistic predictions.
Each element of the update feature as outlined above is now discussed in
more detail. Returning first the statistical module 53, the statistics
given here are descriptive in nature and are formatted and output in the
repair effectiveness report in the form of an administrative report 54.
The values are preferably computed for the data collection period input by
the user to cover the learning process. These vary by emissions testing
program and may include the following: number of failing vehicles broken
down by type of failure (standard failed) and test regime applied; number
of failure reports 31; description of OBD operations performed | | |
