`Breed
`
`US005809437A
`[11] Patent Number:
`[45] Date of Patent:
`
`5,809,437
`Sep. 15, 1998
`
`[54] ON BOARD VEHICLE DIAGNOSTIC
`MODULE USING PATTERN RECOGNITION
`
`[75] Inventor: gaIvid S. Breed, Boonton Township,
`
`[73] Assignee: Automotive Technologies
`International, Inc., Denville, NJ.
`
`[21] Appl' No_; 476,077
`_
`[22] Flledl
`
`Jun- 7, 1995
`
`[51] Int. Cl.6 ......................... .. G01M 17/00; G06F 11/00
`[52] US. Cl. ............................... .. 701/29; 701/34; 701/35;
`701/45; 364/55101; 73/1173
`[58] Field of Search ....................... .. 364/42403, 424.04,
`364/424~05> 431~11> 551~01> 508; 340/439>
`459> 521; 395/20> 21> 23> 913> 905; 73/116>
`1173; 701/29> 35> 40> 44> 45> 34> 39> 43
`_
`References Clted
`Us PATENT DOCUMENTS
`
`[56]
`
`4,128,005 12/1978 Arnston et a1. ...................... .. 73/1173
`4,418,388 11/1983 Allgof et a?
`" 364/43101
`4’817’418
`4/1989 Asaml et a‘ '
`"" " 73/1181
`5,041,976
`8/1991 Marko et al.
`.. 364/42403
`5 313 407 5/1994 Tieman et a1‘ n
`364/508
`5j325j0s2
`6/1994 Rodriguez .......... ..
`340/438
`5,333,240
`7/1994 Matsumoto et a1. .
`395/23
`5,400,018
`3/1995 Scholl et a1. ..................... .. 340/82554
`
`5,406,502
`5,442,553
`5,481,906
`
`4/1995 Haramaty et a1. ............... .. 364/55101
`8/1995 Parrillo ............. ..
`364/42404
`1/1996 Nagayoshi et a1. ..................... .. 73/116
`OTHER PUBLICATIONS
`
`Liubakka et al., “Failure Detection Algorithms Applied to
`Control System Designs for Improved Diagnositcs and
`Reliability”, SAE Technical Paper Series, Jan. 29 to Mar. 4,
`1988, pp. 1—7.
`James et al., “Microprocessor Based Data Acquistion for
`Analysis of Engine Performace”, SAE Technical Paper
`Series, Feb. 23—27, 1987, pp. 1—9.
`
`Primary Examiner_Tan Q Nguyen
`
`ABSTRACT
`[57]
`A Component diagnostic System for a motor Vehicle having
`at least one component Which emits a signal having a pattern
`containing information as to Whether that component is
`operating normally or abnormally. The system includes at
`least one sensor Which senses the signal and outputs an
`electrical signal representative thereof and corresponding to
`the pattern, a processor coupled to the sensor(s) for process
`ing the electrical signal and determining if the pattern is
`Characteristic of abnormal State of Operation of the
`component, and an output device coupled to the processor
`for affecting a another system Within the vehicle if the
`.
`.
`component is operating abnormally. The processor prefer
`ably is a Pattern recognition System
`
`20 Claims, 4 Drawing Sheets
`
`160 [/i
`
`110% 115
`
`8
`3%
`
`\J 152
`
`\Abrorion Sensor
`
`101
`
`180
`
`/—1%
`
`Component
`
`f es
`105 JV!
`
`151
`
`103
`
`_ 120
`
`9 (l)
`
`w
`
`13°
`
`“3 Kg 30
`
`Se sor
`
`.
`
`.
`
`Drognos’rrc
`Module
`
`.
`
`170
`
`Voltage or
`Current Sensor
`
`AVS EXHIBIT 2001
`Toyota Inc. v. American Vehicular Sciences LLC
`IPR2013-00417
`
`
`
`U.S. Patent
`
`Sep. 15, 1998
`
`Sheet 1 of4
`
`5,809,437
`
`(MW
`W
`
`o2
`
`2:
`
`2
`
`0:
`
`10s as
`
`EmcoaEoo
`
`
`
`68% c9655
`
`OWH
`
`o:
`
`6 mooto> l
`
`
`
`62% E950
`
`v .9
`n
`
`E295
`@582
`
`6w ow
`
`c2
`
`m2
`
`c2
`
`
`
`
`
`
`
`U.S. Patent
`US. Patent
`
`Sep. 15, 1998
`Sep. 15, 1998
`
`Sheet 3 of4
`Sheet 3 0f 4
`
`5,809,437
`5,809,437
`
`mm
`
`mm
`
`o: wfi
`
`min
`m, @m
`
`E
`
`N
`
`@N MN / a /
`mm
`
`llg! 3
`
`_ 0
`
`o:
`
`a
`HN
`/ 2 E
`
`¢ 8
`
`0
`.2 4.
`‘>4
`
`0 2 2
`2
`
`I .
`
`a
`
`'I(
`
`mm 2
`NM om E
`
`
`
`U.S. Patent
`
`Sep. 15, 1998
`
`Sheet 4 of4
`
`5,809,437
`
`1 Manson
`
`-———\
`
`z MICIOI‘IIONI
`
`-_€
`
`3 000mm- -* 220 Gammon
`
`4 omnlasunlmson -——€
`
`5 omummson -—€
`
`5 mnowmrrm -—§
`
`'1 VOLMTII. IE
`
`' m '—_\
`
`9 numnmrmson '_*
`
`10 mm KNOCK mson ' —?
`
`n ommmm
`
`1: monnnrosn'lorzmom—_—)
`
`13 mmc'mnovlml '__J
`
`14 wmLsrmslNm I
`
`15 IAcuomm
`
`I
`
`16 srunoMIm
`
`'___l
`
`11 OIIIYGIINBIWFOR
`
`I
`
`1; manouml '—J
`
`u CLOCK
`
`J
`
`:0 ODOMI-m
`
`'_J
`
`n mus-mammal‘;
`
`n mtnmnsnzsol '——J
`
`n mums:
`
`:4 cm'mllomm
`
`I
`
`'
`
`_ 1 70
`
`DIAGNOSTIC
`MODULE
`
`210
`
`DISPLAY
`
`25
`
`383:
`
`2
`
`ILDLVL
`‘MWBIINSOR lg
`
`OOOLANTLIVILEINSOR '*
`
`TRANBFLUID Wm
`
`“WM '-_’
`
`OOOLANT mm
`
`Fig. 4
`
`
`
`1
`ON BOARD VEHICLE DIAGNOSTIC
`MODULE USING PATTERN RECOGNITION
`
`BACKGROUND OF THE INVENTION
`Every automobile driver fears that his or her vehicle will
`breakdown at some unfortunate time, e.g., when he or she is
`traveling at night, during rush hour, or on a long trip away
`from home. To help alleviate that fear, certain luxury auto
`mobile manufacturers provide roadside service in the event
`of a breakdown. Nevertheless, the vehicle driver must still
`be able to get to a telephone to call for service. It is also a
`fact that many people purchase a new automobile out of fear
`of a breakdown with their current vehicle. This invention is
`primarily concerned with preventing breakdowns and with
`minimiZing maintenance costs by predicting component
`failure which would lead to such a breakdown before it
`occurs.
`When a vehicle component begins to fail, the repair cost
`is frequently minimal if the impending failure of the com
`ponent is caught early but increases as the repair is delayed.
`Sometimes if a component in need of repair is not caught in
`a timely manner, the component, and particularly the
`impending failure thereof, can cause other components of
`the vehicle to deteriorate. One example is where the water
`pump fails gradually until the vehicle over heats and blows
`a head gasket. It is desirable, therefore, to determine that a
`vehicle component is about to fail as early as possible so as
`to minimiZe the probability of a breakdown and the resulting
`repair costs.
`There are various gages on an automobile which alert the
`driver to various vehicle problems. For example, if the oil
`pressure drops below some predetermined level, the driver
`is warned to stop his vehicle immediately. Similarly, if the
`coolant temperature exceeds some predetermined value, the
`driver is also warned to take immediate corrective action. In
`these cases, the warning often comes too late as most vehicle
`gages alert the driver after he or she can comfortably solve
`the problem. Thus, what is needed is a component failure
`warning system which alerts the driver to the impending
`failure of a component long before the problem gets to a
`catastrophic point.
`Some astute drivers can sense changes in the performance
`of their vehicle and correctly diagnose that a problem with
`a component is about to occur. Other drivers can sense that
`their vehicle is performing differently but they don’t know
`why or when a component will fail or how serious that
`failure will be. The invention disclosed herein will, in most
`cases, solve this problem by predicting component failures
`in time to permit maintenance and thus prevent vehicle
`breakdowns.
`Presently, automobile sensors in use are based on speci?c
`predetermined levels, such as the coolant temperature or oil
`pressure, whereby an increase above the set level or a
`decrease below the set level will activate the sensor, rather
`than being based on changes in this level over time. The rate
`at which coolant heats up, for example, can be an important
`clue that some component in the cooling system is about to
`fail. There are no systems currently on automobiles to
`monitor the numerous vehicle components over time and to
`compare component performance with normal performance.
`Nowhere in the vehicle is the vibration signal of a normally
`operating front wheel stored, for example, or for that matter,
`any normal signal from any other vehicle component.
`Additionally, there is no system currently existing on a
`vehicle to look for erratic behavior of a vehicle component
`and to warn the driver or the dealer that a component is
`misbehaving and is therefore likely to fail in the very near
`future.
`
`10
`
`15
`
`25
`
`35
`
`45
`
`55
`
`65
`
`5,809,437
`
`2
`OBJECTS AND SUMMARY OF THE
`INVENTION
`Accordingly, it is an object of the present invention to
`solve the above problems by monitoring the patterns of
`signals emitted from the vehicle components and, through
`the use of pattern recognition technology, forecasting com
`ponent failures before they occur. Vehicle component behav
`ior is monitored over time in contrast to currently used
`systems which merely wait until a serious condition occurs.
`It is another object of the present invention to provide a
`new and improved on-board vehicle diagnostic module
`utiliZing pattern recognition technologies which are trained
`to differentiate normal from abnormal component behavior.
`In this manner, the problems discussed above, as well as
`many others, are alleviated by the vehicle diagnostic module
`described in the paragraphs below.
`The diagnostic module in accordance with the invention
`utiliZes information which already exists in signals emanat
`ing from various vehicle components along with sensors
`which sense these signals and, using pattern recognition
`techniques, compares these signals with patterns character
`istic of normal and abnormal component performance to
`predict component failure earlier than would otherwise
`occur if the diagnostic module was not utiliZed. If fully
`implemented, this invention is a total diagnostic system of
`the vehicle. In most implementations, the module is attached
`to the vehicle and electrically connected to the vehicle data
`bus where it analyZes data appearing on the bus to diagnose
`components of the vehicle.
`Principal objects and advantages of this invention are
`thus:
`1. To prevent vehicle breakdowns.
`2. To alert the driver of the vehicle that a component of the
`vehicle is functioning differently than normal and might be
`in danger of failing.
`3. To alert the dealer, or other repair facility, that a
`component of the vehicle is functioning differently than
`normal and is in danger of failing.
`4. To provide an early warning of a potential component
`failure and to thereby minimiZe the cost of repairing or
`replacing the component.
`5. To provide a device which will capture available
`information from signals emanating from vehicle compo
`nents for a variety of uses such as current and future vehicle
`diagnostic purposes.
`6. To provide a device which uses information from
`existing sensors for new purposes thereby increasing the
`value of existing sensors and, in some cases, eliminating the
`need for sensors which provide redundant information.
`7. To provide a device which is trained to recogniZe
`deterioration in the performance of a vehicle component
`based on information in signals emanating from the com
`ponent.
`8. To provide a device which analyZes vibrations from
`various vehicle components which are transmitted through
`the vehicle structure and sensed by existing vibration sen
`sors such as vehicular crash sensors used with airbag sys
`tems.
`9. To provide a device which provides information to the
`vehicle manufacturer of the events leading to a component
`failure.
`10. To apply pattern recognition techniques based on
`training to diagnosing potential vehicle component failures.
`Other objects and advantages of the present invention will
`become apparent from the following description of the
`
`
`
`5,809,437
`
`3
`preferred embodiments taken in conjunction With the
`accompanying drawings.
`
`BRIEF DESCRIPTION OF THE DRAWINGS
`
`The following draWings are illustrative of embodiments
`of the invention and are not meant to limit the scope of the
`invention as encompassed by the claims.
`FIG. 1 is a schematic illustration of a generaliZed com
`ponent With several signals being emitted and transmitted
`along a variety of paths, sensed by a variety of sensors and
`analyZed by the diagnostic module in accordance With the
`invention.
`FIG. 2 is a schematic of one pattern recognition method
`ology knoWn as a neural netWork.
`FIG. 3 is a schematic of a vehicle With several compo
`nents and several sensors and a total vehicle diagnostic
`system in accordance With the invention utiliZing a diag
`nostic module in accordance With the invention.
`FIG. 4 is a ?oW diagram of information ?oWing from
`various sensors onto the vehicle data bus and thereby into
`the diagnostic module in accordance With the invention With
`outputs to a display for notifying the driver, and to the
`vehicle cellular phone for notifying another person, of a
`potential component failure.
`
`15
`
`25
`
`DESCRIPTION OF THE PREFERRED
`EMBODIMENTS
`
`oil pressure sensor;
`oil level sensor;
`air ?oW meter;
`voltmeter;
`ammeter;
`humidity sensor;
`engine knock sensor;
`oil turbidity sensor;
`throttle position sensor;
`Wheel speed sensor;
`tachometer;
`speedometer;
`oxygen sensor;
`pitch and roll sensor;
`clock;
`odometer;
`poWer steering pressure sensor;
`pollution sensor;
`fuel gauge;
`cabin thermometer;
`transmission ?uid level sensor;
`yaW sensor;
`coolant level sensor;
`transmission ?uid turbidity sensor;
`break pressure sensor; and
`coolant pressure sensor.
`The term “signal” herein refers to any time varying output
`from a component including electrical, acoustic, thermal, or
`electromagnetic radiation, or mechanical vibration.
`Sensors on a vehicle are generally designed to measure
`particular parameters of particular vehicle components.
`HoWever, frequently these sensors also measure outputs
`from other vehicle components. For example, electronic
`airbag crash sensors currently in use contain an accelerom
`eter for determining the accelerations of the vehicle structure
`so that the associated electronic circuitry of the airbag crash
`sensor can determine Whether a vehicle is experiencing a
`crash of sufficient magnitude so as to require deployment of
`the airbag. This accelerometer continuously monitors the
`vibrations in the vehicle structure regardless of the source of
`these vibrations. If a Wheel is out of balance, or it there is
`extensive Wear of the parts of the front Wheel mounting
`assembly, or Wear in the shock absorbers, the resulting
`abnormal vibrations or accelerations can, in many cases, be
`sensed by the crash sensor accelerometer.
`Every component of a vehicle emits various signals
`during its life. These signals can take the form of electro
`magnetic radiation, acoustic radiation, thermal radiation,
`vibrations transmitted through the vehicle structure, and
`voltage or current ?uctuations, depending on the particular
`component. When a component is functioning normally, it
`may not emit a perceptible signal. In that case, the normal
`signal is no signal, i.e., the absence of a signal. In most cases,
`a component Will emit signals Which change over its life and
`it is these changes Which contain information as to the state
`of the component, e.g., Whether failure of the component is
`impending. Usually components do not fail Without Warn
`ing. HoWever, most such Warnings are either not perceived
`or if perceived are not understood by the vehicle operator
`until the component actually fails and, in some cases, a
`breakdoWn of the vehicle occurs.
`In accordance With the invention, each of these signals
`emitted by the vehicle components is converted into elec
`trical signals and then digitiZed (i.e., the analog signal is
`converted into a digital signal) to create numerical time
`series data Which is then entered into a processor. Pattern
`recognition algorithms then are applied in the processor to
`
`For the purposes herein the folloWing terms are de?ned as
`folloWs:
`The term “component” refers to any part or assembly of
`parts Which is mounted to or a part of a motor vehicle and
`Which is capable of emitting a signal representative of its
`operating state. The folloWing is a partial list of general
`automobile and truck components, the list not being exclu
`sive:
`engine;
`transmission;
`brakes and associated brake assembly;
`tires;
`Wheel;
`steering Wheel;
`Water pump;
`alternator;
`shock absorber;
`Wheel mounting assembly;
`radiator;
`battery;
`oil pump;
`fuel pump;
`air conditioner compressor;
`differential gear;
`exhaust system;
`fan belts;
`engine valves;
`steering assembly; and
`engine cooling fan assembly.
`The term “sensor” refers to any measuring or sensing
`device mounted on a vehicle or any of its components
`including neW sensors mounted in conjunction With the
`diagnostic module in accordance With the invention. A
`partial, non-exclusive list of common sensors mounted on an
`automobile or truck is:
`airbag crash sensor;
`microphone;
`coolant thermometer;
`
`35
`
`45
`
`55
`
`65
`
`
`
`5,809,437
`
`10
`
`15
`
`5
`attempt to identify and classify patterns in this time series
`data. For a particular component, such as a tire for example,
`the algorithm attempts to determine from the relevant digital
`data Whether the tire is functioning properly or Whether it
`requires balancing, additional air, or perhaps replacement.
`Frequently, the data entered into the computer needs to be
`preprocessed before being analyZed by a pattern recognition
`algorithm. The data from a Wheel speed sensor, for example,
`might be used as is for determining Whether a particular tire
`is operating abnormally in the event it is unbalanced,
`Whereas the integral of the Wheel speed data over a long time
`period (a preprocessing step), When compared to such sen
`sors on different Wheels, might be more useful in determin
`ing Whether a particular tire is going ?at and therefore needs
`air. In some cases, the frequencies present in a set of data is
`a better predictor of component failures than the data itself.
`For example, When a motor begins to fail due to Worn
`bearings, certain characteristic frequencies began to appear.
`Moreover, the identi?cation of Which component is causing
`vibrations present in the vehicle structure can frequently be
`accomplished through a frequency analysis of the data. For
`these cases, a Fourier transformation of the data is made
`prior to entry of the data into a pattern recognition algorithm.
`Other mathematical transformations are also made for par
`ticular pattern recognition purposes in practicing the teach
`ings of this invention. Some of these include shifting and
`combining data to determine phase changes, differentiating
`the data, ?ltering the data, and sampling the data. Also, there
`exists certain more sophisticated mathematical operations
`Which attempt to extract or highlight speci?c features of the
`data. This invention contemplates the use of a variety of
`these preprocessing techniques and the choice of Which ones
`is left to the skill of the practitioner designing a particular
`diagnostic module.
`When a vehicle component begins to change its operating
`behavior, it is not alWays apparent from the particular
`sensors, if any, Which are monitoring that component. The
`output from any one of these sensors can be normal even
`though the component is failing. By analyZing the output of
`a variety of sensors, hoWever, the pending failure can be
`diagnosed. For example, the rate of temperature rise in the
`vehicle coolant, if it Were monitored, might appear normal
`unless it Were knoWn that the vehicle Was idling and not
`traveling doWn a highWay at a high speed. Even the level of
`coolant temperature Which is in the normal range could be
`in fact abnormal in some situations signifying a failing
`coolant pump, for example, but not detectable from the
`coolant thermometer alone.
`In FIG. 1, a generaliZed component 100 emitting several
`signals Which are transmitted along a variety of paths,
`sensed by a variety of sensors and analyZed by the diagnostic
`device in accordance With the invention is illustrated sche
`matically. Component 100 is mounted to a vehicle 180 and
`during operation it emits a variety of signals such as acoustic
`101, electromagnetic radiation 102, thermal radiation 103,
`current and voltage ?uctuations in conductor 104 and
`mechanical vibrations 105. Various sensors are mounted in
`the vehicle to detect the signals emitted by the component
`100. These include a vibration sensor 130 also mounted to
`the vehicle, acoustic sensor 110, electromagnetic radiation
`sensor 115, heat radiation sensor 120, and voltage or current
`sensor 140. In addition, various other sensors 150, 151, 152,
`153 measure other parameters of other components Which in
`some manner provide information directly or indirectly on
`the operation of component 100. All of the sensors illus
`65
`trated on FIG. 1 are connected to a data bus 160. A
`diagnostic module 170, in accordance With the invention, is
`
`6
`also attached to the vehicle data bus 160 and receives the
`signals generated by the various sensors.
`As shoWn in FIG. 1, the diagnostic module 170 has access
`to the output data of each of the sensors Which have
`information relative to the component 100. This data appears
`as a series of numerical values each corresponding to a
`measured value at a speci?c point in time. The cumulative
`data from a particular sensor is called a time series of
`individual data points. The diagnostic module 170 compares
`the patterns of data received from each sensor individually,
`or in combination With data from other sensors, With patterns
`for Which the diagnostic module has been trained to deter
`mine Whether the component is functioning normally or
`abnormally.
`Central to this invention is the manner in Which the
`diagnostic module 170 determines a normal pattern from an
`abnormal pattern and the manner in Which it decides What
`data to use from the vast amount of data available. This is
`accomplished using pattern recognition technologies such as
`arti?cial neural netWorks and training. The theory of neural
`netWorks including many examples can be found in several
`books on the subject including: TechniquesAna' Application
`Of Neural Networks, edited by Taylor, M. and Lisboa, R,
`Ellis HorWood, West Sussex, England, 1993; Naturally
`Intelligent Systems, by Caudill, M. and Butler, C., MIT
`Press, Cambridge Mass. 1990; J. M. Zaruda, Introduction to
`Arti?cial Neural Systems, West publishing Co., NY, 1992
`and, Digital Neural Networks, by Kung, S. Y., PTR Prentice
`Hall, EngleWood Cliffs, N.J., 1993, all of Which are included
`herein by reference. The neural netWork pattern recognition
`technology is one of the most developed of pattern recog
`nition technologies. NeWer and more ef?cient systems are
`noW being developed such as the neural netWork system
`Which is being developed by Motorola and is described in
`US. Pat. No. 5,390,136 and patent application Ser. No.
`08/76,602. The neural netWork Will be used here to illustrate
`one example of a pattern recognition technology but it is
`emphasiZed that this invention is not limited to neural
`netWorks. Rather, the invention may apply any knoWn
`pattern recognition technology. A brief description of the
`neural netWork pattern recognition technology is set forth
`beloW.
`Neural netWorks are constructed of processing elements
`knoWn as neurons that are interconnected using information
`channels call interconnects. Each neuron can have multiple
`inputs but only one output. Each output hoWever is con
`nected to all other neurons in the next layer. The neurons in
`the ?rst layer operate collectively on the input data as
`described in more detail beloW. Neural netWorks learn by
`extracting relational information from the data and the
`desired output. Neural netWorks have been applied to a Wide
`variety of pattern recognition problems including speech
`recognition, optical character recognition, and handWriting
`analysis.
`To train a neural netWork, data is provided in the form of
`one or more time series Which represents the condition to be
`diagnosed as Well as normal operation. As an example, the
`simple case of an out of balance tire Will be used. Various
`sensors on the vehicle are used to extract information from
`signals emitted by the tire such as the airbag accelerometer,
`a torque sensor on the steering Wheel or the pressure output
`of the poWer steering system. Other sensors Which might not
`have an obvious relationship to tire unbalance are also
`included such as, for example, the vehicle speed or Wheel
`speed. Data is taken from a variety of vehicles Where the
`tires Were accurately balanced under a variety of operating
`conditions also for cases Where varying amounts of unbal
`
`20
`
`25
`
`30
`
`35
`
`40
`
`45
`
`55
`
`60
`
`
`
`5,809,437
`
`7
`ance Was intentionally introduced. Once the data had been
`collected, some degree of preprocessing is usually per
`formed to reduce the total amount of data fed to the neural
`network. In the case of the unbalanced tire, the time period
`betWeen data points might be chosen such that there are at
`least ten data points per revolution of the Wheel. For some
`other application, the time period might be one minute or
`one millisecond.
`Once the data has been collected, it is processed by a
`neural netWork generating program, for example, if a neural
`netWork pattern recognition system is to be used. Such
`programs are available commercially, e. g., from NeuralWare
`of Pittsburgh, Pa. The program proceeds in a trial and error
`manner until it successfully associates the various patterns
`representative of abnormal behavior, an unbalanced tire,
`With that condition. The resulting neural netWork can be
`tested to determine if some of the input data from some of
`the sensors, for example, can be eliminated. In this Way, the
`engineer can determine What sensor data is relevant to a
`particular diagnostic problem. The program then generates
`an algorithm Which is programmed onto a microprocessor.
`Such a microprocessor appears inside the diagnostic module
`170 in FIG. 1. Once trained, the neural netWork, as repre
`sented by the algorithm, Will noW recogniZe an unbalanced
`tire on a vehicle When this event occurs. At that time, When
`the tire is unbalanced, the diagnostic module 170 Will output
`a message to the driver indicating that the tire should be noW
`be balanced as described in more detail beloW. The message
`to the driver is provided by output means coupled to or
`incorporated Within the module 170 and may be, e.g., a light
`on the dashboard, a vocal tone or any other recogniZable
`indication apparatus.
`Discussions on the operation of a neural netWork can be
`found in the above references on the subject and are Well
`understood by those skilled in the art. Neural netWorks are
`the most Well knoWn of the pattern recognition technologies
`based on training, although neural netWorks have only
`recently received Widespread attention and have been
`applied to only very limited and specialiZed problems in
`motor vehicles. Other non-training based pattern recognition
`technologies exist, such as fuZZy logic. HoWever, the pro
`gramming required to use fuZZy logic, Where the patterns
`must be determine by the programmer, render these systems
`impractical for general vehicle diagnostic problems such as
`described herein. Therefore, preferably the pattern recogni
`tion systems Which learn by training are used herein.
`The neural netWork is the ?rst highly successful of What
`Will be a variety of pattern recognition techniques based on
`training. There is nothing Which suggests that it is the only
`or even the best technology. The characteristics of all of
`these technologies Which render them applicable to this
`general diagnostic problem include the use of time-based
`input data and that they are trainable. In all cases, the pattern
`recognition technology learns from examples of data char
`acteristic of normal and abnormal component operation.
`A diagram of one example of a neural netWork used for
`diagnosing an unbalanced tire, for example, based on the
`teachings of this invention is shoWn in FIG. 2. The process
`can be programmed to periodically test for an unbalanced
`tire. Since this need be done only infrequently, the same
`processor can be used for many such diagnostic problems.
`When the particular diagnostic test is run, data from the
`previously determined relevant sensors is preprocessed and
`analyZed With the neural netWork algorithm. For the unbal
`anced tire, using the data from the crash accelerometer, the
`digital acceleration values from the analog to digital con
`verter in the crash sensor are entered into nodes 1 through n
`
`10
`
`15
`
`25
`
`35
`
`45
`
`55
`
`65
`
`8
`and the neural netWork algorithm compares the pattern of
`values on nodes 1 through n With patterns for Which it has
`been trained as folloWs.
`Each of the input nodes is connected to each of the second
`layer nodes, h-1,h-2, .
`.
`. , h-n, called the hidden layer, either
`electrically as in the case of a neural computer, or through
`mathematical functions containing multiplying coefficients
`called Weights, in the manner described in more detail in the
`above references. At each hidden layer node, a summation
`occurs of the values from each of the input layer nodes,
`Which have been operated on by functions containing the
`Weights, to create a node value. Similarly, the hidden layer
`nodes are in like manner connected to the output layer
`node(s), Which in this example is only a single node O
`representing the decision to notify the driver of the unbal
`anced tire. During the training phase, an output node value
`of 1, for example, is assigned to indicate that the driver
`should be noti?ed and a value of 0 is assigned to not doing
`so. Once again, the details of this process are described in
`above-referenced texts and Will not be presented in detail
`here.
`In the example above, tWenty input nodes Were used, ?ve
`hidden layer nodes and one output layer node. In this
`example, only one sensor Was considered and accelerations
`from only one direction Were used. If other data from other
`sensors such as accelerations from the vertical or lateral
`directions Were also used, then the number of input layer
`nodes Would increase. Again, the theory for determining the
`complexity of a neural netWork for a particular application
`has been the subject of many technical papers and Will not
`be presented in detail here. Determining the requisite com
`plexity for the example presented here can be accomplished
`by those skilled in the art of neural netWork design.
`Brie?y, the neural netWork described above de?nes a
`method, using a pattern recognition system, of sensing an
`unbalanced tire and determining Whether to notify the driver
`and comprises the steps of
`(a) obtaining an acceleration signal from an accelerometer
`mounted on a vehicle;
`(b) converting the acceleration signal into a digital time
`series;
`(c) entering the digital time series data into the input
`nodes of the neural netWork;
`(d) performing a mathematical operation on the data from
`each of the input nodes and inputting the operated on
`data into a second series of nodes Wherein the operation
`performed on each of the input node data prior to
`inputting the operated on value to a second series node
`is different from that operation performed on some
`other input node data;
`(e) combining the operated on data from all of the input
`nodes into each second series node to form a value at
`each second series node;
`(f) performing a mathematical operation on each of the
`values on the second series of nodes and inputting this
`operated on data into an output series of nodes Wherein
`the operation performed on each of the second series
`node data prior to inputting the operated on value to an
`output series node is different from that operation
`performed on some other second series node data;
`(g) combining the operated on data from all of the second
`series nodes into each output series node to form a
`value at each output series node; and,
`(h) notifying a driver if the value on one output series
`node is Within a chosen range signifying that a tire
`requires balancing.
`
`
`
`9
`This method can be generalized to a method of predicting
`that a component of a vehicle Will fail comprising the steps
`of:
`(a) sensing a signal emitted from the component;
`(b) converting the sensed signal into a digital time series;
`(c) entering the digital time series data into a pattern
`recognition algorithm;
`(d) executing the pattern recognition algorithm to deter
`mine if there exists Within the digital time series data a
`pattern characteristic of abnormal operation of the
`component; and
`(e) notifying a driver if the abnormal pattern is recog
`niZed.
`The particular neural netWork described and illustrated
`above contains a single series of hidden layer nodes. In some
`netWork designs, more than one hidden layer is used,
`although only rarely Will more than tWo such layers appear.
`There are of course many other variations of the neural
`netWork architecture illustrated above Which appear in the
`referenced literature. For the purposes herein, therefore,
`“neural network” Will be de?ned as a system Wherein the
`data to be processed is separated into discrete values Whic