Fault detection and isolation system and method

This application claims the benefit of U.S. Provisional Patent Application No. 60/247,849 entitled FAULT DETECTION AND ISOLATION SYSTEM AND METHOD and filed Nov. 9, 2000.

This application claims the benefit of U.S. Provisional Patent Application No. 60/247, 849, entitled FAULT DETECTION AND ISOLATION SYSTEM AND METHOD and filed Nov. 9, 2000.

The present invention is in the field of vehicle control system design. More particularly, the present invention is a model-based fault detection and fault diagnosis system and method for monitoring overall vehicle system performance.

In recent years, increasing interest and requirement for improved vehicle performance, reliability, and safety has focused attention on the use of Fault Detection & Isolation (FDI) when designing vehicle control systems. Fault detection and isolation is becoming one of the most important aspects in vehicle system control design. In order to meet the increasing demand for better performance and reliability, model-based FDI schemes are being developed to address complete vehicle systems, to detect faults in sensors and actuators, and to apply appropriate corrective action without adding new hardware to the vehicle. However, the high complexity of most vehicle systems makes the standard FDI model-based technique difficult to apply without unacceptable computational effort.

The present invention is a novel system and method based on a hierarchical structure of the FDI scheme that reduces the computational effort of prior art systems. The FDI scheme uses the available sensors in a vehicle system and divides them into subsystems of smaller dimensions containing one or more modules that are related or interconnected. The same module may appear in a different subsystem, but the set of all subsystems does not have to contain all of the modules. For this structure, an FDI scheme comprising several detector units is created. Each detector unit receives information from the sensors and outputs a residual that is sent to a high-level detector unit which processes the data and performs the residual evaluation for the selected subsystem. Finally, each subsystem outputs a decision that is sent to a supervisor hazard detector performing the final diagnosis.

FIG. 1 is a block diagram of a vehicle model for an example embodiment of the present invention;

FIG. 2 is a block diagram of the general structure of a model-based FDI method;

FIG. 3 is a block diagram of a residual generator in accordance with an example embodiment of the present invention;

FIG. 4 is a block diagram of a hierarchical diagnostic system in accordance with an example embodiment of the present invention;

FIG. 5 is a block diagram for the structure of a detector unit in accordance with an example embodiment of the present invention;

FIG. 6 is a block diagram of a general module in accordance with an example embodiment of the present invention;

FIG. 7 is a block diagram of a residual evaluation unit in accordance with an example embodiment of the present invention;

FIG. 8 is a block diagram of the FDI scheme of the present invention;

FIG. 9 is a graph of steering wheel angle input;

FIGS. 10-12 are graphs of estimated and actual state variables; and

FIGS. 13-18 are graphs of experimental results for a J-turn.

The present invention may be implemented in accordance with software components that provide the features and functionality described herein. Referring to FIG. 1, a vehicle may be represented, in general, as a block diagram as shown in FIG. 1 constituting of two main subsystems: the core subsystem 124 and the external subsystem. The core subsystem 124 comprises the vehicle 114 , tire 120 , powertrain 118 , steering 112 , suspension 116 , and brake 108 modules. The vehicle module 114 comprises a 16DOF vehicle model. The vehicle model further comprises a vehicle body, (i.e., the sprung mass), and four wheels, (i.e., the unsprung masses). The model contains three translational degrees of freedom—longitudinal, lateral, vertical, and three rotational degrees of freedom—roll, pitch, and yaw for the sprung mass. Each of the unsprung masses has vertical, spin, and steering angle degrees of freedom. The tire module 120 has as inputs the longitudinal slip, the lateral slip, the vertical load, and the camber angle which gives as output the longitudinal and lateral force as well as aligning moment. The powertrain module 118 comprises the engine, the transmission, and the differential models. The engine uses a lookup table with throttle position and engine speed as inputs and gives as output the engine torque. The transmission model inputs the engine torque and transforms the torque based on the selected gear. The differential model proportions the torque from the transmission to the drive wheels. The steering module 112 describes the elastic and geometric properties of the steering system. The suspension module 116 comprises the model of the suspension that may be of four different types: linear spring and damper, nonlinear spring and damper, semiactive suspension, and active suspension. The brake module 108 generates the wheel torques as a function of the driver brake pedal force and brake controller commands.

The external subsystem comprises the environmental module 122 , driver module 110 , sensor module 100 , brake controller module 102 , suspension controller module 104 , and communication module 106 . The environmental module 122 determines the interfaces between the vehicle and the environment. The driver module 110 determines the interface between the driver and the vehicle. This module provides information such as brake pedal force, steering angle, throttle position, and desired gear to the core module. The sensor module 100 models the sensor dynamics. The outputs of this module are sent to the controller module. The brake controller 102 and suspension controller 104 contain algorithms used to control the brake, and the suspension systems. The communication module 106 models communication delays which occur in communication links between controllers.

In the model-based FDI system and method of the present invention, analytical redundancy is used rather than physical redundancy. This analytical redundancy is contained in the static and dynamic relationship between the input and the output variables of the system. The sensitivity of a diagnostic method to modeling error is one of the key issues in the application of model-based FDI methods. In most cases, model-based FDI methods can be described by the block diagram shown in FIG. 2 .

When an accurate model of the plant is available, the general process of the model-based FDI consists of the three stages depicted in FIG. 3 . At the first stage, observations 160 acquired through sensor measurements are compared to analytical values of the same variables in a primary residual generator 162 . The error between measured and calculated variable is called a primary residual. This residual reflects the system behavior, and has nominal zero mean value under normal conditions. At the second stage, the primary residuals that usually deviate from zero due to noise, modeling error and faults, are communicated to a secondary residual generator 164 and converted in secondary residuals by means of filtering, statistical testing ,or spectral analysis to obtain signals that can be used to analyze and isolate faults. Finally, the secondary residuals are communicated to a decision maker 166 and analyzed to isolate the fault and a diagnosis 168 or decision is taken.

In accordance with the present invention, the vehicle system is decomposed into subsystems of smaller dimension containing one or more modules strictly related or interconnected. Referring to FIG. 4, for this structure, the FDI scheme comprises a plurality of fault detector units 186 , 188 , 192 , 194 , 198 , 200 . Each fault detector unit 186 , 188 , 192 , 194 , 198 , 200 outputs a residual that is sent to a residual evaluation unit 184 , 190 , 196 that performs the residual evaluation for the selected subsystem. Finally, each subsystem 184 , 190 , 196 outputs a decision that is sent to a supervisor fault detector 182 performing the final diagnosis 180 . As shown in FIG. 4, some different subsystems for the vehicle are shown and each is constituted by a residual evaluation unit 184 , 190 , 196 and a plurality of fault detector units 186 , 188 , 192 , 194 , 198 , 200 . The scheme for a fault detector unit 222 is depicted in FIG. 5 .

In general, a module may be represented as in FIG. 6 where:

u 0l, i= 1 . . . m are the input vectors

Δu i , i=1 . . . m are the input fault vectors

θ 0l , i=1 . . . m are the nominal parameter vectors

Δθ l , l=1 . . . m are the parameter fault vectors

x i , i=1 . . . m are the state vectors

Δy is the output fault vector

y is the output measured vector.

The module can be described by the following equations { x . 1 = f 1  ( x 1 , u 1 , θ 1 ) y = h 1  ( x 1 , u 1 , θ 1 ) + Δ     y , x 1 ∈ Γ 1 ⋮ ⋮ { x . m = f m  ( x m , u m , θ m ) y = h m  ( x m , u m , θ m ) + Δ     y , x m ∈ Γ m  (1)

with u 0l ,=u 0l +Δu i , θ i =θ 0l +Δθ i , i=1 . . . m, and where Γ i is a subset in which the i th model equations are valid. A fault detection unit is associated with each module. Each fault detection unit contains a multimodel representation of the type { x ^ . 1 = g 1  ( x ^ 1 , u 1 , θ ^ 1 , y ) y ^ 1 = h 1  ( x ^ 1 , u 1 , θ ^ 1 ) , x ^ 1 ∈ Γ 1 ⋮ ⋮ { x ^ . m = g m  ( x m , u m , θ ^ m , y ) y ^ m = h m  ( x ^ m , u m , θ ^ m ) , x ^ m ∈ Γ m  (2)

characterized by the fact that, without any fault, the following conditions hold

{circumflex over (x)} i →x i for t→∞, i =1 . . . n   (3)

The primary residuals 230 , 240 , 246 are sent to the high-level fault detector decision unit 236 as shown in FIG. 7 . In this decision unit 236 , the residual evaluations for the subsystem are performed at the residual evaluation units 234 , 244 , 250 and the result from the decision unit 236 is sent as input to the supervisor fault detector 238 .

The method of the FDI scheme of the present invention comprises the following steps:

1. Partition of the vehicle model into subsystems containing one or more interconnected modules. The same module may appear in more then one subsystem, but the set of all subsystems, in general, does not have to contain all the modules.

2. Associate a fault detector unit to each module or smaller partition and define a multimodel representation and selection of a residual generation method for every subsystem. The method for residual generation may be of different type, but commonly used approaches are the parity space method, the observer method, and the parameter identification method.

3. Define an appropriate residual evaluation method for each subsystem.

To illustrate the method for a specific case, consider the subproblem of fault detection for three important sensors:

the lateral acceleration sensor;

the steering wheel angle sensor;

the yaw rate sensor; and for two parameters:

the front cornering stiffness; and

the rear cornering stiffness.

The structure of this example FDI scheme is shown in FIG. 8 .

The FDI scheme shows a fault detection unit where only one multimodel representation for a simplified front wheel steered, small angle, bicycle model structure is considered. The dependence of the vehicle lateral velocity and yaw rate and the longitudinal velocity on the steering input is modeled. A simplified tire force model is adopted, whereby the lateral forces of the front and rear tires are linearly related to the front and rear slip angles, through Cf and Cr the front and rear cornering stiffness. The model is valid for nonsevere maneuvers, (i.e., for a lat ≧0.2 g, where g is the acceleration due to gravity). The nonlinear model can be described by the equations { v . x = F x M + v y  Ψ . v . y = - 2 M  ( C f + C r )  v y v x - 2 M  ( aC f - bC r )  Ψ . v x - v x  Ψ . + 2     C f MG     δ Ψ ¨ = - 2 I     ( aC f - bC r )     v y v x - 2 I  ( a 2  C f + b 2  C r )     Ψ . v x + 2  aC f IG     δ 

 a is     the     distance     from     front     wheel     to     C .    G .    of     the     vehicle b is     the     distance     from     rear     wheel     to     C .    G .    of     the     vehicle C f is     the     front     cornering     stiffness C r is     the     rear     cornering     stiffness    M is     the     vehicle     mass I is     the     vehicle     moment     of     inertia G is     the     gear     ratio F x is     the     longitudinal     force v x is     the     vehicle     longitudinal     velocity v y is     the     vehicle     lateral     velocity    δ is     the     steering     angle Ψ . is     the     yaw     rate (4)

For this model, it is possible to design the following sliding mode nonlinear observer based only on the yaw rate measurement x ^ . = ( ∂ H  ( x ^ ) ∂ x ^ ) - 1  M  ( x ^ )  sign  ( V  ( t ) - H  ( x ^ ) ) + B     δ (5)

where H  ( x ) =    [ h 1  ( x )  h 2  ( x )  h 3  ( x ) ] h 1  ( x ) = Ψ .    = r h 2  ( x ) =    r . h 3  ( x ) =    r ¨ V  ( t ) =    [ v 1  ( t )  v 2  ( t )  v 3  ( t ) ] v 1  ( t ) =    r  ( t ) v i + 1 =    ( m i  ( ( x ^ ) )  sign  ( x  ( v i  ( t ) - h i  ( x ^  ( t ) ) ) ) eq ,    i = 1 , 2 M  ( x ^ ) =    diag  ( m 1  ( x ^ )  m 2  ( x ^ )  m 3  ( x ^ ) )

The following table shows the error signatures.

To simplify the problem, consider only the case of single faults. The residual vector is

R={a lat −â y1 δ−{circumflex over (δ)}a lat −â y2 C f −C f a lat −â y3 C r −Ĉ r }  (6)

With the choice made above, the error signature described in the Table 1 may be derived.

Some simulation and experimental results obtained from the previous FDI scheme using sliding mode observers illustrate the system and method of the present invention. The tests are carried out for a vehicle with the parameter data set as in table 2.

Referring to FIG. 9, the steering input for a vehicle lane change maneuver at a longitudinal velocity of 25 mph (11 m/s) and without any fault is shown.

The relative state variable estimations (dashed line) are represented in FIGS. 10-12. It is possible to notice that, after a fast transient, the estimates track the true variable with a very small error.

In FIGS. 13-18, the experimental results for a Jturn at constant forward velocity and step change in the steering angle are presented. A steering input fault of 1.25 times the commanded input has been applied during the test. FIGS. 13 and 14 show the residuals for lateral acceleration and steering angle respectively obtained from Unit A 1 . In dashed line are indicated the estimate values from the observer, a flag 0 (threshold evaluation) may be associated to the lateral acceleration residual while a flag 1 is associated to steering angle residual.

The residuals for lateral acceleration and front tire cornering stiffness obtained from Unit A 2 are depicted in FIGS. 15 and 16 while in FIGS. 17 and 18 the lateral acceleration and the rear cornering stiffness are compared with the measured values. At the end, the following residual signature is observed

R ={0 1 0 1 1 1}  (7)

which indicates a steering inputs or C f fault.

The present invention supports implementation of a vehicle health monitor to increase the reliability of a passenger vehicle with experimental validation of the observer design and FDI scheme. While particular embodiments of the invention have been illustrated and described, various modifications and combinations can be made without departing from the spirit and scope of the invention, and all such modifications, combinations, and equivalents are intended to be covered and claimed.