3.1. Design of Kalman-like Filter
The position sensor in the AET system converts the actual opening of the AET into a voltage signal and feeds it to the controller for realizing the closed-loop control. Owing to the poor driving environment and complexity of the engine’s internal structure, multiple noise sources exist in the engine. For example, the aerodynamic, mechanical, combustion, and electrical noise affect the measurement accuracy of the AET position sensor [
31,
32], thereby affecting the accuracy of the AET control.
The position sensor noise is a superposition of multiple random noises that are non-Gaussian white noises. The detected noise probability density curve in the laboratory environment is shown in
Figure 3.
In order to deal with the disturbance of the non-Gaussian white noise, the system estimation tool employs the Kalman-like filter in this paper. The Kalman-like filter uses an approximate minimum mean square error point process to estimate the state of the Markov chain system, and basic steps of its filtering algorithm are as follows.
Remark 1. The noise signal collected in this study is not only the measurement error of the position sensor but also the electrical noise in the transmission circuit of the position sensor. This is because in the actual process of automobile operation, the superposition of various noises impacts the AET controller (in addition to the impact of the position sensor measurement error). This paper also presents the follow-up control experiment on this platform.
Remark 2. Herein, non-Gaussian white noise is defined as noise for which the probability density function does not obey the normal distribution while the power spectral density obeys the uniform distribution.
Covariance matrix prediction:
Covariance matrix update:
where
is the state vector,
is the state transition matrix,
is the observation matrix,
and
are the gain matrix and the observation vector, respectively,
and
are the prediction error covariance matrix and the measurement noise covariance matrix, respectively,
is the system noise covariance matrix,
is the system noise matrix, and
is the measurement noise matrix. The structure block diagram of the Kalman-like filter is shown in
Figure 4.
As can be seen from Equations (12)–(14) and
Figure 4, there are three differences between the two filters:
(1) In the standard Kalman filter algorithm, the control quantity will affect the state variation in the system, whereas in the Kalman-like filter algorithm, the control quantity only affects the precision of the control and will not affect the state variation in the entire estimation system.
(2) The gain of Kalman filter and Kalman-like filter does not depend on the measured value but is related to the state value of the system. Because the system state of the Kalman filtering algorithm requires a Gaussian distribution, the system state of the Kalman-like filter operates for a discrete time, and given the finite state of the Markov chain, the direct result of this correlation is that the Kalman-like filter used herein constitutes a nonlinear filter, unlike the standard Kalman filter.
(3) Under the standard Kalman filter, the conditional distribution of system states is proven to be Gaussian distribution, the system state estimation is conditional mean, and the filtering error covariance matrix is conditional covariance matrix. However, in our setup, this distribution is consistent with the system state estimate [
33]. The electronic throttle valve position estimation steps can be described as follows.
Step (1) The actual electronic throttle opening angle is defined as the state vector , and the observation vector is obtained.
Step (2) The state prediction value of the system and the covariance matrix prediction value are obtained.
Step (3) The Kalman-like filter gain matrix is calculated.
Step (4) The system status update value and the covariance matrix update value are obtained.
Step (5) is set, and the steps are repeated starting from Step 1.
3.2. Self-Tuning Backstepping Control Strategy
The control goal of the AET is to make the real opening angle follow the reference angle accurately and quickly. According to the nonlinear characteristics of the system, the backstepping control method is used to guarantee the closed-loop stability of the system, and the fuzzy control is used to select parameters of the backstepping controller.
The backstepping method is used to construct the Lyapunov function step-by-step to obtain the corresponding virtual control, and then backstep to the equation that contains the real control input to obtain the control law of the system. The basic steps of designing an AET controller based on the backstepping algorithm are as follows:
Define the first error variable:
where
is the real opening of the electronic throttle valve plate, and
is the reference value of the throttle angle.
The Lyapunov function is:
Its derivative is:
where
is the derivative of
.
To make
a negative definite, select
as the virtual control. The ideal virtual control value is:
where
is a positive constant.
Define the second error variable:
The Lyapunov function is:
The derivative of
is:
To make
a negative definite, let:
where
is a positive constant.
A tracking error system was established for the backstepping control method, and its robustness was analyzed in [
34]. The derivative of the two error variables is given as follows:
The state space expression of the AET tracking error is expressed as:
The Lyapunov function is:
The derivative is obtained as:
According to Young’s inequality:
where
is a constant and
.
Substitute inequality (29) into Equation (28):
where
, and
. Let:
The following inequality can thus be obtained:
According to Equations (27) and (33):
Thus, the following conclusion can be drawn:
According to Equations (26), (31), and (35), it can be concluded that the decay rate of the tracking error is related to the selection of , and the larger the value of , the faster is the decay rate. In addition, the tracking static error is related to the selection of , and the larger the value of , the smaller is the tracking static error. However, excessive values of and would result in a high gain of the controller, which can cause problems in amplifying the disturbance signal and the overshoot oscillation. Hence, herein, the fuzzy control algorithm is chosen to adjust the values of the two parameters and of the backstepping control in real time, and the fuzzy rules are formulated to select reasonable parameter values.
Remark 3. According to the stability proof presented herein, the error converges when the parameters and are positive constants. The parameters adjusted by the fuzzy rule conform to the condition of error convergence and can produce a better tracking effect than the fixed gain. The effect of experimental tracking can also be demonstrated as described above.
The fuzzy logic system comprises four parts: fuzzy module, fuzzy rule base, fuzzy module inference engine, and defuzzification module, as shown in
Figure 5. In the fuzzy logic system, the tracking error
and the derivative of the tracking error
are set as the input of the fuzzy control. The two parameters
and
of the adaptive control are used as output variables, and the range of the input and output parameters is defined according to the actual system output.
The principle of parameter self-tuning is based on the relationship between the parameters , , the tracking error , and its derivative . If the absolute value of the tracking error is large, the parameter should be given a large positive value, so that the static tracking error can be eliminated as little as possible. If the absolute value of the derivative of the tracking error is large, the parameter should be given a large positive value to increase the error decay rate. If the absolute values of and are relatively small, the parameters and should be given relatively small positive values, respectively. It ensures that the controller has appropriate gain parameter when approaching a stable state, thereby avoiding the occurrence of system output ripple and oscillation.
Based on the above principles, the fuzzy sets that define the tracking error and the derivatives of the tracking error are {NB (negative big), NS (negative small), ZO (zero), PS (positive small), PB (positive big)}. The fuzzy sets of the two parameters and for the adaptive control are {PS (positive small), PM (positive medium), PB (positive big)}. In fuzzy control, membership functions of triangle, trapezoid, Gauss, and other shapes are usually used to describe the fuzzy degree of input variables and output variables. Considering that the input variables are bounded error parameters, the triangle membership function method is used to fuzzify the input and output variables. The performance of the control system is adjusted by appropriately adjusting the parameter value of the membership function. The inference type of fuzzy control adopts Mamdani inference algorithm in this study.
The fuzzy rule base
is a set of fuzzy rules, where each IF-THEN fuzzy rule
is in the form of:
where
is the number of rules,
is the input membership function, and
is the output membership function. The specific fuzzy rules are listed in
Table 2 and
Table 3 below. Using the average weighting method for defuzzification, the output of the fuzzy controller can be designed as:
where
is the value of the membership function at
.
After defuzzification, the two parameters
and
of the backstepping controller are the outputs. The specific value range is shown in
Figure 6. The backstepping controller selects the parameters
and
output by the fuzzy controller and calculates the optimal control voltage
according to the tracking error
, thereby outputting the ideal output angle
of the AET system.
Remark 4. The selection criteria of parameters are as follows:
Multiply both sides of Equation (32) by :
Integrating Equation (38) on [0, t] can be obtained as follows:
Multiply both sides by
to obtain:
In this paper, the minimum control objective is set as, from full opening to full closing of the valve plate, the adjustment time with a steady-state error of 2% being allowed within 200 ms. The following formula can be obtained by substituting it from Equation (41): .
Combined with Equation (31), it can be obtained: . After the value of is fixed, the value of is gradually increased by trial and error, and a relatively satisfactory compromise can be reached between the response speed and control precision within the interval of [45, 50]. Then, is fixed within the interval of [45, 50], and the value of is gradually increased to obtain that of , which achieves a satisfactory system performance within the interval of [65, 70].