Academia.eduAcademia.edu

Figure 3 – uploaded by M. Steinbuch

See full PDFdownloadDownload figure
Fig. 3. Reference governor design (top) for the controlled belt restraint system (bottom) that the acceleration signal of the vehicle a,., can be measured during the crash, and it is proposed to use the entire motion history to predict future behavior. The crash prediction is updated at a frequency f, = 1/7,, with T;, the prediction period.

Figure 3 Reference governor design (top) for the controlled belt restraint system (bottom) that the acceleration signal of the vehicle a,., can be measured during the crash, and it is proposed to use the entire motion history to predict future behavior. The crash prediction is updated at a frequency f, = 1/7,, with T;, the prediction period.

Related Figures (7)

“ig. 1. Schematic representation of a Continuously Controlled Restraint system Restraint systems could deliver a near optimal protection when they are manipulated as a function of measured signals during the full duration of the crash. In that case, the system is referred to as a continuously controlled restraint system. The continuous manipulation can be performed through a control algorithm that aims at minimizing one or more injury criteria (IC). IC’s are defined as indices of one or more biomechanical responses of the occupant, such as head accel- eration, chest compression, etc. [14]. The control algorithm provides the control values to an actuator based on measured or estimated biomechanical responses. The actuator then continuously alters the restraint system element, e.g. load limiter force, airbag vent size or belt position. A schematic representation of the components of a controlled restraint system is shown in Fig. 1.
“ig. 1. Schematic representation of a Continuously Controlled Restraint system Restraint systems could deliver a near optimal protection when they are manipulated as a function of measured signals during the full duration of the crash. In that case, the system is referred to as a continuously controlled restraint system. The continuous manipulation can be performed through a control algorithm that aims at minimizing one or more injury criteria (IC). IC’s are defined as indices of one or more biomechanical responses of the occupant, such as head accel- eration, chest compression, etc. [14]. The control algorithm provides the control values to an actuator based on measured or estimated biomechanical responses. The actuator then continuously alters the restraint system element, e.g. load limiter force, airbag vent size or belt position. A schematic representation of the components of a controlled restraint system is shown in Fig. 1.
Fig. 4. Left: The optimal setpoint (solid) for the chest acceleration given a vehicle acceleration (dash-dotted). Right: The relative chest displacement does not violate the constraints The algorithm is implemented in Matlab, and results are generated for /; = —0.03 m, lz = 0.2 m, f, = 100 Hz, t. = 0.2 s, T = 0.1 ms and with full a priori crash information Of Zyeh and Vyen. So the crash prediction step in Fig 3 is not used here. Fig. 4 shows the crash pulse, and the optimal setpoint r, which is indeed near constant, without violating constraints. Each optimal solution was found within a few milliseconds on an average workstation!, which is smaller than the optimization period 7p = 10 ms.
Fig. 4. Left: The optimal setpoint (solid) for the chest acceleration given a vehicle acceleration (dash-dotted). Right: The relative chest displacement does not violate the constraints The algorithm is implemented in Matlab, and results are generated for /; = —0.03 m, lz = 0.2 m, f, = 100 Hz, t. = 0.2 s, T = 0.1 ms and with full a priori crash information Of Zyeh and Vyen. So the crash prediction step in Fig 3 is not used here. Fig. 4 shows the crash pulse, and the optimal setpoint r, which is indeed near constant, without violating constraints. Each optimal solution was found within a few milliseconds on an average workstation!, which is smaller than the optimization period 7p = 10 ms.
As mentioned above, the vehicle motion measurement is sampled at the time instances t € {0, T, 2T,...,t-}. We assume that the vehicle velocity reaches zero velocity at the time instances t € {0.15,0.16,...,¢.} s. Now the functions A, :R— R”~? and A, : R — R”~? are used to create the two following matrices The matrices A, and B, are used to obtain a discretized version of the objective function (10), and the optimization problem can be written as
As mentioned above, the vehicle motion measurement is sampled at the time instances t € {0, T, 2T,...,t-}. We assume that the vehicle velocity reaches zero velocity at the time instances t € {0.15,0.16,...,¢.} s. Now the functions A, :R— R”~? and A, : R — R”~? are used to create the two following matrices The matrices A, and B, are used to obtain a discretized version of the objective function (10), and the optimization problem can be written as
Fig. 5. Mean eso values over 18 different crash pulses The information about the maximum prediction error at every time instant is implemented in the setpoint optimiza- tion problem by using a so-called robustness parameter, as will be shown next in the simulation setup.
Fig. 5. Mean eso values over 18 different crash pulses The information about the maximum prediction error at every time instant is implemented in the setpoint optimiza- tion problem by using a so-called robustness parameter, as will be shown next in the simulation setup.
Results with the combined RG, i.e. the robust setpoint optimization and the vehicle prediction, are generated at update frequencies fy = f, = 100 Hz, and with a sample time of JT’ = 0.1 ms. Fig. 6 shows the results of the RG for the same EuroNCAP frontal impact pulse as in Fig. 4. It illustrates that the calculated setpoint, based on a vehicle motion prediction, is desirably close the optimal setpoint, obtained with a priori crash information. This implies that without knowledge of the crash, almost optimal behavior can be predicted in terms of the injury parameter Amax for the considered impact pulse. Note there is a slight constraint violation on the upper bound, due to errors in the vehicle prediction. Fig. 6. Left: The optimal setpoint for the chest acceleration with (dash- dotted) and without (solid) prior knowledge of the crash pulse. Right: Relative chest displacement
Results with the combined RG, i.e. the robust setpoint optimization and the vehicle prediction, are generated at update frequencies fy = f, = 100 Hz, and with a sample time of JT’ = 0.1 ms. Fig. 6 shows the results of the RG for the same EuroNCAP frontal impact pulse as in Fig. 4. It illustrates that the calculated setpoint, based on a vehicle motion prediction, is desirably close the optimal setpoint, obtained with a priori crash information. This implies that without knowledge of the crash, almost optimal behavior can be predicted in terms of the injury parameter Amax for the considered impact pulse. Note there is a slight constraint violation on the upper bound, due to errors in the vehicle prediction. Fig. 6. Left: The optimal setpoint for the chest acceleration with (dash- dotted) and without (solid) prior knowledge of the crash pulse. Right: Relative chest displacement
Fig. 7. Chest acceleration (left) and belt force (right) responses from a system with a conventional restraint (dash-dotted) and a closed loop controlled system with a Reference Governor (solid) an identical occupant model with a conventional restraint system, i.e. a 4 KN load limiter, which is obviously uncon- trolled. One can see that the Ama, criterion has reduced by approximate y 60%. To emphasize the interest of these results, note that that the only inputs to the RG algorithm is the initial ve hicle velocity vg and the measured vehicle acceleration at the current time, @yen(t.). On the contrary, it should also be mentioned that this injury reduction is achieved with an ideal belt actuator and perfect sensors, which of course will not be realizable.
Fig. 7. Chest acceleration (left) and belt force (right) responses from a system with a conventional restraint (dash-dotted) and a closed loop controlled system with a Reference Governor (solid) an identical occupant model with a conventional restraint system, i.e. a 4 KN load limiter, which is obviously uncon- trolled. One can see that the Ama, criterion has reduced by approximate y 60%. To emphasize the interest of these results, note that that the only inputs to the RG algorithm is the initial ve hicle velocity vg and the measured vehicle acceleration at the current time, @yen(t.). On the contrary, it should also be mentioned that this injury reduction is achieved with an ideal belt actuator and perfect sensors, which of course will not be realizable.
Related topics:
EngineeringReference GovernorVehicle System
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved