車輛動態估測與預測系統

Vehicle Dynamics Estimation and Prediction Systems

許齡元(Ling-Yuan Hsu)

指導教授 ： 陳宗麟

國立交通大學/工學院/機械工程系所/博士(2011年)

https://doi.org/10.6842/NCTU.2011.00975

摘要

本論文針對車輛動態模型已知及車輛動態模型未知兩種案例，分別提出兩套預測車輛動態的解決方案。前者透過以車輛動態模型為基礎的狀態觀察器來估測當下時間的車輛動態與道路角度資訊，並用此資訊與已知之車輛模型來預測未來時間的車輛動態；後者透過感測器混合系統（Sensor Fusion System）來估測當下時間的車輛動態與道路角度資訊，然後自行建立一車輛模型，利用感測器混合系統所獲得之車輛動態與道路角度資訊來鑑定出此一車輛模型中的系統參數，再藉由上述資訊（車輛模型、車輛動態、道路角度）來預測未來時間的車輛動態。兩套預測車輛動態的解決方法皆可以判斷車輛在未來時間內是否有翻覆危機，並可以應用於各式車輛即時（Real Time）控制系統中。
　　在以車輛動態模型為基礎的車輛動態估測／預測系統中，本論文主要探討：（1）如何透過可觀察性矩陣（Observability Matrix）來決定所需要的感測器，進而獲得預測時所需要的當下時間的車輛動態資訊；（2）發展一套新式的狀態觀察器－循環式狀態觀察器（Switching Observer）用以降低非線性狀態觀察器的設計難度及減少計算量，因此可適用於本論文的車輛系統（高階、非線性系統）。由於上述作法中車輛動態模型的準確與否，將嚴重影響車輛動態估測／預測的準確性。因此本論文提出另一作法，即在車輛參數未知的狀況下進行車輛動態估測與預測。此作法主要探討：（1）如何選取適當的感測器及感測器混合技術，在不使用車輛模型的狀況下，進行車輛動態與道路角度估測；（2）如何利用所得到的車輛動態及道路角度資訊進行車輛參數、道路摩擦係數鑑定；（3）如何整合上述資訊進行車輛動態預測。
　　為了以軟體驗證所提出的車輛動態估測／預測系統的可行性，本論文首先建立一具20個動態、6自由度的「完整車輛模型（Full-State Vehicle Model）」來模擬真實車輛的行為。此車輛模型相異於先前文獻在於其包含道路角度資訊，且可以描述車輛翻覆（Rollover）行為。
　　在車輛參數已知的例子中，藉由觀察性矩陣的分析得知最少需採用四種感測器（側向加速度感測器、縱向速度感測器、橫擺角度感測器以及四側懸吊系統位移量感測器）即可在道路角度未知的狀況下、平常駕駛或是輪胎抬離地面狀況下，即時估算出車輛動態，且成功的預測車輛於未來時間的動態。由模擬結果得知，在動態估測部分：車輛姿態與道路角度的估測誤差皆分別小於0.5度與3.59度，且不論車輛於未來時間是否翻覆，本作法都能夠粗略地預測出於未來時間的動態，其預測之「相對誤差（Relative Accuracy）」分別為0.21%與4.3%。在車輛參數未知的例子中，車輛動態估測系統在輪胎抬離地面時會無法捕捉部份車輛動態，而在輪胎未離地的狀況下，位移與姿態估測誤差分別小於0.3公尺與0.11度，道路角度的估測誤差小於0.15度，車輛於未來時間未翻覆的案例中，其預測誤差為0.51%；於未來時間翻覆的案例中，其預測誤差為27.3%。在翻覆過程中預測結果較不準確，主要是因為在參數鑑定過程中採用過於簡化的輪胎模型。
　　本論文亦嘗試將所獲得的車輛即時動態及未來動態應用於車輛軌跡跟隨控制系統中。所採用的控制策略乃是採用差動式煞車（Differential Brake），不使用方向盤，進行軌跡跟隨。相較於先前文獻，本作法的特點在於：（1）可應用於前輪驅動、前輪轉向的車輛系統；（2）使用階層式架構來簡化控制策略的設計，並選擇順滑模態控制策略以確保系統具強健穩定性（Robust Stability）；（3）在確保強健穩定性下進行最小控制輸入的最佳化設計，且所獲得的最佳控制輸入為一解析解，避免數值搜尋（Numerical Search）的耗時與不確定性。本論文採用兩種車輛模型（完整車輛模型與Carsim轎車模型）來驗證所提出的控制法則的可行性，當車輛初始速度為每小時90公里時，控制系統皆可成功地調節車輛進行二次車道變換。當採用當下時間之車輛動態資訊進行控制時，其側向位移誤差小於0.032公尺。但是當採用未來時間之車輛動態資訊進行控制時，雖然可以提早0.5秒控制車輛、減少52.42%的最大車輛橫擺角速度以及降低37.34%的控制輸入總和，但是卻犧牲軌跡跟隨之精度（側向位移誤差為0.1307公尺）。

關鍵字

車輛動態 ； 估測系統 ； 預測系統 ； 參數鑑定 ； 車輛翻覆 ； 差動式煞車 ； 階層式架構 ； 順滑模態控制 ； 最佳化問題 ； 車輛軌跡跟隨控制 ； 感測器混合 ； 循環式觀察器 ； 卡曼濾波器

並列摘要

In this dissertation, two methods for vehicle dynamics estimation and prediction are proposed for known vehicle model and unknown vehicle model, respectively. The former estimates the current vehicle dynamics and road angle information from a model-based state observer. And then, it predicts the future vehicle dynamic information based on the same vehicle model, vehicle dynamics and road angles. The latter estimates the current vehicle dynamics and road angle information from the sensor-based sensor fusion system. And then it uses the above vehicle dynamics to identify the parameters of a vehicle model. Lastly, it predicts vehicle future dynamics based on the selected vehicle model, vehicle dynamics and road conditions (angles and friction coefficients). Both methods not only can predict the risk of the vehicle rollover in the future time, but also can be applied to the vehicle real-time control system to increase the driving safety.
　　In the approach for the known vehicle model, the proposed method focuses on two things: (1) using the observability matrix of the vehicle model to determine the collaborated sensors so that all the vehicle dynamics can be estimated using the minimum number of sensors; (2) developing a novel state observer techniques (switching observer) to both lower the complexity of designing an state observer for a high-order nonlinear system and reduce the computation load for the real-time implantation. The above model-based approach can predict the vehicle dynamics. However, its feasibility is largely affected by the accuracy of the vehicle model. Therefore, this dissertation proposed another method for unknown vehicle model. The method for unknown vehicle model is focuses on: (1) selecting a set of sensors so that most of the vehicle dynamics can be estimated without a vehicle model; (2) selecting a vehicle model and using the above vehicle dynamics to identify the parameters in the vehicle model; (3) using the above vehicle model, vehicle dynamics, and road conditions to predict the vehicle dynamics in the future time.
　　In order to verify the feasibility of the proposed method by simulations, a full-state vehicle model was employed to mimic the dynamics of a real vehicle on the road. This vehicle model differs from most existing vehicle models in including the road angles, and it can describe the rollover behaviors.
　　In the approach for the known vehicle model, the observability analysis suggests the minimum numbers of the incorporated sensors are four, which are lateral acceleration sensor, longitudinal velocity sensor, yaw angle sensor, and four suspension displacement sensors, According to the simulation results, the proposed method can estimate vehicle dynamics even when two of the tires are off the ground. The estimation accuracy for vehicle attitude estimation is less than 0.5 deg; the estimation accuracy for the road angles is less than 3.59 deg. The prediction accuracy is of 0.21% when the vehicle does a left turn, and it reduces to 4.3% when the vehicle rollover happens. For the case when vehicle model is unknown, the analysis indicates that some of the vehicle dynamics cannot be correctly estimated when the vehicle tire is off the ground. In other simulation conditions, the estimation accuracies of the vehicle displacements and attitudes are less than 0.3 m and 0.11 deg, respectively, while the estimation accuracy of the road angles is less than 0.15 deg. The accuracy of the dynamics prediction is of 0.51% when the vehicle does a left turn, and it reduces to 27.3% when the vehicle rollover happens. The inaccuracy of the dynamics prediction in the rollover case mainly comes from using an over-simplified tire model.
　　This dissertation also discusses the effect of using future vehicle dynamics in a vehicle trajectory control system. The proposed trajectory following system uses the differential brake technique to regulate the vehicle trajectory without using the steering wheel. It differs from the existing approaches in: (1) it can be applied to a front-drive, front steer vehicle; (2) the control algorithm is developed using a hierarchical architecture, the sliding model controls are used to ensure the robust stability for the entire system; (3) it achieves the minimum control inputs while preserving the robust performance. Additionally, this optimal solution is obtained analytically instead of from numerical search. Two vehicle models (full-state vehicle model and Carsim vehicle model) are employed to verify the robustness of the proposed control system. In the simulation, the initial vehicle velocity is 90 km/hr, the proposed control system can use the current vehicle dynamics information to regulate the vehicle finishing the double lane change with the accuracy of the lateral displacement less than 0.032 m. When using the future vehicle dynamics information, the control algorithm can control the vehicle in advance by 0.5 second, lower the maximum vehicle yaw rate by 52.42%, and reduce the total control efforts by 37.34%. However, the accuracy of the trajectory following is also lowered to 0.1307 m.

並列關鍵字

Vehicle Dynamics ； Estimation System ； Prediction System ； System Identification ； Vehicle Rollover ； Differential Brake ； Hierarchical Architecture ； Sliding Mode Control ； Optimization Problem ； Trajectory Following Control ； Sensor Fusion ； Switching Observer ； Kalman Filter

參考文獻

[2] C. R. Carlson and J. C. Gerdes, “Optimal rollover prevention with steer by wire and differential braking,” in Proc. ASME Dynamic Systems and Control Division, Washington, DC, 2003, pp. 345-354.

[3] E. N. Scanchez, L. J. Ricalde, R. Langari, and D. Shahmirzadi, “Recurrent neural control for rollover prevention on heavy vehicles,” in Proc. IEEE International Conference on Neural Network, 2004, pp. 1841-1846.

[5] P. Gaspar, Z. Szabo, and J. Bokor, “Prediction based combined control to prevent the rollover of heavy vehicles,” in Proc. 13th Mediterranean Conference on Control and Automation, Limas sol, Cyprus, 2005, pp. 575-580.

[6] L.-Y. Hsu and T.-L. Chen, “Vehicle full-state estimation and prediction system using state observers,” IEEE Transactions on Vehicular Technology, vol. 58, no. 6, pp. 2651-2662, 2009.

[7] L.-Y. Hsu and T.-L. Chen, “Estimating Road Angles with the Knowledge of the Vehicle Yaw Angle,” Transactions on ASME Journal of Dynamic Systems, Measurement and Control, vol.132, no. 3, pp. 031004, 2010.