| 摘要: | 本研究提出一套無需依賴慣性感測元件(IMU)與全球定位系統(GPS)之機車縱向車速估測方法，應對煞車、滑移與輪胎鎖死等動態條件而設計。鑑於傳統車速估測常仰賴輪速感測器或慣性量測單元，然於輪胎打滑或鎖死情形下輪速無法反映實際車速，導致估測誤差放大與應用可靠性下降，因此本研究致力於發展具高穩健性的無IMU車速估測。本方法首先建立完整的車輛縱向動態模型，推導車速、輪速與外力間之物理關係，並透過自由滾動實驗辨識輪胎半徑、前後輪之轉動慣量、黏滯阻力係數，使車輛縱向動態模型能準確反映實際動態行為。本研究進一步分別利用自適應卡爾曼濾波器(Adaptive Kalman Filter, AKF)與低通濾波器(Low Pass Filter, LPF) 進行訊號處理，AKF著重於動態調整與高精度估測，LPF則提供平滑化效果並作為比較基準。接續探討訊號融合方法，由融合前後輪等效車速，並根據輪胎狀態動態調整估測權重，以強化煞車滑移階段下之車速估測準確性與穩定性。此外，為補足輪胎與地面間縱向力之估測，本研究構建多項式形式之外力模型，並藉由自迴歸模型(AR)進行階數選擇，以兼顧估測精度與模型簡化。考量實際感測環境中存在訊號雜訊，本研究亦模擬不同訊號雜訊比(SNR)條件下之敏感度分析，以評估輪速品質對估測表現之影響與限制。本研究於模擬與實車驗證中針對多種煞車情境（如單輪煞車、雙輪鎖死、自由滾動）進行測試，並以實車量測車速作為基準比較估測誤差。本研究共進行16組實車測試，各別使用AKF與LPF之濾波融合策略可有效率波雜訊，分別比較AKF形式之車速估測系統架構與LPF形式之車速估測系統架構於不同車輛操作情境下之車速估測成效。整體誤差統計顯示，AKF與LPF平均誤差分別為-0.2931 km/h與-0.2794 km/h，標準差為0.2553 km/h與0.2470 km/h，均方根誤差(RMSE)為0.4239 km/h 與0.4143 km/h，驗證本方法能穩定控制估測誤差於±1 km/h內。驗證本方法具備良好之實用性與動態適應能力。進一步依據操作區段比較，AKF於煞車區段表現最為穩定，標準差與RMSE值均低於LPF，而在驅動與自由滾動區段，LPF則因訊號較平穩而展現更一致之濾波效果。整體而言，AKF結合動態模型與共變異數自調整機制，能即時響應動態變化，本研究最終採用自適應卡爾曼濾波器形式之車速估測系統架構，建立一套高準確度與穩健性之無IMU之車速估測方法。;This study proposes a motorcycle longitudinal speed estimation method that does not rely on inertial measurement units (IMU) or global positioning systems (GPS), and is specifically designed to handle dynamic conditions such as braking, wheel slip, and wheel lock. Traditional speed estimation approaches often depend on wheel speed sensors or IMUs; however, under slip or lock conditions, wheel speed no longer reflects the true vehicle speed, leading to significant estimation errors and reduced reliability. To address this, a robust non-IMU estimation framework is developed.
First, a complete longitudinal dynamics model of the motorcycle is established to derive the physical relationships among vehicle speed, wheel speed, and external forces. Free-run experiments are then conducted to identify key parameters including tire radii, front and rear wheel moments of inertia, and viscous damping coefficients, ensuring the model accurately captures real dynamics. Signal processing is performed using both an Adaptive Kalman Filter (AKF), which emphasizes dynamic adjustment and high-precision estimation, and a Low-Pass Filter (LPF), which provides smoothing and serves as a comparative baseline. A fusion strategy is further introduced, combining front and rear equivalent wheel speeds with dynamically adjusted weights based on tire states to improve accuracy and stability during braking and slip conditions.
In addition, a polynomial external force model is constructed to supplement the estimation of longitudinal tire–road forces, with autoregressive (AR) model order selection balancing precision and model simplicity. To account for sensor noise in practical environments, sensitivity analyses under varying signal-to-noise ratio (SNR) conditions are performed to evaluate the impact of wheel-speed quality on estimation performance. Both simulation and on-road validation are conducted under diverse braking scenarios (e.g., single-wheel braking, dual-wheel lock, free-run), with true vehicle speed measurements serving as the benchmark for error analysis.
A total of 16 experimental tests are carried out. Results show that both AKF- and LPF-based fusion strategies effectively suppress noise. The overall statistical errors demonstrate mean errors of –0.2931 km/h (AKF) and –0.2794 km/h (LPF), standard deviations of 0.2553 km/h and 0.2470 km/h, and root mean square errors (RMSE) of 0.4239 km/h and 0.4143 km/h, respectively, confirming that the proposed method maintains estimation errors within ±1 km/h. Segment-wise analysis reveals that AKF achieves superior stability in braking regions with lower standard deviation and RMSE than LPF, while LPF performs more consistently during driving and free-run conditions due to smoother signals.
Overall, the AKF approach, which integrates dynamic modeling with covariance self-adaptation, demonstrates real-time responsiveness to dynamic changes. This study concludes by adopting the AKF-based framework as the final architecture, establishing a highly accurate and robust non-IMU speed estimation method for motorcycles. |
