Home Technical Arts Contact
Vehicle Dynamics & Control

Model-Based Acceleration Thresholds for Longitudinal Wheel Slip Detection and Control

Akshay Kaundinya

Department of Mechanical Engineering, University of Michigan, Ann Arbor • Active Safety & Automotive Controls

Abstract Summary

Traditional automotive active safety systems, such as Traction Control Systems (TCS), rely on estimating the longitudinal slip ratio (\(\kappa\)) to regulate wheel spin. However, computing slip ratio requires an accurate estimate of the vehicle's true ground speed; a metric that becomes highly unreliable or mathematically singular during deep slip or simultaneous all-wheel slip events on low-friction surfaces. This project demonstrates a robust, velocity-independent traction control architecture that triggers on model-based wheel angular acceleration thresholds derived from a 5-DoF half-car model and a Pacejka tire curve. Sensor differentiation noise is mitigated using a custom three-stage discrete-time filter cascade, and driveline chatter is eliminated using an asymmetric slew-rate-limited PI control law.

Problem Formulation & 5-DoF Vehicle Model

Standard TCS architectures compute slip ratio \(\kappa = \frac{\dot{\theta}r - v_x}{\max(\dot{\theta}r, \epsilon_{\text{tol}})}\). When all wheels break traction simultaneously (e.g., on ice or snow), there is no non-driven reference wheel to gauge chassis velocity \(v_x\), causing traditional estimators to fail.

Governed by wheel rotational dynamics \(J\ddot{\theta} = M_{\text{drive}} - F_x r\), there exists a physical limit to the angular acceleration a tire can sustain under peak road friction. If measured angular acceleration exceeds this threshold, the tire has inevitably exceeded its peak tractive force. To capture the crucial effect of dynamic load transfer on these thresholds during acceleration, a 5-Degree-of-Freedom longitudinal half-car model was derived using Lagrangian mechanics:

Chassis Pitch & Bounce Coupling

Equations of motion governing dynamic load shifts between axles:

$$\ddot{x} = \frac{F_{xF} + F_{xR} - F_{\text{drag}}}{m + m_F + m_R}$$

Dynamic Normal Force Equations

Instantaneous contact patch normal loads determining peak friction:

$$F_{yR} = m_R g + k_R(h - y + d\theta) - c_R(\dot{y} - d\dot{\theta})$$

Longitudinal tractive forces \(F_{xF}\) and \(F_{xR}\) are evaluated using the nonlinear Pacejka Magic Formula \(F_x = D \sin[C \arctan(B\kappa - E(B\kappa - \arctan(B\kappa)))]\), where peak factor \(D = \mu F_y\) updates dynamically with suspension pitch and bounce.

Three-Stage Signal Processing Pipeline

Computing angular acceleration via raw numerical finite differencing of discrete wheel speed sensors (Hall-effect or variable reluctance) violently amplifies quantization noise (producing catastrophic spikes exceeding \(\pm 600\text{ rad/s}^2\)). Standard FIR/Butterworth low-pass filters introduce severe phase delay that causes feedback control instability. To extract clean acceleration with minimal latency, a novel 3-stage cascade was implemented:

Stage 1

Kinematic Kalman Filter

Formulated across state vector \(x_k = [\omega_k, \alpha_k, j_k]^T\) (velocity, acceleration, jerk). Process noise heavily penalizes velocity/acceleration states while keeping jerk free (\(q_{\text{jerk}}=50\)), forcing reliance on internal kinematics rather than noisy differentiation.

Stage 2

Causal Savitzky-Golay

Applies a 3rd-order polynomial fit over an 11-point causal moving window to strip mid-band quantization noise without introducing the linear phase lag characteristic of standard digital filters.

Stage 3

Statistical Shrinkage

Mitigates polynomial boundary variance ("tail-wagging") by pulling the acceleration estimate toward a 10-point local mean with weight \(\beta = 0.01\), completely dampening high-frequency limit cycles.

Asymmetric PI Control & Slew-Rate Limiting

The controller regulates the estimated acceleration (\(\hat{\alpha}_k\)) to the model-based target threshold (\(\alpha_{\text{target}, k}\)). A Proportional-Integral law computes powertrain torque reduction (\(\Delta M\)):

$$\Delta M = \max\left(0, K_p e_k + K_i \sum_{j=1}^{k} e_j \Delta t\right)$$

Key Architectural Mitigations:

  • Zero Steady-State Offset: While a purely proportional (P) controller arrests initial wheel spin, it leaves a persistent steady-state offset above the friction limit. Adding the integral term eliminates this tracking error completely.
  • Exclusion of Derivative Term: Differentiating an estimated acceleration would compute angular jerk, introducing destructive command noise that destabilizes the plant.
  • Asymmetric Slew-Rate Limiter: Rapid torque reinstatement on a nonlinear tire causes violent >10 Hz "bang-bang" chatter that damages vehicle drivelines. The controller cuts torque instantaneously during slip detection, but strictly limits torque recovery rate via \(\dot{M}_{\max}\):
    \(\Delta M_k = \max(\Delta M_{k-1} - \dot{M}_{\max}\Delta t, \Delta M_{\text{raw}})\)

Simulation Results & Multi-Drivetrain Validation

The architecture was validated in MATLAB/Simulink across low-friction surface conditions (\(\mu = 0.25\)) under aggressive torque demands.

Rear-Wheel Drive (RWD) Response

Subjected to sudden step-like torque demands. The integrated slew-rate-limited PI controller halted runaway wheel spin, suppressed high-frequency chatter, and locked the wheel acceleration directly to the boundary of maximum tractive force.

All-Wheel Drive (AWD) with Load Shift

Evaluated on a 50/50 static torque split. As acceleration shifted normal load rearward, front normal force dropped. The decentralized controllers independently adjusted thresholds, cutting front torque more aggressively while allowing rear wheels to deliver higher tractive effort.