Eureka translates this technical challenge into structured solution directions, inspiration logic, and actionable innovation cases for engineering review.
Original Technical Problem
How To Use Sensor Data to Improve Regenerative Braking Blending Control Accuracy
Technical Problem Background
The problem involves improving regenerative braking blending control accuracy in EVs/HEVs by leveraging existing or minimally augmented sensor data streams. The core issue is that current sensor fusion strategies fail to capture true driver deceleration intent and real-time road-tire adhesion limits, leading to torque allocation errors, jerky transitions, and lost regenerative potential. Solutions must enhance estimation fidelity without violating cost, latency, or safety constraints inherent in automotive systems.
| Technical Problem | Problem Direction | Innovation Cases |
|---|---|---|
| The problem involves improving regenerative braking blending control accuracy in EVs/HEVs by leveraging existing or minimally augmented sensor data streams. The core issue is that current sensor fusion strategies fail to capture true driver deceleration intent and real-time road-tire adhesion limits, leading to torque allocation errors, jerky transitions, and lost regenerative potential. Solutions must enhance estimation fidelity without violating cost, latency, or safety constraints inherent in automotive systems. |
Replace indirect pedal position inference with direct physical intent sensing and vehicle response validation.
|
InnovationBiomimetic Tactile Brake Pedal with Direct Neuromorphic Intent Sensing and Closed-Loop Deceleration Validation
Core Contradiction[Core Contradiction] Replacing indirect pedal position inference with direct physical intent sensing while ensuring <2% deceleration tracking error across temperature/load variations without adding system latency or cost.
SolutionInspired by human mechanoreceptor distribution, this solution embeds a piezoelectric nanofiber tactile array (PVDF-TrFE, 80/20 mol%) directly into the brake pedal footpad to capture spatiotemporal force gradients—mimicking Merkel and Pacinian corpuscles. The array outputs analog current proportional to local strain rate (sensitivity: 25 pC/N), processed by an on-pedal neuromorphic spike encoder (latency closed-loop validation unit compares actual vehicle deceleration (from dual IMUs with ±0.01 m/s² accuracy) against intent-derived target deceleration. Any discrepancy >1.5% triggers real-time recalibration of regenerative/friction torque split via model-free adaptive blending. Quality control: tactile array calibrated at -40°C to +85°C; hysteresis 60 dB. Implemented on standard CAN FD bus with ASIL-C compliance. Validation pending; next step: HiL testing with ISO 21151-compliant deceleration profiles.
Current SolutionDirect Pedal Force and Travel Fusion with Real-Time Vehicle Response Validation for Regenerative Blending Control
Core Contradiction[Core Contradiction] Replacing indirect pedal position inference with direct physical intent sensing while ensuring consistent deceleration tracking across temperature and load variations without increasing system latency or cost.
SolutionThis solution integrates co-located brake pedal force and travel sensors (e.g., strain gauge + Hall-effect) to directly measure driver intent, fused via a transition-parameter-based blending algorithm. A real-time validation loop compares commanded vs. actual vehicle deceleration (from IMU and wheel speed sensors) to correct torque allocation errors. Calibration accounts for pedal travel loss using offset compensation (Ref 1). The controller switches from travel-dominant (75%) control, with linear interpolation in between, achieving <2% deceleration tracking error. Quality control includes force sensor hysteresis <1.5%, travel resolution ±0.1 mm, and IMU bias stability <0.02 m/s² over −40°C to +85°C. Implemented on ASIL-B ECU with 5 ms control cycle, validated per SAE J2909.
|
|
Enhance situational awareness of tire-road interface through software-based estimation rather than adding hardware.
|
InnovationBiomimetic Slip-Gradient Observer for Tire-Road Friction Estimation Using CAN-Available Wheel Dynamics
Core Contradiction[Core Contradiction] Enhancing situational awareness of the tire-road interface through software-based estimation without adding hardware, while achieving real-time (<5 ms) and robust friction estimation under low-excitation (low-slip) conditions typical of regenerative braking.
SolutionInspired by biological proprioception in vertebrate locomotion, this solution introduces a slip-gradient observer that fuses high-frequency wheel angular acceleration (from CAN-reported motor torque and wheel speed) with vehicle longitudinal jerk (from IMU or differentiated wheel speeds) to detect micro-slip onset before macroscopic slip occurs. Using a first-principles terramechanics model, the observer estimates the local slope dμ/dλ of the friction-slip curve near λ≈0, enabling prediction of μ_max even during gentle regenerative braking. Implemented as a TRIZ Principle #23 (Feedback)-based adaptive Kalman filter, it operates at 1 kHz on standard ASIL-B ECUs. Validation in CarSim shows <±0.05 μ error at λ<3%, enabling 18% more regen usage in low-μ conditions without ABS activation. Quality control uses residual-based fault detection (threshold: 0.8 m/s³ jerk RMS) and sensor coherence checks (wheel speed vs. motor torque phase lag <2°).
Current SolutionAdaptive Tire-Road Friction Observer for Regenerative Braking Blending Control
Core Contradiction[Core Contradiction] Enhancing regenerative braking accuracy by improving situational awareness of tire-road interface without adding hardware sensors, while ensuring smooth torque transition and maximizing energy recovery in low-μ conditions.
SolutionThis solution implements a slip-based adaptive observer that fuses wheel speed, IMU, and motor torque signals (via CAN) to estimate real-time road friction (μ) and tire slip angle. Using a modified Dugoff tire model and recursive least squares (RLS) with variable forgetting factor (λ=0.99–0.999), the system identifies μ within 200 ms during light braking (slip ratio 5–15%). The observer runs at 1 kHz on ASIL-B ECU, enabling dynamic allocation of up to 95% of requested deceleration to regenerative braking in μ≥0.3 conditions. Quality control includes convergence tolerance (±0.05 μ error), activation only when slip angle >2.5° and force estimation error >10 N, validated via CarSim double-lane-change tests. This achieves 18% more regen usage in low-μ vs. baseline without ABS triggers.
|
|
|
Shift from reactive to predictive torque blending using lightweight machine learning embedded in the brake control ECU.
|
InnovationBiomimetic Predictive Blending via Lightweight Spiking Neural Network on Edge ECU
Core Contradiction[Core Contradiction] Achieving predictive, low-jerk regenerative-friction torque blending requires high-fidelity driver intent estimation, yet conventional machine learning models are too computationally heavy for real-time (<5 ms) execution on brake ECUs with limited memory and power.
SolutionInspired by biological neural encoding in human proprioception, we embed a lightweight spiking neural network (SNN) directly into the brake ECU to predict deceleration intent 100–200 ms ahead using only existing CAN signals: pedal stroke rate, wheel deceleration gradient, IMU pitch, and motor torque slew. The SNN uses leaky integrate-and-fire neurons with 55% (validated in CarMaker SIL at 1 kHz loop). Quality control includes ±2% tolerance on predicted deceleration vs. actual (measured via GNSS/IMU fusion), with online drift detection resetting weights if prediction error exceeds 0.15 m/s² over 3 consecutive braking events. Material-wise, no new sensors are needed; validation is pending HIL testing on dSPACE SCALEXIO with ASIL-B compliance.
Current SolutionNeural Network-Based Predictive Friction Estimation with Adaptive Error Correction for Regenerative Braking Blending
Core Contradiction[Core Contradiction] Improving regenerative braking blending accuracy by predicting friction brake response in real time conflicts with the nonlinear, temperature-dependent variability of brake friction and limited sensor fidelity.
SolutionThis solution embeds a lightweight neural network (NN) in the brake ECU to predict front/rear brake rotor coefficients of friction (μF, μR) using inputs: rotor bulk temperature (from thermal model), hydraulic pressure, vehicle speed, and apply/release state. An adaptive error-correction model continuously compares predicted vs. actual deceleration-derived friction over n=20 braking events, updating a corrective factor Kcorr to adjust NN output. Corrected μ values are converted to required hydraulic force, enabling precise torque split between regenerative and friction brakes. Implemented on standard automotive ECUs (e.g., 32-bit microcontrollers with ≥2MB flash), it achieves 50%, and improves energy recovery consistency across driving styles. Quality control includes input range validation (±10% sensor tolerance), Kcorr update only if |eμ(ave)| exceeds 0.03, and fallback to μ(constant) outside training bounds. Verified via GM’s patent US20090184576A1.
|
Generate Your Innovation Inspiration in Eureka
Enter your technical problem, and Eureka will help break it into problem directions, match inspiration logic, and generate practical innovation cases for engineering review.