Sign up for our weekly email to stay on top of the latest news and insights!
Presented to you by SAE BRAKE.
This technical paper, provided by SAE International, details how machine learning models enable virtual development of high-performance brake systems. Authored by David Antanaitis from General Motors, it chronicles efforts to capture complex tribological behaviors like friction and wear under extreme conditions, supporting faster vehicle integration and features such as fade warnings.
Key Highlights
- Machine learning models predict non-linear brake behaviors, enhancing accuracy for high-energy scenarios.
- Neural networks are trained on dynamometer data to generalize friction and fluid displacement.
- Virtual simulations allow calibration of fade warning algorithms without physical prototypes.
- Physics-based inputs including temperature, speed, and power improve model reliability.
- System integration validates performance in racetrack conditions, aiding high-power vehicle design.
Physics of Brake Fade
Brake fade refers to changes in friction, compliance, and wear as the system heats up, involving processes like resin decomposition and glazing. In setups with aluminum calipers and cast iron discs, fade leads to higher pressures and fluid requirements for target torque. Vacuum-boosted systems exhibit increased pedal travel, while electro-hydraulic ones mask this, prompting the need for Fade Warning Assist to inform drivers of overheating risks.
Brake Corner Modeling Data
Training data derive from “fade envelope” tests, covering speeds up to 300 kph, decelerations to 2.0g, and temperatures to 700°C. Predictors include temperature measured at disc bulk, hydraulic pressure affecting clamp load, rotational speed linked to sliding effects, braking power for gradients, cumulative in-stop work to capture friction curves, and total cumulative work correlating with wear patterns like taper.
Access the full SAE paper here
Target Signals for Models
Models target apparent friction, linking pressure to torque, and fluid displacement, indicating volume needed for pressurization. These signals provide insights into actuation demands during fade, facilitating system-level predictions.
Machine Learning Development
Using MATLAB tools, shallow neural networks with one hidden layer excel in predicting friction and displacement. Models achieve high correlations, such as 99.99% for friction, but refinements exclude pressure from friction inputs to prevent instability. Additional data like ramp profiles and wear modifiers enhance generalization.
Validation and Refinement
Validated against novel conditions, including flat-track tests simulating Virginia International Raceway, models deliver stable outputs. Revised friction models incorporate temperature, speed, work, and power; displacement uses lookups with wear estimates, accurately reflecting in-stop dynamics.
System Model Predictions
Integrated into Simulink, front and rear models simulate a 650 hp sport sedan and a 1,000 hp variant. The baseline reaches 573°C front temperatures with 15 cc fluid consumption; the higher-power version hits 741°C and 19 cc, underscoring fade vulnerabilities. Results align with physical tests, supporting feature optimization.
Fade Warning Calibration
A simplified algorithm tracks fluid displacement against baselines and wheel lock pressures. Simulations set thresholds: +5 cc and 140 bar for level 1 warnings, +10 cc and 160 bar for level 2, activating appropriately for high-power setups while avoiding false alerts in baselines.
Future Directions
The paper suggests expanding models to handle hysteresis in slip control and predict integration effects using surrogate data. Machine learning holds promise for simulating friction interface design changes, accelerating brake engineering advancements.
Like this article? Join brake professionals from all over the world at SAE’s Brake Colloquium & Exhibition. This annual event is your gateway to the forefront of braking technology and innovation. With a rich history spanning over four decades, the brightest minds in the braking industry gather, creating a platform where ideas are cultivated and partnerships thrive.
Subscribe Today!
Sign up for our weekly eNewsletter and get a free copy of our quarterly digital magazine.
