Abstract
Electro Mechanical Brake (EMB) is a high-efficiency pure electric vehicle braking system based on electronic, mechanical, and communication network technologies. Due to high costs and safety/reliability concerns, mass production remains a challenge. EMB systems rely heavily on sensor feedback; thus, accurate fault detection and diagnosis (FDD) are critical. This paper establishes an FDD model using Support Vector Regression (SVR) for detection and Support Vector Classification (SVC) for diagnosis. To enhance performance, the Clonal Selection Algorithm (CSA) optimizes SVM parameters. Experimental results show a 62.5% increase in fault detection rate and a 10% improvement in classification accuracy compared to traditional SVM models.
Introduction
As the automotive industry transitions to pure electric vehicles, traditional hydraulic braking is being replaced by Electro-Mechanical Brake (EMB) systems. As a "brake-by-wire" technology, EMB removes the physical link between the brake pedal and the caliper, relying instead on sensors to measure driver intent and motors to apply pressure. However, this total dependence on sensors (current, motor speed, and clamping force) introduces risks. Sensor failures due to electromagnetic interference or mechanical wear can lead to unpredictable braking. Traditional linear filters and standard neural networks often fail to handle the non-linear complexities or suffer from over-fitting. This research proposes a robust diagnostic framework to identify faults before they compromise safety.
Methods
The proposed system utilizes a three-loop control architecture (current, speed, and pressure loops) and a two-stage diagnostic process optimized by bio-inspired algorithms.
Fault Detection and Diagnosis Framework
The core of the system is a hybrid SVM model:
- Fault Detection (SVR): Support Vector Regression creates a "digital twin" of the sensors. By comparing real-time data against predicted outputs, the system generates "residuals." A significant spike in the residual indicates a fault.
- Fault Diagnosis (SVC): Once a fault is detected, Support Vector Classification identifies the specific type: stuck faults, constant gain (scaling errors), or constant bias (offset errors).
Parameter Optimization via CSA
The Clonal Selection Algorithm (CSA), inspired by the biological immune system's response to antigens, is used to optimize the SVM's hyperparameters (penalty factor $C$ and kernel parameters). By "cloning" and "mutating" candidate solutions, CSA avoids the local optima traps common in traditional optimization methods, ensuring higher model reliability.
Results
The model was validated using simulated data from a mathematical state-space model of a Brushless DC Motor. The CSA-optimized model demonstrated significant improvements over baseline SVM approaches.
Key performance metrics include:
- Detection Reliability: The fault detection rate increased by 62.5% over standard SVM.
- Classification Accuracy: The system correctly identified fault types with an average accuracy of 96.7%, a 10% improvement.
- Residual Analysis: As shown in Figure 2, when a sensor fails, the residual "jumps" and holds a steady, elevated value. The CSA-SVR model produces a much cleaner signal with lower steady-state error than standard SVR, making it a more reliable trigger for safety systems.
Conclusion
The integration of CSA with SVM provides a high-performance solution for EMB sensor fault diagnosis. While currently based on simulated data, this work establishes a foundation for fault-tolerant control in electric vehicles. Future research will focus on real-world testing under varying environmental conditions and implementing automated fail-safe logic to maintain braking capability after a sensor failure.
Summary updated