| dc.description.abstract | Impressive progress in vehicle control technologies has equipped modern vehicles with advanced driver assistance and stability control systems to help the driver handle unfavorable driving conditions. These control systems rely on sensor measurements and/or information estimated based on these measurements to generate control commands for vehicle actuators. Therefore, any failure in sensors and actuators can degrade the performance of these control systems and cause instability in vehicle operation. Sensor failures make the control systems generate undesired control commands and actuator failures prevent desired control commands from being applied. To achieve safe and satisfactory vehicle operation, a real-time health monitoring system is crucial for vehicle sensors and actuators. The reliability of health monitoring is determined by its sensitivity to faults and robustness to disturbances. A reliable health monitoring system is responsible for timely detecting the occurrence of a fault in the target vehicle, accurately identifying the source of the fault, and properly determining the type and magnitude of the fault. This information is crucial for reconstructing sensor data or scaling actuator output, which ultimately could result in a fault-tolerant vehicle system.
This thesis proposes a hybrid model/data fault detection and diagnosis system to monitor the health status of any sensor or actuator in a vehicle. The proposed approach works based on residuals generated by comparing sensor measurements or control inputs with their estimations. The estimations are obtained by a hybrid estimator that is developed based on the integration of model-based and data-driven estimators to leverage their strengths. Due to the poor performance of data-driven estimators in unknown conditions, a self-updating dataset is proposed to learn new cases. Estimation based on updated datasets necessitates the use of data-driven estimators that do not require any pre-training. As case studies, the proposed hybrid fault detection and diagnosis system is applied to a vehicle’s lateral acceleration sensor and traction motor. The experimental results show that the hybrid estimator outperforms the model-based and data-driven estimators used individually. These results also confirm that the proposed hybrid fault detection and diagnosis system can detect the faults in the target sensor or actuator, and then reconstruct the healthy value of the faulty sensor or find the failure level of the faulty actuator.
For cases where a set of sensors and actuators should be monitored to evaluate their health status, this thesis develops a general data-driven health monitoring system to detect, isolate, and quantify faults in these components. This method checks the coherency among the target vehicle's variables by using the vehicle's data. The coherency among the vehicle variables means that the variables reflect the physical principles governing the target vehicle's motion and the causality between its states. Each variable corresponds to a component in the vehicle. When a fault occurs in one of the vehicle's components, the coherency among the variables is no longer valid. This idea is incorporated to detect faults. After fault detection, to isolate the faulty component, the developed system explores the coherency in the subsets of the vehicle's variables to find which variable is not coherent with others. Once the faulty component is determined, the health monitoring system uses the remaining healthy components to reconstruct the true value of the faulty sensor or find the failure level of the faulty actuator. The experimental results show that the developed health monitoring system appropriately detects, isolates, and quantifies faults in the test vehicle's sensors and actuators.
The developed data-driven health monitoring system requires a pre-collected dataset for each vehicle to monitor the health status of its sensors and actuators. To relax this requirement, this thesis proposes a universal health monitoring system for the vehicle IMU sensor (measuring the longitudinal/lateral accelerations and yaw rate) by involving vehicle parameters in the health monitoring process. The proposed universal health monitoring system is able to monitor the health status of the target vehicle's IMU sensor using the other vehicles' data. The performance of the universal health monitoring system is evaluated by simulations. The simulation results are in line with what is expected. | en |
