Soft Computing Approach for Sensorless Control in Brake-By-Wire Systems with Electro-Mechanical Brake

Abstract

In this thesis, we propose a new soft computing-based approach for sensorless fault-tolerant control in brake-by-wire (BBW) systems. Research on BBW systems in the automobile industry is actively proceeding. In or-der to mount and drive the electro-mechanical brake (EMB) used as the brake actuator in the hybrid vehicles and electric vehicles for operation reliability it is imperative that the clamping force data is not lost even if a failure occurs in the electrical and electronic sys-tems. In this study, the mathematical modeling of the mechanical part and the electric mo-tor of the EMB system was first established and the cascaded PI controller was designed based on the EMB model. The mechanical part consisted of a reduction gear, screw, in-ner/outer pads, and caliper. A permanent magnet synchronous motor (PMSM) was used for the electric motor and an electronic control unit (ECU) including the micro-controller and the inverter was constructed and experiments were performed. The EMB controller is configured as a cascaded PI control type, and the clamping force controller, speed control-ler, and the current controller are located in the order of the external controller to the inter-nal controller. The gain of the controller is designed to be easily adjusted using the param-eters of the motor. Also, the vector control method was applied to the PMSM to ensure optimal torque operation. Our goal is to apply a new hybrid-type system identification and estimation methods against failure or for sensorless control that can occur in the EMB electronic pedal sensor system and the clamping force sensor by applying the soft computing techniques such as a neural network, fuzzy and genetic algorithm. First, we propose a novel identification of an electronic brake pedal system for a vir-tual sensor system based on a hybrid approach using the group method of data handling (GMDH) and the genetic algorithm (GA). The main idea in the GMDH is to build an ana-lytical function in a feed-forward network based on a quadratic node transfer function whose coefficients are obtained using a regression technique. The analytical GMDH model has been found, and application of this model is very quick and inexpensive compared to other identification techniques. To develop the best network architecture for the GMDH, the GA is arranged in a new approach to design the whole architecture of the GMDH. Second, we study estimation of the clamping force in the EMB actuator part. The main sensors used in the EMB control system are a clamping force sensor to measure clamping force, a rotor position sensor to measure motor rotation angle, and a current sen-sor to measure the current of the three-phase motor. It is necessary to judge the failure of each sensor or developing without sensors in terms of cost and implementation and replace with an appropriate estimation value in the case of failure. In this study, the estimation of the clamping force is more accurate considering the hysteresis at the time of applying and releasing, and the dynamic stiffness model and torque balance model are combined by us-ing a novel Kalman filter optimized by the GA. The application of the GA improves the estimation accuracy by optimizing the noise covariance matrices of the Kalman filter and enables on-line estimation when using a high performance parallel processor. Finally, we verified the performance of the proposed algorithm through experiments. ⓒ 2017 DGIST