The online actuator fault diagnosis method is proposed with totally utilizing the characteristics of over-actuated system; fault diagnosis method consists of fault isolation and identification method. The system is defined as over-actuated system when the number of actuators is greater than the number of state variables. The over-actuated system can enhance the robustness of the system because the redundant degree-of-freedoms are provided as input matrix of the system has null space; it is referred as actuator redundancy. The multi-objective can be achieved by using the technique for handing of actuator redundancy such as control allocation; the separated tuning between control of system behavior and distribution of total control demand to individual actuators can be facilitated. However, the increasing number of actuators also increases the possibility of single or multiple actuator faults occurrence; fault isolation and identification procedure becomes more complicated. Therefore, in this paper, the actuator fault is focused for fault diagnosis. The existing fault isolation methods are difficult to apply to the system which has many actuators; the conventional methods based on structural residual need not only a lot of computational power but also complicated procedure for multiple faults isolation. With assumption that the number of fault occurred at the same time is sufficiently smaller than the number of actuators of over-actuated system, much simpler multiple faults isolation method can be derived. Due to there are fewer output residuals than actuators, the system can be considered to underdetermined. Also, as the number of faults is sufficiently smaller than the number of actuators, the concept of sparsity can be adopted for fault isolation. A sparse representation is one in which a small number of solutions contain a large proportion of the energy. The L1-norm minimization method which is widely used to image processing or signal processing can be the method for obtaining sparsest solution. It has advantages compared to structural residual analysis; the number of fault is automatically determined and the number of calculation is small regardless of the number of actuators. For a high degree of system reliability, fault identification as well as isolation must be considered in the presence of fault. To achieve identifiability, a proper perturbation input signal is needed; it also leads to output response excitation. The systems, in which users are not on board, may not be necessary to control the system and identify the fault simultaneously. However the systems operated in online with users should be always controlled to ensure user’s safety although the fault identification is needed in faulty situation. Existing input signal design methods are not suitable for online fault identification because the output response excitation cannot be avoided; the advantages of redundant actuators are not utilized perfectly even for over-actuated system application. If actuator redundancy is utilized perfectly, the perturbation input signal can be designed for online fault identification without loss of control performance. The proposed optimal control allocation method distributes the total control efforts to maximize the sensitivity of the system output to parameters related to the fault position without output response excitation. Therefore online fault identification for safety critical system can be achieved. The two major contributions of proposed fault diagnosis method for over-actuated system are as follows: first contribution is that the most possible and smallest number of faults can be isolated with small number of calculations, second contribution is that the actuator excitation to improve performance of fault identification can be realized without loss of control performance by totally utilizing redundant actuator. The extremely over-actuated electric vehicle that uses four independent driving motors and four independent wheel steering actuators is adopted as an application to verify proposed fault isolation and identification method. The simulation and experiment results support the performance of the proposed approach. In practically, the output excitation cannot be eliminated exactly when the proposed input design method is applied to faulty electric vehicle due to the fault effect. But even in faulty case, the proposed approach is the best way to suppressing loss of control performance for applying perturbation input signal.
과 구동 시스템은 제어 하고자 하는 상태 변수의 수보다 구동기의 수가 더 많은 시스템을 말한다. 과 구동 시스템의 입력 행렬의 랭크 (rank) 보다 열의 수가 더 많기 때문에 입력 행렬에 영 공간 (null space)이 존재하게 되고, 구동기 중복 (actuator redundancy)를 제공한다. 그에 따라 과 구동 시스템은 기존의 시스템에 비해 강건성 (robustness)를 강화 할 수 있다. 하지만, 구동기의 수가 증가함에 따라 구동기의 결함 진단 가능성 또한 증가한다. 그리고 기존의 결함 진단 과정을 그대로 적용하기에는 과정이 어려워지고 복잡해지는 문제점이 있다. 따라서 본 연구에서는 과 구동 시스템의 특성을 이용하여 실시간 구동기 결함 진단 (fault diagnosis) 방법을 제안하였다.
과 구동 시스템의 결함 진단의 과정 중 결함의 위치를 정확하게 파악하기 위한 결함 분류 (fault isolation) 과정을 L1-norm 최소화 방법을 활용하여 제안하였다. 구동기의 수와 동시에 발생하는 결함의 수가 증가함에 따라 계산 과정이 복잡해지는 기존의 방법에 비해, 희소성 (sparsity)이 만족되는 과소결정 시스템 (underdetermined system)이면 항상 일정한 수의 적은 양의 계산 양만을 요구하는 장점이 있다. 다음으로 결함의 크기와 형태를 정확히 파악하기 위한 결함 분석 (fault identification) 과정에 있어서 피셔 정보 행렬 (Fisher information matrix)을 활용하여 결함에 대한 출력의 민감도가 최대가 되도록 하는 섭동 신호를 제안하여 정밀도를 향상시켰다. 구동기 중복의 특성을 활용하여 시스템의 거동에 최소한의 영향만을 주면서 섭동 신호를 각각의 구동기에 적용하는 구조를 제안하여 사용자의 안전성이 실시간으로 보장이 되어야 하는 시스템에 적용할 수 있는 장점이 있다. 1/5 크기의 축소 모형 자동차를 활용한 실험을 통해 제안한 결함 분류와 분석 방법의 성능이 우수함을 확인하였다.
