Fault detection in the electrohydraulic actuator using Extended Kalman Filter
Chinniah, Yuvin Adnarain
MetadataShow full item record
In this thesis a fault detection technique for a high performance hydrostatic actuation system was developed and evaluated. The Extended Kalman Filter (EKF) was used for parameter identification and was applied to an Electrohydraulic Actuator (ERA) and the performance of the technique is discussed. The ERA is a high performance, closed loop actuation system consisting of an AC variable speed electric motor, a bi-directional gear pump, an accumulator, check valves, a cross-over relief valve, connecting tubes and a custom made symmetrical actuator. The ERA has potential applications in the aerospace industry for flight surface actuation and in robotics. Failures in the ERA can pose a safety hazard and unscheduled maintenance can result in costly downtime. Fault detection in the ERA will increase its safety and efficiency. The proposed preventive maintenance approach involves monitoring the ERA by estimating two parameters of interest, namely the effective bulk modulus and the viscous damping coefficient. Lowering of the effective bulk modulus, as a result of air entrapment, will affect the response of the ERA and may cause stability issues, by lowering the bandwidth of the system. Changes in the damping coefficient for the actuator can indicate deterioration of the oil, wear in the seals or changes in external friction characteristics. The two parameters were estimated using the EKF and changes in the estimated values were related to faults in the system. Prior to applying the EKF to the ERA prototype, an extensive simulation study was carried out to investigate the feasibility of the approach as well as the level of accuracy to be expected with the experimental system. The simulation study was used to verify that changes in the two parameters were detected and accurately estimated. In this study, an attempt was also made to visit some of the problems reported with the use of the EKF for fault detection purposes, namely the difficulty in setting the correct values in the matrices to initialize the EKF algorithm and the presence of biases in the estimates. The problem was believed to be linked to system observability which was investigated in this research. It was found that using observable state space models for the EKF improved the ability of the EKF to estimate parameters, both in terms of accuracy of the estimations and repeatability of experimental results. System observability was investigated in this work by first using simple mechanical systems and then using the more complex ERA system. An iterative approach was presented whereby parameters were not estimated at the same time but iteratively and using different models. System observability was maintained by reducing the number of states and by using the correct type and number of system measurements. Also, the use of observable systems eliminated the need to choose parameter values, in the initial state vector of the EKF, close to the desired parameter values, as was very often done in previous research. No a-priori knowledge about the parameters was assumed in this research. Biases in the estimates (this has been reported in previous studies) are believed to be due to the filter facing a local minima problem. This problem is linked to the error covariance matrix not converging to a global minimum. In the Kalman Filter, the main objective of the error covariance matrix is to compute the Kalman gain, which is in turn used to correct an estimate with the latest sensor measurement. Errors in the Kalman gain may lead to biases in the estimates. In this study, it was also found that although the system is not observable, it can be detectable, although the converse is not true, and as such, changes in parameters can be detected but not necessarily accurately estimated. Observability ensures uniqueness of the estimate. The effective bulk modulus and viscous damping coefficient were estimated successfully, both in simulations and using experimental data. Faults were introduced in the ERA prototype and changes in the parameters were detected and estimated. The friction characteristic of the actuator for the ERA was also investigated. A novel empirical friction model was proposed. The EKF was used to estimate iteratively (to maintain system observability), the coefficients of that friction function which was believed to be a realistic representation of friction effect in the prototype. Simulation and experimental results were presented. In summary, the application of the EKF technique to the ERA has produced very promising results.