Model-free Adaptive Fault-tolerant Control For Subway Trains

With the rapid development of urban rail transit technology,the reliability and safety of subway trains have arisen concern in recent years.The health of the traction/braking actuators in subway trains has become crucial factor affecting the safety of the subway train operation.Therefore,it is very important to design a fault-tolerant controller for actuator fault.The speed and traction/braking force constraints are considered to maintain the safety of the subway train.The main contributions of this work are as follows.Firstly,the single mass-point subway train model and the multiple point-mass with single-coordinate subway train model are established with considering actuator faults,and the constraints of speed and traction/braking force are analysed.The mass of subway trains,resistances,in-train forces and the factors of actuator faults are not easy to obtain accurately in subway train models.The single mass-point train model and multiple point-mass with single-coordinate train model are transformed into compact form dynamic linearization(CFDL)data models with help of the concept pseudo-partial derivative(PPD)and pseudo gradient(PG)proposed under the framework of model-free adaptive control(MFAC),respectively.Then,a fault detection method is established for the actuator fault by least-squares support vector machine(LSSVM).Select the traction/braking force,speed,and speed tracking error of the subway train as samples attributes.And take whether the actuator fault occurs as the labels of fault samples and normal samples,respectively.The classification model can be obtained by training the samples in different fault coefficients.The data generated during the subway train operation is then brought into the classification model to detect fault.The experimental results are given to verify the effectiveness of the fault detection method proposed in this paper.Finally,the fault-tolerant controllers are proposed for the actuator faults in single point-mass train model and multiple point-mass with single-coordinate train model,respectively.The radial basis function neural network(RBFNN)is used to compensate the actuator fault by using the approximation ability of RBFNN for unknow nonlinear function.Then data-driven model-free adaptive fault-tolerant control schemes are designed by only using saturated input/output data of subway trains under the constraints of speed and traction/braking force.Only RBFNN-weights and PPD or PG are needed to updated in control process.The convergences and effectiveness of proposed fault-tolerant schemes are verified by rigorous theoretical analysis and numerical simulations,respectively.