Prognosis and Adaptive Control Allocation for Stochastically Deteriorating Over-Actuated Systems

A feedback control system is over-actuated when it is equipped with multiple actuators to achieve the same objective in different ways depending on the control intensity allocated to each actuator. Under the impacts of usage and operating conditions, some actuators degrade more quickly than others, and subsequently can no longer effectively contribute to the performance of the entire system. Allocation laws are thus required to appropriately redistribute the control intensities on the actuators taking into account their health state. The design of such a law can rely on the remaining useful life prognosis of the entire over-actuated system, as well as its actuators.

Most researches on this topic assume that actuator degradation can be gathered by direct inspections. In this PhD work, we place ourselves in a more realistic situation where the input-output data of the over-actuated system are the only information available at each inspection. This consideration leads to new challenges in the stochastic modeling of deteriorating over-actuated systems and in the design of optimal control allocation laws/maintenance policies for their actuators.

 

+ Identification and synthesis of degradation indices for the actuators: Since it is impossible to directly measure the degradation of each actuator, the central question is how to derive degradation indices of the entire over-actuated system and its actuators from its system input-output data. The major difficulty lies in the principle of the feedback control, which offers the desired performance for an over-actuated system, but at the same time conceals its degradation.

+ Modeling the evolution of actuator degradation taking into account their stochastic dependencies: To achieve the required performance, an over-actuated system coordinates its actuators by distributing the control intensity among them. The degradation indices obtained for an actuator reflect not only its own degradation, but also its interaction with the degradation of other actuators. It is therefore more reasonable to model the joint evolution of actuator degradation using multivariate stochastic processes, rather than describing their individual degradation using independent univariate processes. In this context, the major issues lie in the way to model the stochastic dependencies between the actuators, to integrate them into the actuator degradation models, and to estimate the parameters of the model from the synthesized degradation indices.

+ Prognosis of the actuators remaining useful life: Given the degradation models from the previous step, the central issue in the remaining useful life prognosis of the actuators is to determine an individual failure threshold for each of them. This involves reconstructing a critical degradation threshold from the input-output data of the over-actuated system, beyond which the actuator in question can no longer contribute effectively to the overall performance of the system. Once the failure threshold has been determined, the prognosis of the actuator remaining useful life returns to specifying the probability law of the remaining duration before its degradation reaches this threshold.

+ Design of control allocation laws taking into account the remaining useful life of the actuators: This involves studying various strategies to optimally adapt the control intensity for each actuator to a quantile of its remaining useful life, while preserving the overall performance of the over-actuated system. The major challenge comes from the intrinsic complexity of the constrained multi-variable and multi-objective optimization problem.

Funding category: Contrat doctoral

UTT salary

PHD title: Doctorat en Sciences pour l'Ingénieur, spécialité Optimisation et Sûreté des Systèmes

PHD Country: France