Dynamic learning-based fault tolerant control for robotic manipulators with actuator faults
Introduction
With the increasing level of industrial automation, robotic systems, as an important component of several complex engineering systems, play an irreplaceable role. In such systems, owing to increasingly strict requirements for operation and productivity, robot manipulators generally approach their design limits. Such conditions may often cause failures in the robot system, such as sensor faults, actuator faults, or unpredictable nonlinear dynamic changes. Therefore, automated health monitoring diagnostics of robot systems and intelligent fault-tolerant control (FTC) of faulty robot systems play a vital role in using robot manipulators without human intervention [1].
Over the past decades, numerous fault diagnosis (FD) algorithms for robot systems have been investigated based on model-based approaches [2], [3] in which an accurate system model is required. However, this requirement is generally difficult to satisfy in most practical applications, particularly in complex nonlinear systems [4]. Subsequently, using the general approximation characteristics of neural networks (NNs), approximation-based learning methodologies have been proposed to learn the anomalies and faults from a monitored robot system [5], [6], and the residual vector is then generated. If a fault is detected, the residual vector is classified using NNs, and then the occurring fault is isolated [7], [8]. In [9], robust FD methods using the learning method for robot manipulators were proposed. By using NNs, not only the occurrence of faults could be detected but also the types of faults could be identified and isolated.
Fault information (FI) obtained from fault detection and isolation (FDI) procedures is useful for FTC, and the influence of a fault is compensated by reconfiguring the controller with the obtained FI. Thus, even if a fault occurs, the faulty system can continue to operate reliably [10]. The basic idea of the FTC scheme based on FDI is that an FD module is employed to detect and isolate the fault (in which a series of online adaptive estimators is included, one is used for fault detection, and the others are used for isolation) based on the FI acquired during the FDI procedure. Then, the FTC component is implemented to reconfigure the controller for a better control performance [11], [12], [13]. Various FTC schemes have been studied for robot systems [9], [14], [15], [16], [17], [18]. For instance, a reconfigurable FTC approach for a robot arm was developed using a sliding-mode observer technique [16]. In [9], [14], based on FI, a nominal computed torque controller (CTC) was reconfigured to control a faulty robot system. However, the use of CTC often cannot guarantee the desired control performance owing to system uncertainties and faults [15]. Some adaptive-based control methods have been proposed to improve the control performance of robot systems during normal (fault-free) operations [19], [20]. In [16], [21], reconfigurable FTC schemes were presented for mechanical systems in which both actuator faults and uncertainty were considered. Huang et al. [22] studied the fault compensation control problem of robot manipulators with actuator failures and disturbances. For autonomous underwater vehicles, Zhu et al. [23] provided a model parameter-free control strategy to address the trajectory-tracking problem under actuator failures. With NNs [24], [25], Shen et al.[26] proposed an NN-based active FTC scheme with nonlinear faults, and an observer-based FTC method was proposed for induction motors with possible actuator faults [27]. In [28], a robust adaptive FTC problem was investigated for robot systems with actuator failures using the sliding mode technique in which NNs were used to compensate for failures and uncertainties. For the aforementioned FD and FTC schemes, NNs are often utilized to estimate the fault model and reconfigure the controller online. However, the learning ability in the adaptive process, that is, the real intelligent capabilities for knowledge acquisition and re-utilization, has not been fully explored. As a result, when performing even the same control task, the estimated parameters have to be re-adapted online, which leads to the consumption of computing resources and limits the implementation in practical applications.
Learning ability, that is, acquiring knowledge from a dynamic environment and using learned knowledge to improve control performance, is the core of an intelligent control system [29], [30]. In recent years, Wang and Hill proposed a deterministic learning (DL) [also known as dynamic learning] method to acquire knowledge from system identification and adaptive neural control (ANC) [31], [32]. Based on DL, the inherent unknown dynamics of the system can be accurately approximated by RBF NNs in the system identification or stable control processes. Using the DL method, numerous interesting results regarding learning control and applications of small FD have been proposed [33], [34], [35], [36], [37], [38]. However, these learning control schemes are mainly considered in a single control environment; that is, it is often assumed that no faults would occur. For robot systems, Chen et al. [39] attempted to investigate FD for robotic manipulators, in which a conventional CTC was used as the executive controller. Although fault isolation is achieved in some sense, owing to the lack of decision-making and control strategies, it is impossible to achieve compensation control for the faulty system. To realize high-performance control in fault environments, it is important to develop intelligent FTC for complex robot systems.
In this paper, we present a novel active FTC scheme for robot manipulators with actuator faults, which will provide a new concept for intelligent FTC of systems in fault environments. The proposed scheme comprises two modules: a rapid FDI module and an FTC module. In the FDI module, first, under the normal controller, the uncertainties and unknown dynamics of each actuator fault (including the fault-free mode) are accurately identified via DL, utilizing the acquired knowledge, and a series of dynamical estimators are constructed to represent different fault modes of the robot manipulator. Then, by comparing the monitored robot system with the constructed estimators, a set of residuals is generated, and rapid FDI can be achieved simultaneously using the minimum residual principle. In the FTC module, an adaptive neural controller is first designed for the fault-free mode and all possible actuator fault modes, and the closed-loop inherent dynamics of the faulty manipulators are accurately identified by the RBF NNs. Using the learned constant NNs, a set of candidate experience-based fault-tolerant controllers (EBFTCs) is constructed, where each controller corresponds to a specific type of actuator fault mode. Then, when a fault reoccurs, based on the rapid FDI results, the correct EBFTC can be directly selected and used to achieve stability and improve control performance.
The main contributions of this study are as follows: 1) under the proposed FTC scheme, both the uncertainties and actuator faults can be accurately modeled/learned by RBF networks with the satisfaction of persistent excitation (PE), accurate classification of different fault modes can be achieved, and a specific knowledge-based controller library is constructed for all possible fault modes; 2) no online adjustment of any estimated parameters is required in the processes of both FDI and FTC, the calculation burden is significantly reduced, and rapid FD can be achieved; and 3) with the help of rapid FDI, the correct experience-based controller corresponding to the current fault can be directly selected, and the knowledge embedded in the controller can rapidly and correctly compensate for the effect of the actuator fault. Different from existing FTC approaches [11], [12], [24], [26], the proposed scheme is based on the acquisition of fault knowledge, rapid and accurate FD, and precise compensation control. Under the action of the experience-based controller, the FI can be effectively retained, avoiding the cover-up of the fault under the action of adaptive control, particularly for small faults.
The remainder of this paper is organized as follows: Section 2 presents the problem formulation and the preliminaries. The rapid FDI strategy is designed in Sections 3 and 4, fault-tolerant controllers are designed, and the implementation of the rapid FTC scheme is addressed. Section 5 presents simulation studies to illustrate the validity of the proposed FTC scheme. Finally, Section 6 concludes the paper.
Section snippets
Problem formulation
Consider -link robot dynamics as follows [9]:where denotes the position vector, denotes the velocity vector, denotes the acceleration vector, represents the matrix of inertia, represents the matrix of centripetal and Coriolis forces, denotes the vector of gravitational force, denotes the unknown friction vector, represents the input torque vector,
Fault detection and isolation module
This section comprises two phases: fault training (learning) and recognition (diagnosis). Note that rapid recognition of the occurrence and type of fault is a crucial and challenging issue because the occurrence of the fault not only implies the dynamic changes of the controlled system but is also influenced by the execution controller. To address this challenge, the experience-based controller corresponding to the normal (fault-free) mode is set as the execution controller in the following
Fault-tolerant control design
The FTC design includes two phases: the fault-tolerant controller design phase and implementation of the FTC. The objective of the controller design is to construct a candidate EBFTC bank, where each fault mode corresponds to a specific knowledge-based controller. The implementation of FTC is, when a trained fault occurs abruptly, to quickly select the correct EBFTC based on the rapid FDI results to control the faulty robotic system; thus, improved control performance can be achieved.
Simulation studies
In this study, to demonstrate the feasibility of the rapid FTC scheme, the two-link planar robot arm shown in Fig. 2 is considered, and the dynamics of the robot manipulator are given in Abdelatti et al. [34] as follows:where , , , , and . The relevant parameters of the robot arm are selected as
Conclusion
This paper proposed a novel FTC scheme for robot manipulators with actuator faults and provided a new design idea for intelligent FTC. In the FDI module, first, both the uncertainty and each possible actuator fault of the robot under the normal controller are accurately identified via DL, the acquired knowledge is used to construct a series of dynamical estimators for the classification of different actuator failures, and the fault can be detected and isolated rapidly using the estimators. In
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
This work was supported in part by the National Natural Science Foundation of China under Grant 62203262, Grant 61890922, and Grant 62203263, and in part by the Natural Science Foundation of Shandong Province under Grant ZR2022QF124, Grant ZR2020ZD40, and Grant ZR2022QF062.
References (49)
- et al.
A model-based approach to robot fault diagnosis
Knowledge-Based Syst.
(2005) - et al.
Actuator fault detection in nonlinear uncertain systems using neural on-line approximation models
Eur. J. Control
(2009) - et al.
Neural network-based fault diagnosis for spacecraft with single-gimbal control moment gyros
Chin. J. Aeronaut.
(2022) - et al.
Neural network state observer-based robust adaptive fault-tolerant quantized iterative learning control for the rigid-flexible coupled robotic systems with unknown time delays
Appl. Math. Comput
(2022) - et al.
Dual neural networks based active fault-tolerant control for electromechanical systems with actuator and sensor failure
Mech. Syst. Signal Process.
(2023) - et al.
Adaptive model-parameter-free fault-tolerant trajectory tracking control for autonomous underwater vehicles
ISA Trans.
(2021) - et al.
Adaptive neural network state constrained fault-tolerant control for a class of pure-feedback systems with actuator faults
Neurocomputing
(2022) - et al.
Neuroadaptive observer-based discrete-time command filtered fault-tolerant control for induction motors with load disturbances
Neurocomputing
(2021) - et al.
Deterministic learning from neural control for a class of sampled-data nonlinear systems
Inf. Sci.
(2022) - et al.
Adaptive actuator failure compensation control of uncertain nonlinear systems with guaranteed transient performance
Automatica
(2010)
Learning from neural control of nonlinear systems in normal form
Syst. Control Lett.
A dynamic fault tolerance framework for remote robots
IEEE Trans. Robot. Autom.
Model-based health monitoring for a vehicle steering system with multiple faults of unknown types
IEEE Trans. Ind. Electron.
Neural network based fault detection in robotic manipulators
IEEE Trans. Robot. Autom.
Fault detection and isolation in robotic manipulators via neural networks: a comparison among three architectures for residual analysis
J. Robot. Syst.
A methodology for fault diagnosis in robotic systems using neural networks
Robotica
Fault detection, isolation, and accommodation control in robotic systems
IEEE Trans. Autom. Sci. Eng.
Neural-network-based robust fault diagnosis in robotic systems
IEEE Trans. Neural Netw.
Diagnosis and Fault-Tolerant Control
Adaptive fault-tolerant control of nonlinear uncertain systems: an information-based diagnostic approach
IEEE Trans. Autom. Control
Decentralized fault tolerant control of a class of interconnected nonlinear systems
IEEE Trans. Autom. Control
Local joint information based active fault tolerant control for reconfigurable manipulator
Nonlinear Dyn.
Finite time fault tolerant control for robot manipulators using time delay estimation and continuous nonsingular fast terminal sliding mode control
IEEE Trans. Cybern.
Reconfigurable tolerant control of uncertain mechanical systems with actuator faults: a sliding mode observer-based approach
IEEE Trans. Control Syst. Technol.
Cited by (10)
Adaptive fault-tolerant control for high-order fully actuated system with full-state constraints
2023, Journal of the Franklin InstituteSome m-fold symmetric bi-univalent function classes and their associated Taylor-Maclaurin coefficient bounds
2024, Journal of Inequalities and ApplicationsA learning-based approach to fault detection and fault-tolerant control of permanent magnet DC motors
2023, Journal of Engineering and Applied ScienceA Multi-Robot-Based Architecture and a Trust Model for Intelligent Fault Management and Control Systems
2023, Electronics (Switzerland)