Mark Karpenko

Diagnosis of process valve actuator faults using a multilayer neural network

This paper investigates the ability of a multilayer neural network to diagnose actuator faults in a Fisher-Rosemount 667 process control valve. A software package that comes with the valve is used to obtain experimental figures of merit related to the position response of the valve given a step command. The particular values of the dead time, peak time, percent overshoot, steady state error, 63% and 86% rise times, and gain are shown to depend on the severity of three commonly occurring faults: incorrect supply pressure, actuator vent blockage, and diaphragm leakage. The relationships between these parameters form fault signatures for each operating condition that are subsequently learned by a multilayer feedforward neural network. The results show that the trained network has the capability to detect and identify various magnitudes of the faults of interest. In addition, it is observed that the network has the ability to estimate fault levels not seen by the network during training. The approach presented in this paper allows the existing instrumentation to be utilised without modification. Thus, the proposed methodology is practical to implement.

Fault-tolerant control of a servohydraulic positioning system with crossport leakage

This brief details the design of a fault-tolerant control (FTC) scheme for a servohydraulic positioning system with a faulty actuator piston seal that introduces internal (crossport) leakage between the actuator chambers. It is shown that the leakage fault changes the plant type from 1 to 0, decreases the open-loop gain, and increases the effective damping. A fixed-gain linear time-invariant control law is synthesized via quantitative feedback theory (QFT) to guarantee satisfaction of a priori-defined reference tracking and stability requirements, despite the occurrence of the leakage fault. Experiments verify the ability of the designed fault-tolerant controller to compensate for the degrading effects of this fault. Experiments also demonstrate the superior tracking performance of the FTC scheme as compared to a control loop in which the effects of the leakage fault are not considered in the controller design.

Robust Position Control of an Electrohydraulic Actuator With a Faulty Actuator Piston Seal

The development of a fault tolerant control (FTC) strategy to compensate for the degrading effects of fluid leakage across a faulty actuator piston seal in an electrohydraulic positioning system is presented. Due to relatively large variations in the dynamics of the plant, accomplishing the FTC task with a single controller requires a compensator of relatively high gain. Hence, the problem is first reformulated by discretizing the desired range of fault tolerance into a number of distinct levels. Next, a set of low gain local controllers is synthesized via quantitative feedback theory, such that the resulting closed-loop systems all conform to a priori defined performance specifications. Each controller is designed to compensate for a specific level of leakage. A simple switching algorithm is then employed to determine the appropriate control action by scaling each controllers output based upon an estimate of the leakage level. Experimental results illustrate the ability of the designed FTC scheme to compensate for the degrading effect of the leakage fault.

Hardware-in-the-loop simulator for research on fault tolerant control of electrohydraulic flight control systems

This paper describes the development of a hardware-in-the-loop (HIL) simulator to support the design and testing of novel fault tolerant control and condition monitoring schemes for fluid power systems emphasizing flight control applications. The simulator uses a distributed architecture to share, in a synchronized manner, the demanding computational load associated with the real-time simulation amongst a number of desktop workstations connected by a dedicated Ethernet network. The simulator runs a high-fidelity model of the F-16 fighter aircraft that is augmented in this paper by the addition of realistic nonlinear models of the hydraulic flight control surface actuators and a model of the nonlinear control surface aerodynamic loads. A specially designed state-of-the-art hydraulic test rig, which has the capacity to experimentally simulate common failure modes of a typical fluid power circuit, is used to emulate a F-16 horizontal tail actuator. The experimental actuator can thus be exercised against the realtime simulation of a F-16 aircraft operating under a variety of normal or faulty conditions. To add further realism to the simulation, a second experimental hydraulic actuator is used to generate the aerodynamic disturbing load. Novel fault tolerant control and diagnosis algorithms can therefore be verified in a realistic application scenario. Pilot-in-the-loop simulations are supported by the inclusion of a graphical visualization of the aircraft motions. The results of a typical HIL experiment, for a normally functioning hydraulic system, are presented to illustrate the operation of the simulator

A neural network based fault detection and identification scheme for pneumatic process control valves

This paper outlines a method for detection and identification of actuator faults in a pneumatic process control valve using a neural network. First, the valve signature and dynamic error band tests, used by specialists to determine valve performance parameters, are carried out for a number of faulty operating conditions. A commercially available software package is used to carry out the diagnostic tests, thus eliminating the need for additional instrumentation of the valve. Next, the experimentally determined valve performance parameters are used to train a multilayer feedforward network to successfully detect and identify incorrect supply pressure, actuator vent blockage, and diaphragm leakage faults.

Quantitative Fault Tolerant Control Design for a Leaking Hydraulic Actuator

This paper documents the design of a low-order, fixed-gain, controller that can maintain the positioning performance of an electrohydraulic actuator operating under variable load with a leaking piston seal. A set of linear time-invariant equivalent models of the faulty hydraulic actuator is first established, in the frequency domain, by Fourier transformation of acceptable actuator input-output responses. Then, a robust position control law is synthesized by quantitative feedback theory to meet the prescribed design tolerances on closed-loop stability and reference tracking. The designed fault tolerant controller uses only actuator position as feedback, yet it can accommodate nonlinearities in the hydraulic functions, maintain robustness against typical parametric uncertainties, and maintain the closed-loop performance despite a leakage fault that can bypass up to 40% of the rated servovalve flow across the actuator piston. To demonstrate the utility of the fault tolerant control approach in a realistic application, the experimental fault tolerant hydraulic system is operated as a flight surface actuator in the hardware-in-the-loop simulation of a high-performance jet aircraft.

Decentralized Coordinated Motion Control of Two Hydraulic Actuators Handling a Common Object

In this paper, reinforcement learning is applied to coordinate, in a decentralized fashion, the motions of a pair of hydraulic actuators whose task is to firmly hold and move an object along a specified trajectory under conventional position control. The learning goal is to reduce the interaction forces acting on the object that arise due to inevitable positioning errors resulting from the imperfect closed-loop actuator dynamics. Each actuator is therefore outfitted with a reinforcement learning neural network that modifies a centrally planned formation constrained position trajectory in response to the locally measured interaction force. It is shown that the actuators, which form a multiagent learning system, can learn decentralized control strategies that reduce the object interaction forces and thus greatly improve their coordination on the manipulation task. However, the problem of credit assignment, a common difficulty in multiagent learning systems, prevents the actuators from learning control strategies where each actuator contributes equally to reducing the interaction force. This problem is resolved in this paper via the periodic communication of limited local state information between the reinforcement learning actuators. Using both simulations and experiments, this paper examines some of the issues pertaining to learning in dynamic multiagent environments and establishes reinforcement learning as a potential technique for coordinating several nonlinear hydraulic manipulators performing a common task.

Neural network detection and identification of actuator faults in a pneumatic process control valve

This paper establishes a scheme for detection and identification of actuator faults in a pneumatic process control valve using neural networks. First, experimental performance parameters related to the valve step responses, including dead time, rise time, overshoot, and the steady state error are obtained directly from a commercially available software package for a variety of faulty operating conditions. Acquiring training data in this way has eliminated the need for additional instrumentation of the valve. Next, the experimentally determined performance parameters are used to train a multilayer perceptron network to detect and identify incorrect supply pressure, actuator vent blockage and diaphragm leakage faults. The scheme presented here is novel in that it demonstrates that a pattern recognition approach to fault detection and identification, for pneumatic process control valves, using features of the valve step response alone, is possible.

Equivalent Time-Invariant Modelling of Electrohydraulic Actuators with Application to Robust Control Synthesis

Abstract An important aspect of robust control development around hydraulic actuators is establishing a set of equivalent linear time-invariant (LTI) models that describe the dynamics of the system over the desired envelope of operation. The nonlinearities inherent in the hydraulic functions must be recast into an equivalent linear form in order to make the robust control problem amenable to solution by linear techniques. This paper develops a simple model-based approach for evaluating equivalent LTI frequency response functions of an electrohydraulic actuator by Fourier transformation of acceptable actuator input-output data. The efficacy of the numerical procedure is compared with two other available methods, namely small-signal analysis and Golubevs least-squares approach. It is shown that the proposed approach can describe large signal effects and at the same time properly characterize the features of the hydraulic actuator frequency response that are important for robust control design, without the need for a priori information about the asymptotic behaviour or structure of the equivalent LTI transfer function. The applicability of the proposed numerical technique towards development of practical controllers for fluid power systems is demonstrated by the results of a typical robust control design example for an experimental electrohydraulic positioning system.

Coordination of hydraulic manipulators by reinforcement learning

In this paper, a reinforcement learning method is applied to coordinate a pair of horizontal hydraulic actuators engaged in the cooperative positioning of an object. The goal is to enable the actuators to discover how to intelligently select control actions that tend to reduce the interaction forces directed along the axis of motion, while maintaining the desired trajectory. First, a detailed and realistic dynamic model of the entire system is derived. A multi-layer reinforcement learning neural network control architecture is designed next to regulate the interaction force during positioning. To regulate the interaction force, the neural network measures the interaction force and proposes a modification to the a priori prescribed formation constrained position trajectory. Each actuator system is outfitted with such a neural controller so that a decentralized reinforcement learning control system results. Simulations demonstrate the efficacy of the approach towards reducing the interaction forces and minimizing the associated object internal force in a single degree of freedom