1. EHPV® Technology
Advanced Control Techniques for Electro-Hydraulic Control Valves
by Patrick Opdenbosch
Goals
Develop a smarter and self-contained valve.
Investigate algorithms for off-line and on-line learning of the input-output map.
Develop robust trajectory learning controller
Improve the valve’s performance
Explore algorithms and applications via theory, simulation, and implementation on a Hardware-in-the-Loop simulator.
System Description and Modeling
System Features:
Multidisciplinary: Mechanical, Electromagnetic, and hydraulic.
Complex nonlinear behavior
Bi-directional capability
“Zero” leakage
Mechanical pressure compensation
On-line Learning
Control is based on making the valve’s flow conductance, Kv, follow a desired trajectory.
Linearized error dynamics:
Control Law:
Coil cap
Modulating
spring
Coil
Estimation Scheme
Control Input
Modeling Principle
Armature
isol Kv
Electromagnetic
System
Mechanical
System
Hydraulic
System
Forward Step Responses
Pilot
Abstract
Electro-Hydraulic Poppet Valves are used for flow/pressure control in hydraulic machinery. Currently, the valve control is achieved by changing the valve’s conductance coefficient Kv (output) in an open loop manner via PWM current (input). This input current is computed from an inverse input-output map obtained through individual extensive offline calibration. Without any online correction, the map cannot be adjusted to accurately reflect the behavior of the valve as it undergoes continuous operation. The intention is to develop a control methodology to have the valve learn its own inverse mapping at the same time that the transient performance is improved.
Input
Output
Cage
(housing)
Main Poppet
Forward:
Nose to side
Estimation can be done via:
Modified Broyden Method
Recursive Least Squares with Covariance Reset
Recursive Least Square with Forgetting Factor
Compensation
NLPN learning can be accomplished by:
Delta-Rule Method
Recursive Least Squares with Covariance Reset
Recursive Least Square with Forgetting Factor
Reverse:
Side to nose
Experimental Steady State Characteristics: Forward and Reverse Kv as a function of Pressure differential and input current
Reverse Step Responses
Off-line Learning
The input-output map of the EHPV, as well as the inverse of this map, is learned off-line by using a Neural Network structure called Nodal Link Perceptron Network (NLPN). The learning consists of adjusting the “weights” of basis functions (composed of activation functions).
Upper Figure:
Open-loop Response (nominal map only)
Lower Figure:
Estimated Parameters J and Q
Upper Figure:
Closed-loop Response
Lower Figure:
Estimated Parameters J and Q
Project Tasks
The main project tasks are summarized as
Dynamic modeling of EHPV.
Off-line learning of input-output mapping for both forward and reverse flow modes
On-line learning of input-output mapping for both forward and reverse flow modes
Controller robustness
Performance evaluation
Triangular Activation Function
Test-bed
Hydraulic Schematic
General NLPN Structure
Reverse Input-Output Mapping
Forward Input-Output Mapping
EHPV
Pressure Sensors
Least Squares Estimation of Weights:
Swing
Arm
Orifice Flowmeter Calibration Map
Reverse Output-Input Mapping
Forward Output-Input Mapping
Orifice Flowmeter
Position
Sensor
Future Work
Controller robustness and performance is to be evaluated when using the EHPV in different combinations for metering modes.
Load
Needle Valve
Hydraulic
Piston
Collaborators
Fixed Arm
Temperature Sensor
J.D. Huggins
Sponsors: HUSCO International and FPMC Center
Advisors:
Dr. Nader Sadegh, Dr. Wayne Book
website:
Spring 2005