Their simplicity and low implementation costs are extremely attractive.
However, there are situations in which conventional controllers are not robust enough to cope with the demands from the plant and its environment. In these situations an alternative control technique is required.
Rule based systems [7] require a detailed knowledge of the plant, its environment, and possible operating problems. In a number of engineering environments it is very difficult, if not impossible, to pre-determine the possible modes of system degradation that would result in a need for control adaptation.
A form of continued system evaluation is required in such situations to analyse the environment and instigate the required control counter measures.
Neural network based control provides the capability of learning the dynamics of a complex non-linear system. Fuzzy control can emulate experienced operators in controlling a complex nonlinear system while adapting with the use of self-organizing algorithms. The key difference between the two is the ability of the self-organizing fuzzy logic controller to function without a definitive process model. When applied to the water hydraulic actuation system, which will have constantly changing dynamics due to fouling, leakage, backlash, etc., it is the ability of the fuzzy controller to act without prior and accurate knowledge of the system that makes it attractive.
The SOFLC architecture used for the water hydraulic actuation system [8] differs from conventional SOFLC implementations in that it has a small and fix number of rules which are altered by adjusting the centers and widths of the rules. The SOFLC for this application has 12 adaptable rules and also configured for use with a single-output (actuator displacement), multi-input (valve control signals) system.
An initial study of the SOFLC was carried out within the MATLAB programming environment using Simulink modelling software. The Simulink software was used to produce a detailed model of the water hydraulic system incorporating such non-linear effects as slip-stick friction during piston motion, backlash within the gearing, valve and actuator leakage as a function of the load pressures across the components, and nozzle loss effects within the valves [9].
Figure 4 shows the top level structure of the Simulink programme modelling the hydraulic actuation system, SOFLC, and PWM functions. Inputs to the SOFLC function are the simulation time and the present state of the model in the form of angular displacement. The SOFLC function is used to call the various functions, implement the control system and activate the self-learning algorithms.
6. CONTROL SYSTEM EVALUATION USING COMPUTER SIMULATION
A series of stepped input demands were used to train and evaluate the SOFLC's ability to control the simulation of the water hydraulic system.
The self-learning mechanism in the control system was trained using repeat cycles (epochs) of an 180° stepped demand in actuator angle. Figure 5 shows the dynamic response of the controlled system after five repeat runs and it can be seen that by the fifth epoch there has been a considerable improvement in controller performance. The speed of system response has become slightly faster and oscillation around the steady state position is very small, Figure 6 shows the squared error between the demanded and achieved actuator angle after each of the training cycles and it can be seen that it has become considerably smaller by the fifth epoch.
The robustness and accuracy of the fuzzy logic control system when applied to the water hydraulic actuation system was analysed using a number of different failure modes: internal leakage across the control valves, internal leakage across the actuator, internal flow disruption, and a combination of each failure mode. In each case the control system successfully adapted the fuzzy rule set to account for the changing state of the water hydraulic system. Figure 7 shows the response of the system following a demand for the actuator to move from 0｡?o 20｡?nd then to -60｡?fter 8 seconds and with a 3mm leakage path introduced across the actuator piston seals. This level of leakage was found to be the upper leakage limit for actuator stability with the self-learning mechanism disabled. The response of the system steadily improves after ten epochs. The training algorithms reduce steady state error at 20｡?ut still has an oscillatory response about the -60｡?osition as shown by the angle squared error, Figure 8.
The standard deviation of the response from 8.5 seconds to 15 seconds gives an indication to the magnitude of the oscillation about the -60｡?osition. After the first epoch the standard deviation was 4.7 and after the fifth epoch the standard deviation was 3.8, indicating a gradual reduction in oscillation due to training.
In an actual 'real-life' application, system degradation through leakage more often than not will be gradual and the sudden change simulated represents an extreme case. The controller will reorganise its fuzzy rules continuously to gradually account for incremental system changes.
Changes in the set apex positions during the running of the simulation indicated a relatively high degree of change to each of the fuzzy rules and hence a high degree of self-organization required for efficient control.
A conventional PID controller was also implemented and analysed to determine its effectiveness. After a number of hours spent tuning the three-term controller, the best response obtained still remained oscillatory and had a significant steady state error.
The performance of the SOFLC system with the simulated plant was sufficiently encouraging to implement the control algorithms on the actual water hydraulic test rig.
7. CONTROL EVALUATION OF WATER HYDRAULIC TEST RIG
The initial evaluation of the water hydraulic actuation system control was carried out with the self-organizing algorithms switched off and the control system acting as a Fuzzy Logic Controller (FLC) using the fuzzy sets generated by the training process. This was due to software problems relating to the interfacing of the MATLAB and LabVIEW processing environments and the resulting sampling errors.
The National Instruments LabVIEW programming environment was used to run the control software and communicate with the water hydraulic hardware system. To enable the FLC to be used to control the actuator, the MATLAB M-files were translated to code compatible with LabVIEW.
The LabVIEW code used to communicate with the water hydraulic system and instigate the controlling action was a combination of LabVIEW 'G' code and MATLAB nodes imbedded in LabVIEW. Figure 8 shows the LabVIEW FLC architecture. The input interface converts the voltage (Vact) generated by the actuator output potentiometer into a scaled angular displacement (θact) and sets the desired angular displacement (θDEMAND).
The PWM function converts the proportional output of the FLC system (Un) into the logic input required by the water hydraulic system.
