There is an alternative to the classical single block fuzzy control system. Consider the modular architecture shown in Figure 17.3b. Notice that the condition variables are not all handled simultaneously within a fuzzy block. Rather, the condition variables are manipulated in series based on domain knowledge indicating which input variables are most closely related. Using knowledge of the dynamics of the system to divide a process into separate control tasks allows adequate control with much fewer rules than required by the classical fuzzy controller architecture. This is a major contribution of the current chapter. If each input is described with three membership functions, the distributed controller in Figure 17.3b requires only 3² + 3² + 3² = 27 rules. If the inputs are not tightly coupled, the modular controller can achieve acceptable performance with a mere fraction of the number of rules required by a more traditional fuzzy control architecture.

Developing a simplified yet suitable architecture is often quite difficult. While it is desirable to generate a controller with a small number of rules, extreme care must be taken to preserve important relationships between the state variables. A careful balance between reducing the number of rules and achieving an accurate representation of the coupling in the system, accomplished through the careful development of modular units or fuzzy blocks, is needed to create an efficient and effective controller.

Designing the helicopter controller architecture is largely a process of trial and error tempered with a minimal amount of human expertise. After several design iterations, the configuration shown in Figure 17.4a through Figure 17.4d was adopted. The controller is composed of four sections, one for each control apparatus present in the aircraft. The four sections represent control of the longitudinal cyclic (δ_lon) (Figure 17.4a), lateral cyclic (δ_lat) (Figure 17.4b), pedals (δ_tr) (Figure 17.4c), and collective (δ_col) (Figure 17.4d). Note that in these figures the extension “_dot” represents a derivative with respect to time. Notice in the figure(s) that each control goal is managed through the interaction of several fuzzy blocks.

Figure 17.4a  Section of the fuzzy controller used for longitudinal cyclic control.

Figure 17.4b  Section of fuzzy controller used for lateral cyclic control.

Figure 17.4c  Section of the fuzzy controller used for pedal control.

Figure 17.4d  Section of fuzzy controller used for collective control.

Although the architecture described in the above figure(s) may seem imposing, if not intimidating, the resulting system is far simpler than a control system containing the number of rules that would be required in a traditional fuzzy architecture. In the above system, the longitudinal cyclic control motion is governed by the fuzzy block labeled Longitudinal Attitude. Given the error in pitch, q, and q_dot, the Longitudinal Attitude block infers a change in longitudinal cyclic. The Longitudinal Switch determines which of two control strategies is employed based on the error in forward velocity. If the forward velocity of the helicopter is far from the goal velocity, then the error in pitch is determined by the Longitudinal Acceleration block. Otherwise, the error in pitch is determined by the Longitudinal Hold block. Notice that the Longitudinal Acceleration block infers a desired pitch angle while the Longitudinal Hold block infers an error in pitch angle. While at first this arrangement might seem overly complex it is designed to model the decision process of a pilot. When a pilot is near the desired forward velocity, he or she achieves the velocity goal by making small adjustments to the pitch of the aircraft. When a pilot is far from the desired forward velocity, he or she will typically maintain a desired pitch attitude to accelerate or decelerate the helicopter. In both cases, the result is a function of the current airspeed, the error in forward velocity and the forward acceleration.

The task of achieving lateral cyclic control is split into two control branches (Figure 17.4b), both of which resemble the longitudinal section of the controller. The two branches represent the different functions of lateral cyclic at hover and at forward flight. At hover, the lateral cyclic is used to determine lateral velocity of the aircraft. In forward flight, however, the desired lateral velocity is almost always zero. In this case, the lateral cyclic is used to roll the aircraft into turns to achieve new headings. Notice in Figure 17.4b that the top branch of the lateral controller uses lateral velocities to achieve control at hover. The bottom branch of the lateral controller uses heading information to achieve heading control in forward flight.

Control of the tail rotor (Figure 17.4c) is also a function of the current forward airspeed. While it is not obvious from the figure, the rules in the Hover Heading block become ineffective as airspeed increases. The opposite is true for the Lateral Trim block. At hover, the Hover Heading block controls the tail rotor. In forward flight, the Lateral Trim block controls the tail rotor. Between hover and forward flight, the actions of the two control blocks are linearly combined. The purpose of the Lateral Trim block is to make very small adjustments to the tail rotor in forward flight to eliminate lateral motion, or “side-slip.”

Finally, collective control (Figure 17.4d) is a function of the current airspeed, error in climb rate, and climb acceleration of the helicopter. The collective control section is the simplest yet most effective section of the controller.

Fuzzy blocks or switches are used throughout the controller to provide specialized control for different flight regimes. An example is the Forward Flight/Hover Taxi switch in the lateral cyclic controller (Figure 17.4b). If the forward velocity of the helicopter is less than 6 m/s, then the result from the top section of the lateral controller is passed through the switch. If the forward velocity is greater than 12 m/s, the result of the bottom section of the lateral controller is passed through the switch. When forward velocity is between 6 m/s and 12 m/s, a linear combination of the results of the top and bottom sections of the lateral controller are provided as output from the switch. The fuzzy rules in the switches are predetermined, not manipulated by the search algorithm during rule discovery.