CEmACS Project
home | introduction | participants | publications | links | internal
Information Society Technologies Sixth Framework Programme

Control Design

The solution to complex nonlinear multivariable control problems, like integrated chassis control and collision avoidance, requires the combination of the advantages of a number of different control approaches. This is due to the fact that most of the established approaches in control theory only focus on certain aspects of the problem and thus will not address it completely. For instance, nonlinear control methods usually offer good solutions for global stabilisation and adaptation but they only poorly address robustness and local performance issues. On the other hand classical frequency domain multivariable control techniques achieve excellent local performance and robustness with respect to unmodelled dynamics but nonlinear off-equilibrium global stability and constraint handling are not addressed. Consequently, the focus of the control part of the project will be on the four complementary techniques described next.

Classical Multivariable Control Analysis and Design.

An appealing small-signal frequency-domain framework that holds great promise for the integrated control of complex dynamical systems such as embedded automotive systems is known as Individual Channel Analysis and Design. The ICAD methodology provides a framework within which control concepts and methods from classical control engineering - Nyquist-Bode plots, gain and phase robustness margins - may be rigorously applied to strongly cross-coupled multi-input multi-output (MIMO) systems. It thus preserves continuity with industrial practice while providing much needed analytic insight of a graphical nature into both the uncontrolled and controlled dynamics of growing complex systems. How this works within ICAD is that, without loss of structural information, an m-input m-output feedback control problem can be decomposed into m individual single-input single-output (SISO) channels; that is each controlled output is naturally paired to a reference or command input. Each individual channel thereby has its own SISO performance specification since the customer's requirements are most clearly stated in terms of the dynamical response of an output to its associated input. ICAD is first and foremost an analysis capability of the potential and limitations for the embedded control of complex systems. Moreover, ICAD, as well as guiding control systems design for complex systems, can integrate and assess the robustness and performance of other subsystem designs (Classical, Explicit Constrained Optimal Control, H-Infinity, LQR) within the overall composite system. We have a range of previous applications of ICAD to the integrated control of large complex systems, including aerospace (helicopter and fixed wing) control and distributed embedded generation power systems. The aerospace work is particularly relevant here in the context of embedded automotive control. This is because there are similar control requirements in the face of higher levels of systems integration (flight control systems, engines, utilities), stronger cross-coupling, many more inputs/outputs (control surfaces), actuator redundancy; and not least, more demanding performance requirements in the form of tighter control (manoeuvreability). While, in principle, some of the theoretical issues associated with ICAD have already been tackled, we can identify a number of key theoretical developments which must take place for exploitation of the approach within complex automotive applications. Thus, further sub-objectives in line with the theoretical and application-orientated objectives of the project are identified:
  1. Which outputs are best controlled by which inputs? Develop systematic methods for assigning input-output pairs based upon multivariable structure elucidated by ICAD.
  2. Investigate issues of redundancy in actuators and actuator failure compensation.
  3. Evaluate local controllers of all the partners for application studies.

Hybrid Control Systems

Vehicle control systems may have the following requirements:
  1. They must be designed to operate robustly in environments that are subject to abrupt change, \item they should provide high performance over a wide range of operationg conditions and in the presence of severe actuator and state constraints and,
  2. they should be robust to component and subsystem failures.
These requirements imply the use of hybrid control techniques combining continuous and discontinuous control. The area has received much attention form the academic community over the past decade. Within the project we will focus our investigation of hybrid control on two topics:

Multiple Controllers are used for stabilising systems with structural changes (caused by internal failures or external events) and severe nonlinearities and for improving performance in adaptive systems(see WP3.4). Important examples are graceful degradation in the event of component failures and fast response of the control system to abrupt changes such as varying road conditions or wheels loosing contact to the ground in the event of rollover. In such events the control strategy will be changed by switching between different controllers or controller parameters. The two main issues arising in multiple controller approaches are stability of the overall system and the handling of transients which may be caused by the structural changes. Whith respect to the stability issue we will exploit recent approaches to control design and realisation techniques by adopting them to the automotive control setting. With respect to transient handling there are reference trajectory resetting techniques. In addition to this, the fact that multiple actuators are available within complex control problems (overactuation) introduces questions of how to combine and coordinate multiple controllers in an optimal way. Some key issues are:

  1. Investigate how to utilize overactuation to achieve maximal robustness against time-variations (for example due to road conditions and other traffic).
  2. Study how to design controllers that recover from undesired states and integrate them with a stationary controller for normal operation near a desired equilibrium (with all four wheels utilized).
The items are highly relevant for the study of car dynamics. Based on extending recent results, the overall objective is to combine local and global controllers to achieve globally acceptable performance.

Constraint handling and anti-windup Recent studies have shown that the solution to a wide class of constrained dynamic optimization problems can be explicitly computed and represented as a piecewise linear state feedback. This can be exploited to implement constrained optimal feedback control simply by evaluating a piecewise linear mapping, or through lookup tables. For automotive embedded systems where real-time optimization is generally avoided due to high demands for computations and software reliability issues, this is an interesting approach for multi-variable constrained control problems. Currently, such techniques have been tested in automotive applications. However, these methods are not straightforward to apply in complex problems with a large number of states and inputs, due to the rapid increase in complexity of the solution when the dimension of the problem increases. Also, the use of such methods for nonlinear problems is fairly undeveloped. We will extend recently developed ideas of explicit constrained control to be applied in more complex automotive applications. Control allocation is commonly used in advanced vehicle motion control systems in order to generate a total longitudinal force, lateral force, and yaw moment requested from a higher level controller, using a redundant set of actuators. This is useful in automotive vehicles for example in extreme maneuvers during collision avoidance using a redundant combination of steer-by-wire and brake-by-wire. One of the main objectives of the control allocation module is to explicitly handle constraints such that one is guaranteed to attain the requested total control force whenever possible. This is important from a safety point of view when the vehicle operates close to its physical limitations. Another great benefit of control allocation is that it allows simple reconfiguration in case of an actuator failure. Recent research in this direction, indicate that nonlinear control allocation can be implemented using some versions of optimizaton-based control that can be implemented with high reliability and low computational complexity, and is therefore suited for embedded automotive real-time control systems. Actuator position and rate constraints leads to a hybrid behaviour of such control allocation solutions, since the strategy will essentially switch when different combinations of actuators saturate as a response to a time-varying total force request from the higher level control system. We will extend the results \cite{Joh04a} and provide reliable numerical implementation for use in automotive applications.

Multivariable Control Systems with time delay

Loop shaping with delay robustness The literature contains numerous criteria for robustness analysis with respect to time-varying delays in a feedback system. However most of the criteria are based on advanced optimization and are difficult to use in controller design. Within the project we will develop improved frequency domain design criteria that also take into account a bound on delay variation rate.

Linear delay compensation schemes Sometimes it is possible to put a time stamp on each measurement signal that is sent over the network, so that the network induced delay can be measured on arrival. Such information can be used to improve control performance considerably. The objective in this project would be to introduce a linear delay compensator in analogy with anti-windup schemes. Intuitively, a measurement that arrives late should have less impact on the control action than one that is very fresh. Within WP3 this intuititive rule will be formalised and its consequences will be analysed.

Non-linear compensation schemes for automotive applications Nonlinear control loops in automotive applications offer very concrete challenges where measurement delays can be crucial. One example is in control of ABS brakes, where maximal braking force requires very fast feedback in order to prevent wheel-lock. Another example is in the stabilization of vehicle dynamics, where again fast feedback is crucial under safety-critical circumstances. Consequently, an objective of WP3 will be to quantify the delay margin in order to optimize performance without sacrificing safety.

Nonlinear and adaptive control

Automotive vehicles are characterized by highly non-linear, uncertain and time-varying dynamics. This is due to inherent dynamic properties such as Coriolis forces that are significant in extreme maneuvers, but more importantly it is due to the complex phenomena of tyre/road friction. In addition to being characterized by strongly nonlinear characteristics, friction depends on environmental properties such as the road conditions, and vehicle properties such as tyre type, wear and pressure. Any advanced automotive control system, such as collision avoidance or active body control, must therefore either address this directly, or rely on lower-level decentralized control systems that hide the nonlinearity and uncertainty from the perspective of the high-level control function. The issue of adaptive and nonlinear control must therefore be addressed explicitly at some level in the control hierarchy. The current state-of-the-art is dominated by ad hoc solutions that are expensive to maintain and requires considerable effort to redesign for a new vehicle model or series. Within the present project we will focus on model-based nonlinear and adaptive control alternatives, which has the benefit that the control software is generic in the sense that it is parameterized in terms of vehicle parameter, such that only a small effort is required for redesign. Particular focus will be given to improving transient performance in adaptive control via switched control. Adaptive control systems will adapt their paramters to any changes in the dynamics or characteristics of the system under control or its environment. A major challenge for any adaptive control system is to achieve fast response to rapid, or instantaneous, changes in the system under control. The relevant concepts to understand this is uncertainty and information, as the decision to change the parameters of the controller will be based on uncertain information, due to model structure errors, measurement errors and unknown external disturbances acting on the system. In order to reduce sensitivity to such uncertainty, any practical adaptive control systems will include a large degree of filtering and low adaptation gain that will lead to reduced speed of response due to the phase lags being introduced by the filters and the low gain of the adaptation. In order to improve on the transient performance in such cases, it has been suggested to switch parameter estimates within the adaptive controller based on a model-based monitoring of the system under control and the references therein. Yet, the fundamental issues related to robustness to uncertain and missing information are still far from fully understood, and within the project we will seek to provide new insight on how to design for high transient performance in nonlinear adaptive control systems. Fault-tolerant control is a closely related issue that. The methods will also enable the implementation of a systematic approach to fault-tolerant control and automatic reconfigurability to accomodate sudden changes in the vehicle and its environment, and failures in sensors, actuators, or communication. This line of work may benefit strongly from the development of nonlinear vehicle observers, as such observers can be used as an integral part of the system performance monitoring, although alternative estimation methods are also available.

Vehicle Active Safety

Integrated Chassis Control

Control Design

Vehicle State Observation

DaimlerChrysler AG Hamilton Institute University of Glasgow Lund University SINTEF