Annual Review in Automatic Programming最新文献

This paper describes the application of a real time fuzzy controller to a column flotation pilot plant. The objective of this controller is to regulate three important process variables: air holdup, level of collection zone/froth zone interface in the column and bias water flow rate that are related with the product quality (concentrate grade) and recovery. The fuzzy controller was developed based on heuristic rules and it was tuned by a trial and error approach. The controller is coupled with a real-time graphics interface providing an easy process monitoring, with on line set point changes of all the controlled variables.

Learning control systems are expected to have several advantages over conventional approaches when dealing with complex, high-dimensional processes. One example is the task of controlling grasping operations of a multifingered, mul-tijoined robot gripper, which has been designed and implemented at our robotics lab (the Darmstadt-Hand). The Advanced Gripper Control with Learning Algorithms-AGRICOLA- presented in this paper is able to maintain a stable grasp even if disturbances are applied. Also it works for objects of different sizes for which the grasping has not been learned. Compared to the conventional stiffness approach the performance of the learning system is equal but the design is much easier, since less knowledge about the gripper-hardware has to be taken into account. The main part of the learning control loop is an associative memory storing the grasping behaviour as determined by the choice of an objective function.

The development of large-scale software is generally organized in the form of a project. Intensive management facilitates such development. Therefore, the key to the success of a project is to develop technology which strongly supports project management. This paper describes the requirements for a system which can support project management of software development. In particular, it focuses on the following points: the activities and the resources (person resources and non-person resources such as tools and machines) which offer support must be specified for a software development project. And the relationship between activities and resources must be established. This paper proposes a process model which meets the requirements. This paper also uses an example to show the behavior of the model and assesses the model's usefulness.

The contribution of this paper is twofold. First, we focus on the application of a particular NARMAX (nonlinear ARMAX) model representation based on local models for adaptive decoupling. Second, in order to improve the robustness of the adaptive control algorithm we introduce a diagonal PI-controller in parallel with the adaptive decoupler. These controllers are separated in the frequency domain, such that the decoupler and PI-controller takes care of control actions at higher and lower frequencies, respectively. The parallel control structure supports incremental control design, in the sense that improved process knowledge is used to successively upgrade control performance. The concept is illustrated by a semi-realistic simulation example.

Human beings are capable of learning to manually control complex nonlinear dynamical systems. It is well known, however, that humans have great difficulty in articulating the rules underlying their skilled behaviour. This paper focusses on the automatic machine induction of control rules from past records of skilled human behaviour. The aim of this endeavour is to install the induced rules as an automatic control programit is anticipated that this will lead to more consistent and reliable control performance.

The approach we study is based on the automatic induction of production rules from examples. The algorithms used are a product of the machine learning sub-field of artificial intelligence research.

We present experimental results describing induction of executable models of skilled human control behaviour. Experiments were performed on physical laboratory apparatus.

The cybernetic interface through which users can communicate with computers “as we may think” is the dream of human-computer interactions. Aiming at interfaces where machines adapt themselves to users' intention instead of users' adaptation to machines, we have been applying neural networks to realize electromyographic(EMG)-controlled prosthetic members—a historical heritage of the cybernetics. This paper proposes that EMG patterns can be analyzed and classified by neural networks. Through experiments and simulations, it is demonstrated that recognition of not only finger movement and torque but also joint angles in dynamic finger movement, based on EMG patterns, can be successfully accomplished.

The increasing complexity of industrial control systems and industrial processes makes it necessary to frequent use tools and techniques for reliability and safety analysis, particularly to early detection and localization of faults. By applying conventional approaches to diagnose faults via both dynamic process and signal models and parameters estimation and knowledge processing the inherent redundances can be used to make it more effective and to detect faults earlier. But in many cases, it is not enough to design the unique fault diagnosis procedure, but it is also essential to provide its flexible adaptation to changes of environment of the controlled system, including structures, parameters, tasks of the operators, their (changing) knowledge, cognitive capabilities etc. The major efforts have to be made to develop new generation of knowledge-based fault detection (FD) systems with both analytical and heuristical knowledge to be easily adapted to the changes in the system's environment. The paper desribes integrated approach to the “homogeneous” introducing the human-like thinking neuron schemes in all the stages of the model based fault diagnosis. Several examples are discussed.

The paper describes a self-learning control system for a mobile robot. Based on sensor information the control system has to provide a steering signal in such a way that collisions are avoided. Since in our case no «examples« are available, the system learns on the basis of an external reinforcement signal which is negative in case of a collision and zero otherwise. We describe the adaptive algorithm which is used for a discrete coding of the state space, and the adaptive algorithm for learning the correct mapping from the input (state) vector to the output (steering) signal.

The paper concentrates on a Petri-net approach to fault diagnosis. The framework which is proposed allows the preconditions, logical links and mutual influences of the activities and states of functions and equipments to be naturally described, and a simulation-based fault isolation reasoning to be implemented. It is also a basis for a multi-model-based approach, since Petri nets and behavioral models can be easily connected. The framework is illustrated with the example of a pneumatic suspension system.