Overview

Configuration and reconfiguration as well as scheduling and rescheduling are central tasks for flexible and adaptable products and production processes. Existing general frameworks, such as Answer Set Programming (ASP) or Constraint Problem Solving (CSP), fail at finding efficient solutions for many large-scale, real world problems. One of the main reasons is that such methods only use problem independent built-in heuristics but do not exploit problem specific heuristics. However, general problem solving frameworks provide powerful problem representation capabilities.

In contrast to general strategies, heuristic search approaches use problem specific heuristics in order to traverse the search space more efficiently. However, these techniques are not available in general problem solving frameworks which employ high order logic as efficient problem representation language (i.e. ASP). For applying heuristic search, the challenge is clearly to discover effective problem specific heuristics for each particular problem of interest. Remarkably, human experts are almost always able to produce such heuristics based on intuition and experience which achieve reasonable results for real world problem instances. Based on these cognitive capabilities to generate heuristics, humans are able to conquer the enormous size of the search space in order to find satisfying solutions. Yet, the human process of producing effective heuristics is rather an art than engineering and typically very time consuming.

In order to tackle the above mentioned challenges the main innovation of the proposed project is the development of new methods for the efficient generation of heuristics in the domains of (re)configuration and (re)scheduling. To this end, project "Heuristic Intelligence" (HINT) chooses an interdisciplinary approach by identifying heuristics in human intelligence and exploiting them in the arena of artificial intelligence. In particular, the human heuristics will be combined and adapted by evolutionary algorithms. These heuristics will be used by new problem solving algorithms which combine the advantages of heuristic search techniques with those of higher order knowledge representation languages provided by general problem solving frameworks. As a result, the innovative methods aim at solving configuration and scheduling problem instances which cannot be solved by current search techniques. The overall performance of the developed methods will be measured for a set of large-scale problems, defined by industry partners. As main show cases we will employ hard problems of Siemens and Infineon such as the (re)configuration of complex technical systems (e.g. railway-safety systems) and dispatching rules for complex semi-conductor production lines. Thus, the results of HINT will significantly reduce the effort for generating problem specific heuristics and the search costs for optimization. Consequently, more efficient production processes and optimized products can be realized including their adaption to changed context conditions.