Metaheuristic approaches for energy efficient production optimisation in manufacturing facilities

Abstract

The science of optimization has been used since many centuries ago, probably since the great Greek mathematicians, physicists and philosophers started the quest to understand the universe. First it was used as a pure mathematical tool, a method for finding solutions to geometrical problems; but soon its usefulness spread from simple problems to the most complex human engineering works. However, some fields of Engineering optimization still remain a “dark art” when applied to very complex processes. In those instances the use of exact optimization methods is not an option, due to its extreme complexity, involving hundreds or thousands of variables, making it a problem far ahead of the capabilities than even modern computers have. This complexity issue will be further compounded and even more difficult to solve when we impose new objectives and constraints, in particular when they are seen as secondary by the human planners. This is the situation that our research found when investigating the inclusion of energy efficiency as a new objective in manufacturing facilities. In this type of industry, every company is nowadays fighting to survive against stern, world-wide, continuously increasing competition. Multiple objectives need to be optimized (decrease the cost, increase production, minimize time-to-market, just to name a few), with multiple constraints that force boundaries in how to achieve those goals. In this scenario, the optimization of energy efficiency is usually ignored by planners who often prioritize more critical constraints, like production targets and delivery dates. When we step into this level of complexity we find that exact methods are unable to give timely solutions, or even produce any solution at all. In such instances other approaches need to be used. In particular, and since a few decades ago, the use of metaheuristic optimization has proved, again and again, to be of extremely high value in otherwise intractable problems. The first landmark on the use of metaheuristics for optimization was, arguably, the work done in the 60’s. It was in that decade when the work on Evolutionary Strategies by I. Rechenberg and H.P. Schwefel, the invention of Genetic algorithms by J. Holland and the Evolutionary Programming by L. J. Fogel et al. gave the green light for the unstoppable use ofthese techniques in multiple fields of Mathematics, Physics and Engineering. But real-world engineering problems are too complex even for these metaheuristic methods working alone. They lack the cognitive abilities of humans: the adaptability, reactivity to unplanned situations, learning capabilities. Humans bring flexibility to the search of the optimal solutions by applying expert knowledge. A joint cognitive system using the best of both machine algorithms and human knowledge can take away part of the complexity of the problem, allowing the human to tune and guide the search for the optimum solution. This joint system of machine and human will in turn bring new challenges to the optimization algorithms, challenges that have been found in the past to be an issue for the acceptance of machine solutions by the human operators. These challenges can be summarized in two main categories: the interpretability of the results by the human user, which can lead to the rejection of good solutions for other, more intuitive ones; and directability, or the ability of the user to incorporate his expert knowledge into the solution. The present thesis will shed light on the use of meta-heuristic techniques as part of a human-machine joint cognitive system for the optimization of resources in realworld manufacturing scheduling, with the main focus on how to integrate energy efficiency in the set of objectives to optimize. This work will review the current state of the art in metaheuristics optimization, scheduling in advanced manufacturing facilities, and the main techniques that will form the base of the suggested architecture. After the initial chapters, we will show the application of these techniques to real world problems. These examples are taken from the experience of the author in several European companies in the sectors of semiconductors, automotive and pharmaceutical. In summary, this dissertation takes a technical step towards the adoption of metaheuristic techniques as part of an integrated, interdisciplinary system where human intelligence and machine power are able to work together to find better solutions for real-life problems.