Information Technology Reference
In-Depth Information
ii) a difficult implementation of the specification requirements; iii) the FIS-based methods are also
computationally less efficient; iv) when testing the more frequent fault conditions, all the developed
algorithms present an acceptable degree of reliability and robustness, but the crisp algorithm, which is
actually adherent to the specifications provided by the expert personnel of the train society, provides
the best guarantee of estimated speed values lying within acceptable limits; v) with respect the merely
crisp algorithm and to its improved version, where some parameters have been optimized through
GA, the soft computing-based algorithms provide less control over the internal parameters of the
estimator, which increase in number but loose in physical meaning (this is especially true for the
neural predictor, which has been applied as a black-box parametric estimator); vi) soft computing-
based procedures, and especially the ones which exploit neural networks, do not guarantee that a
particular input pattern (or series of input patterns) will lead to unacceptable velocity estimates.
Thus finally the rule-based algorithm tuned through GA has been preferred to all the other
approaches for the final implementation.
4.6 Optimisation of the logistic in an automatic warehouse of steel tubes
Optimisation of logistic is one of the fields where intelligent systems have most successfully
been applied. Colla et al. (Colla et al., 2010) tested several AI-based techniques for the
optimisation of products allocation in an automatic warehouse of steel tubes. The
warehouse has been designed to stock a large variety of typologies of steel tubes, differing
in quality of the steel, in the length as well as in the shape and dimensions of the section. As
soon as the tubes are produced, they are grouped in packs and automatically transferred to
a stocking area, where they are located in piles that must be composed by the same typology
of tubes. A non optimal allocation strategy can cause the available space in the warehouse
not to be fully exploited, such as, for instance, the case in which many short (i.e. composed
by a few packs) piles are present in place of a few higher ones. To this aim, firstly some Key
Performance Indicators (KPIs) have been defined in order to derive objective functions to be
optimised by the different allocation strategies. Afterwards, an optimisation problem has
been formulated, for which an analytical model of the problem was really too complex to
define and implement, due to the variability of the workload and to the interaction between
the automatic tube conveyors, as traffic control is only partially centralised (for instance
collision avoidance is managed at local level through suitable sensing and communicating
devices mounted on each conveyor). Traditional derivative-based optimisation models
cannot be applied, while GAs are a very suitable solution for the optimisation problem, as
they allow a decoupling between the problem formulation and the search procedure. The
destination of each tubes pack has been suitably codified in a chromosome and GAs have
been applied in order to minimise a fitness function obtained from a composition of the
above-defined KPIs. Different ways to aggregate the selected KPIs have been tested, from a
simple weighted sum up to a Fuzzy Inference System implementing a complex combination
according to rules derived from the knowledge of the technical personnel working on the
plant. However, this application is intrinsically a Multi-Objective Optimization (MOO)
problem, as the KPIs represent requirements that are often in contrast to each other. Any
kind of aggregation of the KPIs simplifies it to a Single Objective Optimization problem, but
surely the most suitable way to cope with this problem is by exploiting GA-based MOO
algorithms. The Strength Pareto Evolutionary Algorithm (Zitzler & Thiele 1999) has been
successfully applied to this problem and outperforms all the other approaches.
