Anthony J. Grichnik

Composite goal methods for transportation network optimization

We review the state of art in multi-goal real-world supply chain optimization.From the review, we identify four generic strategies.We implement and test seven instances of these strategies.In pursuit of clarity, we formally describe the implemented strategies.We highlight the poor performance of the most adopted method in the literature. Lately the topic of multi-objective transportation network optimization has received increased attention in the research literature. The use of multi-objective transportation network optimization has led to a more accurate and realistic solution in comparison to scenarios where only a single objective is considered. The aim of this work is to identify the most promising multi-objective optimization technique for use in solving real-world transportation network optimization problems. We start by reviewing the state of the art in multi-objective optimization and identify four generic strategies, which are referred to as goal synthesis, superposition, incremental solving and exploration. We then implement and test seven instances of these four strategies. From the literature, the preferred approach lies in the combination of goals into a single optimization model (a.k.a. goal synthesis). Despite its popularity as a multi-objective optimization method and in the context of our problem domain, the experimental results achieved by this method resulted in poor quality solutions when compared to the other strategies. This was particularly noticeable in the case of the superposition method which significantly outperformed goal synthesis.

A Soft Computing Approach to Controlling Emissions Under Imperfect Sensors

NOx emissions from commercial equipment vehicles can be controlled by the introduction of cleansing agents into output of the equipment. This process requires a monitoring of the output by sensors. Typically, these sensors can be of two types, physical and virtual, each of which can be imperfect in their readings. Physical sensors gradually deteriorate in time while virtual sensors have limit range of reliability. We provide an approach for fusing these imperfect sensor readings using the ordered weighted averaging (OWA) operator. The use of the OWA operator allows users to easily provide their preferred tradeoff between costly introductions of too much cleansing agent versus the possibility of penalization for emission violation. We also provide a methodology for determining the state of the physical sensor. We observe that the system described in this paper has been operationally implemented.

Manufacturing Virtual Sensors at Caterpillar, Inc.

Embedded software allows manufacturers to meet technological demands on their products more efficiently. The complexity of onboard electronic systems has grown to the point where embedded software is being used to replace or back up physical sensors in order to meet the engineering and system reliability requirements of Caterpillar and other manufacturers of large industrial equipment. This paper describes a process that allows Caterpillar to develop virtual sensors in a standard, repeatable manner that improves on conventional processes which are often inconsistent, inefficient, or both.