Scalable Offline Optimization of Industrial Wireless Sensor Networks

Sensor networks are increasingly used to control and monitor industrial and manufacturing processes. In this paper, we consider the problem of optimizing a cost function for wireless sensor networks of this kind under energy consumption constraints. We focus, in particular, on the problem of coverage optimization through scheduling. Following existing approaches, we use a mixed integer linear program formulation. We show how to use partitioning techniques to decompose the problem into separate subproblems, solved individually, overcoming the exponential complexity typical of integer linear programming, while minimizing the loss in optimality. In addition, we evaluate the achieved degree of optimality by computing relatively tight bounds with respect to the optimal solution. Finally, we employ simple but effective heuristics to further improve our solution. The results show that our procedure is very efficient and scalable, and is able to find solutions that are very close to optimal. These characteristics make our approach a perfect fit for large and fixed deployments of wireless sensors, typical in factory automation and industrial applications. To show the generality of the approach, we apply our methodology to three different models of varying complexity.

I. Introduction

Among embedded devices, wireless sensor networks (WSNs) are emerging as one of the most interesting innovations in terms of potential areas of application. They have an increasing impact in fields such as security, health care, disaster management, agricultural monitoring, and building automation. This paradigm is of particular interest in an industrial setting (witnessed by the WISA system from ABB [1]), because a dense deployment of sensors enables engineers to more accurately control and optimize process parameters in manufacturing and to more effectively manage logistics [2], [3]. The most relevant feature of a WSN is that it is a dynamic distributed system, in which complex tasks are performed through the coordinated action of a large number of small autonomous devices (nodes). A problem of paramount importance for a WSN is maximizing the lifetime of the network. The lifetime is typically bounded by the amount of energy stored in each node upon its deployment. This is true especially when the system is deployed in a remote or harsh environment and/or when maintenance and battery replacement is costly. For instance, shutting down a production line for battery replacement could have a significant impact on productivity. The objective is, therefore, to design distributed algorithms for data processing and resource management that attain an optimal tradeoff between functionality, robustness and energy consumption.