Optimal sensor allocation by integrating causal models and set-covering algorithms

Massive amounts of data are generated in Distributed Sensor Networks (DSNs), posing challenges to effective and efficient detection of system abnormality through data analysis. This article proposes a new method for optimal sensor allocation in a DSN with the objective of timely detection of the abnormalities in a underlying physical system. This method involves two steps: first, a Bayesian Network (BN) is built to represent the causal relationships among the physical variables in the system; second, an integrated algorithm by combining the BN and a set-covering algorithm is developed to determine which physical variables should be sensed, in order to minimize the total sensing cost as well as satisfy a prescribed detectability requirement. Case studies are performed on a hot forming process and a large-scale cap alignment process, showing that the developed algorithm satisfies both the cost and detectability requirements.