The Transferable Belief Model for Failure Prediction in Wireless Sensor Networks

In recent years, with the advent of smart sensors, the design and deployment of Wireless Sensor Networks (WSN) have become an active area of research. Sensor nodes are typically deployed to sense, measure and gather parameters/conditions such as temperature, pressure, humidity, vibration, etc. Then, the Industrial Internet of Things (IIoT) applies WSN to improve the productivity and efficiency of existing and prospective manufacturing industries. The measured values are transmitted to a data fusion centre for further processing. However, the set of learning data on which the processing must be based for decision making is not always complete due to some failures such as security attack, packet leak, collision, interference, congestion, channel fading, devices errors, incorrect deployment, synchronization issues, environmental blockages or unknown errors. The management of missing data is a common and widespread problem in smart factory using WSN. It is a more common topic in monitoring systems. The matter is to define a Transferable Belief Model (TBM) for failures prediction in WSN. However, most of the existing prediction methods are based on deterministic probabilistic approaches. These approaches are not adapted for representation of imprecise knowledge common in WSN. To remedy these issues, the ultimate goal of this paper is to propose a non-deterministic approach for building a predictive belief function from statistical data for decision making. The fundamental point of methodology is to transfer uncertainties by transforming masses function into densities function, and then belief functions, plausibility functions and commonality functions into integrals of these densities function. The efficiency of the approach is demonstrated using a simulated WSN problem and Monte Carlo simulation. The simulation’s results have shown that the proposed approach achieves satisfactory performance compared to data imputation methods.