Improving Majority-guided Fuzzy Information Fusion for Industry 4.0 Condition Monitoring

The progression from industrial systems to Industry 4.0 leads to systems with an increasing number of sensors and communication links producing large amounts of data. This results in increased engineering efforts for parameterisation and commissioning. Industrial applications additionally place high demands on algorithms. Data are often noisy, imprecise, or incomplete. Particularly in condition monitoring applications fuzzy information fusion systems have proven themselves. Drawbacks of state-of-the-art applications of majority-guided fuzzy information fusion are that they are highly computational complex (i. e. scale poorly) and require expert's knowledge in their implementation. Expert's knowledge is required for designing the fusion topology as well as for modelling information uncertainty with possibility distributions. This knowledge is especially in large industrial applications either not available or only costly to acquire. This contribution proposes an improved implementation of the search for majority observations with reduced computational complexity. In addition, engineering efforts are reduced by automatically grouping information sources and by determining the possibility distributions on training data. Both, grouping sources for fusion topology inference and distribution determination, rely on a correlation-based redundancy metric.