Few strategies to improve sensor fusion

Sensor fusion is a widely used strategy where the system-variables are unobservable. In this work a (back-propagation neural network) BPNN based sensor fusion model has been developed for Tool-Condition-Monitoring (TCM) in flexible manufacturing applications (as a case study) and different performance improvement strategies have been incorporated with notes on generalized applicability to other sensor fusion models. Novel strategies like applicability of signal-level segmentation for temporal registration, effect of number and quality of datasets, usage of cross-correlation indices for sensor-selection, featurespace filtering, outlier-removal, estimation-space filtering etc. have been proposed. Statistical dataset-generation for better sensor-fusion modeling, and applicability of recurrent-type sensor fusion models have been introduced. All the propositions are verified with results obtained from the TCM sensor-fusion model and cost-effective peer-performing sensor collection is suggested.