Approach towards sensor placement, selection and fusion for real-time condition monitoring of precision machines

Moving mechanical parts in a machine will inevitably generate vibration profiles reflecting its operating conditions. Vibration profile analysis is a useful tool for real-time condition monitoring to avoid loss of performance and unwanted machine downtime. In this paper, we propose and validate an approach for sensor placement, selection and fusion for continuous machine condition monitoring. The main idea is to use a minimal series of sensors mounted at key locations of a machine to measure and infer the actual vibration spectrum at a critical point where it is not suitable to mount a sensor. The locations for sensors' mountings which are subsequently used for vibration inference are identified based on sensitivity calibration at these locations moderated with normalized Fisher Information (NFI) associated with the measurement quality of the sensor at that location. Each of the identified sensor placement location is associated with one or more sensitive frequencies for which it ranks top in terms of the moderated sensitivities calibrated. A set of Radial Basis Function (RBF), each of them associated with a range of sensitive frequencies, is used to infer the vibration at the critical point for that frequency. The overall vibration spectrum of the critical point is then fused from these components. A comprehensive set of experimental results for validation of the proposed approach is provided in the paper.