Embedded Implementation of Rainflow-Counting for On-Line Predictive Maintenance

Predictive maintenance can reduce unscheduled downtime and enhance the efficiency and productivity of industrial operations.Data collected on-plant is usually transmitted to local information technology (IT) systems and/or third party cloud services for storing and off-line processing, but if a reliable transmission channel is unavailable, local data processing is an attractive option.This work presents an on-line predictive maintenance system based on an embedded, real-time implementation of Miner's rule and the rainflow-counting algorithm. Memory and computing power limitations due to the use of an embedded system are taken into account, and their effects on the accuracy of the prediction are evaluated by comparing the embedded results with those of a golden standard, represented by a PC equipped with MATLAB and comparatively limitless resources. Guidelines about viable trade-offs between accuracy and computing resource requirements are extracted from simulations and experiments.