Condition monitoring of a mechanical indexing system with artificial neural networks

The paper presents a procedure based on Artificial Neural Networks (ANNs) for the condition monitoring of a mechanical indexing system. Since rotating tables with indexing cam devices work in severe dynamic conditions (high speed and load), backlashes, compliance, and faults cause vibrations that affect the quality of products. An ANN is used to classify the system condition among a number of possible situations with backlashes, compliances, and damages, receiving as input an opportune feature extracted by the table acceleration signal. FFT and PSD are compared in order to identify which performs better for extracting features from signals. The diagnostic procedure was calibrated and tested on data generated by elasto-dynamic models. Several tests with different parameters of the procedure were carried on in order to identify the optimal configuration of the net. As last step, the procedure was tested experimentally on a system with variable backlash and structural defects of controlled type and extent, obtaining acceptable results.