NEURAL NETWORK APPLICATIONS IN MECHATRONICS

摘要

Motion Control (MC) systems typically consist of a power electronicrnconverter, a motor and a controller, and are therefore non-linear, non-stationary andrndiscontinuous. Feed-forward neural networks (NNs) with supervised learning can bernused as non-linear identifiers and adaptive controllers of such MC systems. Moreover,rnthe MC system is often part of a production process where an unexpected outage due to arnmotor fault is expensive in terms of lost production. Neural Networks with unsupervisedrnlearning, also called Self-Organising-Maps, can be used to continuously analyse thernmotor current signature for early detection of incipient faults and thus enable plannedrnmaintenance shut downs, and constrain further damage to the motor. This paperrndiscusses both the application of neural networks (neuro-controllers) to identify andrncontrol ac motors, as well as the application of a neural network-based on-line ac motorrnwinding fault detection scheme.