Fuzzy Logic and Neural Networks for Insulation Fault Diagnosis in Construction Robots Drives

In building industry reliable uninterrupted power supply of construction robots drives is of particular importance, which is largely determined by reliable trouble-free operation of generating equipment. According to statistics, the majority of electricity in the world is produced by hydro and turbine generators, which are low-speed or high-speed synchronous machines. The urgent problem is the development of methods for non-destructive testing and insulation monitoring of synchronous machines. The main method of assessing the real technical condition is insulation control through the analysis of electrical discharge activity (EDA). This method allows detecting defects at an early stage of their development. The actual problem is the development of automatic technical state diagnosis methods for insulation by the EDA parameter. The main parameters that are evaluated in the analysis of EDA is the shape and amplitude of the discharge phenomena. The article proposes a method for determining the discharge phenomena form, based on a neural network classification model. It used two-layer network of direct signal transmission trained by Levenberg-Marquardt algorithm. A method for determining the degree of defect development based on a neuro-fuzzy diagnostic model, differing by a joint analysis of the shape, amplitude and repetition rate of the pulses of a discharge phenomenon, which allow to determine the degree of defect development by relating it to one of the classes of diagnoses is proposed. © 2019 The Authors. Published by Budapest University of Technology and Economics & Diamond Congress Ltd. Peer-review under responsibility of the scientific committee of the Creative Construction Conference 2019.