Synthetic Dataset for Induction Motor Broken Rotor Bar Analysis

The synthetic dataset was developed in the Power Electronics and Electrical Drive Laboratory (Laboratório de Eletrônica de Potência e Acionamento Elétrico – LEPAC, in portuguese) at the Federal University of Espírito Santo (UFES), Brazil. The dataset was generated from a computer simulation of a three-phase squirrel cage induction motor model with the insertion of faults in the rotor, specifically broken bars. The induction motor mathematical model with broken bar fault is based on reference [1].

Real data from 26 induction motors are used as input parameters the adopted computational model, making it possible to generate a dataset containing 390 different simulated scenarios. For each induction motor, 5 fault scenarios are simulated, varying the number of broken bars between 0 (healthy rotor) and 4 broken bars. Furthermore, each scenario was simulated under 3 load levels (50%, 75% and 100% of nominal torque). The data from the 26 induction motors is available in a PDF provided together with the dataset in the compressed file. 

The files are saved in .mat format with the following nomenclature (MX.rYb.torqueZ.mat) where X refers to the induction motor identification (X = 1:26), Y refers to the number of broken bars (Y = 0:4) and Z indicates the percentage loading of the induction motor (Z= 50, 75, 100). Therefore, the nomenclature M3.r2b.torque50.mat, for example, corresponds to the data of induction motor 3 with 2 broken bars and 50% of nominal load torque. 

All simulations were carried out according to the following methodology. The induction motor is started with no load until it reaches steady state and, subsequently, the load torque for each simulation is set. In all scenarios for all induction motors, the steady state with load is reached after 50 seconds of simulation. The data saved in the .mat files are the three-phase motor currents and the simulation time step at a sampled frequency of 8kHz.