Open AccessANN‐based method for parametric modelling and optimising efficiency, output power and material cost of BLDC motor
Author(s) -
Sadrossadat Sayed Alireza,
Rahmani Omid
Publication year - 2020
Publication title -
iet electric power applications
Language(s) - English
Resource type - Journals
ISSN - 1751-8679
DOI - 10.1049/iet-epa.2019.0686
Subject(s) - torque , parametric statistics , penalty method , computer science , artificial neural network , control theory (sociology) , magnet , metric (unit) , parametric model , mathematical optimization , control engineering , engineering , mathematics , artificial intelligence , mechanical engineering , statistics , physics , operations management , control (management) , thermodynamics
This study presents a new method for parametric modelling and optimisation of a permanent‐magnet brushless DC (BLDC) motor. We proposed an artificial neural network (ANN)‐based L P metric technique to combine and optimise different objective functions of a BLDC motor using ANN‐based models and compared with conventional optimisation methods with analytical models. To proceed with this optimisation problem, the L P function should be minimised. For applying constraints to this problem, a simple method called penalty factor is proposed, in which a penalty term was added to the L P function when the constraints are violated. We considered three goals in this optimisation: efficiency maximisation, speed maximisation and material cost minimisation. Since the load is constant torque in our case, more speed means more powerful motor, and to achieve the minimum material cost goal the volume of the magnet is set as an objective function. To find the optimum geometric parameters, we used gradient‐based method subject to non‐linear magnetic constraints. All the obtained results were validated by Ansoft Maxwell. Optimising using the proposed method including ANN‐based models does not require knowledge about complicated electric/magnetic equations. Also, ANN‐based BLDC motor model is more accurate than analytical models and faster than existing models in simulation tools.