Design and identification of a lumped-parameter thermal network for permanent magnet synchronous motors based on heat transfer theory and particle swarm optimisation

A lumped-parameter thermal network (LPTN) for a permanent magnet motor (PMSM) is developed to estimate the most crucial motor temperatures. In this contribution a 60 kW PMSM prototype designed for automotive traction drives is used as the investigation basis. Aiming at real-time motor monitoring well-known analytic equations from the heat transfer theory are used to model the dominant heat paths. Based on a three-dimensional approach in cylindrical coordinates a differential-algebraic state-space model with varying parameters (LPV) is proposed. Due to the chosen level of model abstraction as well as motor material data uncertainties significant estimation errors between the LPTN and experimental test bench measurements result. To improve the estimation accuracy particle swarm optimisation (PSO) is applied for strategic fitting of uncertain model parameters with respect to a maximum likelihood cost function. To avoid converging in suboptimal local minima, which is a typical problem of gradient-based standard optimisation methods, the meta-heuristic PSO is utilised for the resulting multi-variable, non-linear and constrained optimisation problem. For the identification process experimental training data is used which is statistically independent from the (cross-)validation profiles. As a result the maximum estimation error (worst-case) regarding all considered motor component temperatures can be drastically reduced to 8 °C.