Hybrid Neural Current Controllers for Decentralized and Highly Nonsinusoidal Multiphase Drives in Healthy and Faulty Modes

This paper proposes a neural current controller for decentralized and non-sinusoidal permanent magnet multiphase drives. This architecture has some advantages, such as increased reliability and allows a simplified motor design, but requires new algorithms for decentralized control. The neural network current controller features a so-called hybrid operation, with one part trained offline and a second part working online. Using the machine’s analytical model, the network is first trained offline based on Multi-Agent Reinforcement Learning (MARL) and is then implemented in the real time control. In a second step, by using Least Mean Square (LMS) algorithm, an online calibration is applied only to the output layer to take account the uncertainties of the model and allows at the same time reducing the computational burden. This solution offers a quick response thanks to the offline (forward) training and an adaptability of online (feedback) update. The proposal has been validated in simulation and in experimentation on a highly non-sinusoidal 7-phase test bench to demonstrate the feasibility of the proposed approach.