The direct measurement of speed in induction motors is costly and requires maintenance. Thus, sensorless techniques for estimating or predicting the speed in three-phase induction motors represent a feasible and economical solution. This work considers a single time delay neural network as a speed estimator in two different strategies of stator field-oriented induction motor drive: direct current and torque control. Time delay neural network makes the estimated signal robust against noise, that is usually found in switched power systems, and against disturbances on the input signals, since the estimator is not dependent only on instantaneous values. The synchronous speed and the electromagnetic torque, which are usual quantities in field oriented drives, are the inputs of the proposed neural estimator. In order to have a robust estimator facing induction motor parameter variations, the procedure of training and validating the neural networks are conducted with three different induction motors, from simulations to the experimental tests. An embedded system is also presented, and the scheme is tested considering various speed and load torque levels with different control strategies.

在感应电机中直接测量速度成本高昂且需要维护。因此，用于估计或预测三相感应电机速度的无传感器技术代表了一种可行且经济的解决方案。这项工作将单个时间延迟神经网络视为定子磁场定向感应电机驱动的两种不同策略中的速度估计器：直流和转矩控制。时间延迟神经网络使估计的信号对噪声（通常在开关电源系统中发现）和输入信号的干扰具有鲁棒性，因为估计器不仅仅依赖于瞬时值。同步速度和电磁转矩是磁场定向驱动器中的常用量，是所提出的神经估计器的输入。为了有一个鲁棒的估计器面对感应电机参数变化，训练和验证神经网络的过程是用三种不同的感应电机进行的，从模拟到实验测试。还提出了一个嵌入式系统，并在考虑各种速度和负载转矩水平的情况下对不同控制策略的方案进行了测试。