Applied Soft Computing 40 (2016) 616–623 Contents lists available at ScienceDirect Applied Soft Computing j ourna l ho me page: www.elsevier.com/locate /asoc Neural speed estimator for line-connected induction motor embedded in a digital processor Clayton Luiz Graciola a , Alessandro Goedtel a,∗ , Marcelo Suetake b , Rodrigo Rodrigues Sumar a a Federal University of Technology (UTFPR), Elect. Eng. Dept., Av. Alberto Carazzai, 1640, 86300-000 Cornélio Procópio, PR, Brazil b Federal University of São Carlos (UFSCAR), Rodovia Washington Luís, km 235 – SP 310, 13565-905 São Carlos, SP, Brazil a r t i c l e i n f o Article history: Received 5 July 2015 Received in revised form 22 November 2015 Accepted 19 December 2015 Available online 30 December 2015 Keywords: Induction motors Artificial neural networks Feedforward neural networks Multilayer perceptron Parameter estimation a b s t r a c t Estimating the electrical and mechanical parameters involved in three-phase induction motors is fre- quently employed to avoid measuring every variable in the process. Among mechanical parameters, speed is an important variable: it is involved in control, diagnosis, condition monitoring, and can be mea- sured or estimated by sensorless methods. These technologies offer advantages when compared with direct measurement, such as lower cost or more robust systems. This paper proposes the use of artificial neural networks to estimate rotor speed by using current sensors for balanced and unbalanced voltage sources with a wide mechanical load range in a line-connected induction motor. This paper also presents two case analyses: (i) a single current sensor; and (ii) a multiple currents sensors. Simulation and exper- imental results are presented to validate the proposed approach. A neural speed estimator embedded in a digital processor is also presented. © 2016 Elsevier B.V. All rights reserved. 1. Introduction Three-phase induction motors (TIMs) are used in many indus- trial applications such as pumps, fans, machine tools and robotics as a key element in the conversion of electrical into mechanical energy. Several control strategies for these machines are based on electronic drives with sensorless technologies, which is a growing trend in TIM monitoring and control [1]. Conventional methods based on direct measurement of machine variables, such as torque and speed, have some dis- advantages, in addition to the higher cost involved with the driver implementation. Rotor speed can be measured with optical encoders, electromagnetic resolvers or brushless DC tachogenera- tors. However, the use of these electromechanical devices is subject to limitations, such as increased driver costs, reduced mechani- cal robustness, and low noise immunity. They also affect machine inertia and require special attention in hostile environments [2]. The use of sensorless techniques is primarily found in induc- tion motor control drives [3–5]. However, induction motor speed is also an important variable to be considered in condition moni- toring [6–8] and fault detection [9–11]. The main approaches for ∗ Corresponding author. Tel.: +55 4335204096. E-mail address: agoedtel@utfpr.edu.br (A. Goedtel). speed estimation are open-loop estimators using monitored sta- tor voltage and current, state observers, model reference adaptive systems, and artificial intelligence [2]. The conventional numerical methods for speed estimation are based on machine models. In this case, speed can be calculated by using machine model equations which require voltage and current as local machine parameters. Other variables for equation solving are the electrical and mechanical parameters of the machine, e.g., resistances, inductances, and load inertia, which are unavailable in the machine nameplate to feed the equations [12]. The disadvantages of this method are: (i) unavailable parame- ters, such as resistances and inductances in the machine nameplate to feed the equations; (ii) the need to solve machine equations; (iii) most models are linear; and (iv) substantial computing power is required in the application. Condition monitoring is one of the applications of the speed estimators. The work of D’Angelo et al. [11] proposed approach is related to the enhanced resilience of the new motor failure detection procedure against false alarms, combined with a good sensitivity that allows the detection of rather small fault signals. The proposed system monitors the instantaneous values of the motor currents i as , i bs , i cs and the rotor speed ω [11]. Furthermore, control drives plays also an important role of intelligent speed esti- mation. For example, the work of paper [5] proposed an open-loop neuro-fuzzy speed estimator with an innovative development. The http://dx.doi.org/10.1016/j.asoc.2015.12.033 1568-4946/© 2016 Elsevier B.V. All rights reserved.