No abstract

The knowledge of realistic dynamic models to robotic actuators would be of great aid in the synthesis of control laws to robot manipulators, mainly in cases of great precision robotic or even for manipulators with flexible links. In this paper we present a training scheme and propose a structure of neural network (NN) to learn the friction torque of a geared motor drive joint robotic actuator. To train the NN was used experimental data obtained by an harmonic-drive actuator, equipped with an encoder to measure the rotor angular position. It was considered the motor torque and the rotor angular velocity as NN input, while the friction torque was the only output, which was used in the proposition of a non-linear friction compensation mechanism. The experimental results attested the efficiency of the NN friction estimation and compensation with the proposed mechanism

IntroduçãoExistem atualmente diversas aplicações de redes neurais artificiais (RNA), nos mais variados domínios da ciência e tecnologia [(Kaynak, and Ertugru, 1997), (Jung and Hsia, 1998)]. Trata-se de um assunto que tem merecido grande atenção por parte da comunidade científica. Em [Miller, 1995] tem uma importante descrição sobre o histórico das redes neurais artificiais.Neste artigo investiga-se a possibilidade de identificação do torque de atrito de um atuador robótico do tipo moto-redutor, utilizando-se uma Rede Neural Artificial (RNA). A principal motivação reside na dificuldade de se encontrar um modelo dinâmico muito realista e preditivo para atuadores robóticos, dificuldade esta devida principalmente aos atritos não lineares internos aos atuadores [Armstrong, 1988]. Esta aplicação (identificação de atrito via RNA) começou a ser mais estudada apenas recentemente [(Dapper and Zanh, 1999), (Selmic and Lewis, 2000)], tratando-se, portanto, de um assunto bastante atual. Uma vez identificado o atrito não linear, propõe-se um mecanismo de compensação deste a fim de linearizar o atuador. A estratégia de identificação, utilizada conjuntamente ao mecanismo de compensação, constitui a principal contribuição deste trabalho.Como não se dispõe, até o presente momento, de um atuador robótico isolado em uma bancada a fim de se realizar experimentos, utilizou-se um modelo proposto na literatura [Gomes, 95]. Os resultados deste modelo serviram de referência como padrões de treinamento da RNA. A seqüência natural do presente trabalho será adquirir um moto-redutor do tipo harmonic-drive para a obtenção dos resultados experimentais, a fim de validar definitivamente o mecanismo de compensação de atritos proposto e testado em simulação. Como o modelo utilizado foi muito testado experimentalmente, acredita-se que todo o formalismo de identificação e compensação proposto deva proporcionar bons resultados na prática. O Atuador RobóticoUm atuador robótico do tipo moto-redutor pode ser visualizado como um elemento transmissor de movimento contendo uma elasticidade interna de constante K, conforme representado na Fig. 1. O torque motor m T é aplicado ao rotor de inércia r I . Atritos não lineares estão sempre presentes neste tipo de dinâmica, fazendo com que o torque motor não seja transmitido integralmente à inércia s I acoplada à saída do redutor. As equações (1) descrevem a dinâmica deste sistema [Gomes and Chrétien, 1992]. at T corresponde ao torque de atrito não linear, o qual será obtido via RNA. s r θ θ e são, respectivamente, os ângulos do rotor e do eixo de saída do redutor.

Much work have been done lately to develop complex motor control systems. However they always rely on a physical drive/motor/encoder setup for experimental results. This paper presents a hardware DC motor emulator that can be synthesized to FPGAs. The emulator is intended to replace an actual DC motor during the development phase of motor controllers. A torque based input is required and incremental encoder output is provided, so this model can replace both the DC motor, and its power driver without modifications to the motor control system. The proposed emulator is able to reach a clock frequency of hundreds of megahertz and uses very few logic resources in current FPGA technologies. The hardware can be parameterized at synthesis time to make the model suitable for specific needs.

The design of control laws for flexible manipulators is known to be a challenging problem, when using a conventional actuator, i.e., a motor with gear. This is due to the friction of the nonlinear actuator, which causes torque dead zone and stick-slip behavior, thereby hampering the good performance of the control system. The torque needed to attenuate the vibrations, although calculated by the control law, is consumed by the friction inside the actuator, rendering it ineffective to the flexible structure control. Nonlinear friction varies with different operational conditions of the actuator and a friction compensation mechanism based on these models cannot always keep a good performance. This study proposes a new control strategy using wavelet network to friction compensation. Experimental results obtained with a flexible manipulator attest to the good performance of the proposed control law.

This paper presents a non-linear friction compensation mechanism using a combination of neural network (NN) with fuzzy system (neuro-fuzzy compensator), applied to harmonic-drive robotic actuators. The friction compensation torque is constituted by NN output, which is trained off-line. Since the friction changes significantly over time, temperature and equipment operational conditions, the NN loses its performance. To recover this performance, a fuzzy algorithm is proposed to deal with the variation friction parameters. The output of the fuzzy algorithm is a gain that multiplied by the NN output will adjust the friction compensation torque. Experimental results have shown the efficiency of the proposed mechanism.

This work discusses the conceptual design of a maneuvering test platform to be installed at the LABOCEANO ocean basin. The platform features a 35m long main bridge, crossing over the whole basin width. A secondary trolley runs along the main bridge, connected to the model. The combination of the bridge and trolley motions can impose a very flexible range of trajectories to the model. The trolley will be equipped with special devices which enable the imposition of harmonic motion up to four degrees of freedom simultaneously. The platform technical specification includes longitudinal and transversal velocities up to 1 m/s, acceleration up to 0.2 m/s, and maximum trajectory diameter of 26m. The amplitude and frequency of the forced harmonic motion may read 0.5 m and 0.2 Hz for surge and sway; 0.15 m and 2 Hz in heave; 10 degrees and 2 Hz in roll and pitch; and 30 degrees and 0.5 Hz in yaw. The work features a general description of the platform design with special emphasis in the structural, mechanical, control and safety aspects.

No abstract