外文翻譯---智能超快速的低成本risc實施鎳鎘電池充電器 英文

1、Low cost RISC implementation of intelligent ultra fast charger for Ni–Cd batteryPanom Petchjatuporn a, Phaophak Sirisuk b,*, Noppadol Khaehintung a, Khamron Sunat b, Phinyo Wicheanchote c, Wiwat Kiranon da Department of

2、Control and Instrumentation Engineering, Faculty of Engineering, Mahanakorn University of Technology, Bangkok 10530, Thailand b Department of Computer Engineering, Faculty of Engineering, Mahanakorn University of Technol

3、ogy, Bangkok 10530, Thailand c Test Engineering Department, Sanmina-SCI Systems Co. Ltd., Thailand d Department of Telecommunication Engineering, Faculty of Engineering, King Mongkut’s Institue of Technology, Ladkrabang,

4、 Bangkok 10520, ThailandReceived 25 March 2006; received in revised form 24 November 2006; accepted 29 June 2007 Available online 11 September 2007AbstractThis paper presents a low cost reduced instruction set computer (

5、RISC) implementation of an intelligent ultra fast charger for a nickel–cadmium (Ni–Cd) battery. The charger employs a genetic algorithm (GA) trained generalized regression neural network (GRNN) as a key to ultra fast cha

6、rging while avoiding battery damage. The tradeoff between mean square error (MSE) and the computational burden of the GRNN is addressed. Besides, an efficient technique is proposed for estimation of a radial basis functi

7、on (RBF) in the GRNN. Hardware realization based upon the techniques is discussed. Experimental results with commercial Ni–Cd batteries reveal that while the proposed charger significantly reduces the charging time, it s

8、carcely deteriorates the battery energy storage capability when compared with the conventional charger. ? 2007 Elsevier Ltd. All rights reserved.Keywords: Battery charger; Ni–Cd battery; Fast charging; GA; GRNN; RBF1. In

9、troductionA battery is an electrochemical device that converts chemical energy contained in its active materials directly into electric energy through an electrochemical oxida- tion–reduction reactions [1]. At present, a

10、 secondary or rechargeable battery is widely used in many applications including wireless communication devices and portable appliances. This type of battery includes nickel cadmium (Ni–Cd), nickel metal hydride (Ni–MH)

11、and lithium ion (Li-ion) batteries. Among these, Ni–Cd battery is the most popular rechargeable battery due to its low cost, high num-ber of charge–discharge cycles and excellent load perfor- mance [2]. Battery charging

12、is a very crucial factor in a contracted manufacturing, where a large number of battery powered plug-and-play devices are often produced. A fully charged Ni–Cd battery may be supplied by a battery manufacturer. However,

13、in many practical situations, such as shipment delay or firmware upgrading, the battery must be recharged by the manufacturer before shipping to a customer. Because of its simple structure and cost competitiveness, a con

14、stant trickle current charge strategy [1,2] is one of the most common charging techniques for a Ni–Cd battery. The technique requires a long charge time since it employs a very low charging current, i.e. 0.1C where C is

15、the battery capacity, Sometimes the charging lasts longer than 10 h, and thus, the charger is often referred to as an ‘‘overnight0196-8904/$ - see front matter ? 2007 Elsevier Ltd. All rights reserved. doi:10.1016/j.enco

16、nman.2007.06.023* Corresponding author. Tel.: +66 2988 3655; fax: +66 2988 4040. E-mail address: phaophak@mut.ac.th (P. Sirisuk).www.elsevier.com/locate/enconmanAvailable online at www.sciencedirect.comEnergy Conversion

17、and Management 49 (2008) 185–192allowed to go beyond 50 ?C [1]. In addition to the temper- ature, the battery voltage (V) immediately drops or, equiv- alently, the voltage gradient becomes (dV/dt) negative. Thus, two con

18、ditions can be utilized as an indication of the fully charged state, namely (i) the battery temperature is greater than 50 ?C (T > 50) or (ii) the voltage gradient is negative (dV/dt < 0).2.2. Ultra fast charger de

19、signSince its invention, various approaches have been employed for charging Ni–Cd batteries. One of the most popular charging techniques widely used in consumer products is the constant current trickle charge technique,

20、in which the charger supplies a very small, constant current rate (normally 0.1C) to the battery and relies on user inter- vention to stop the charge when the battery returns to full capacity. It is noted that charging a

21、 battery rated at 1 Ah with 0.1C implies the charging current of 100 mA and results in the charging time being at least 10 h. Reduction of the charging period simply requires the charging current to be larger. In practic

22、e, a Ni–Cd battery is capable of han- dling a charge current of 8C [2]. However, it cannot be sup- plied with such a large current for a long period; otherwise the battery will be severely damaged as discussed above. The

23、refore, the charging current must be adjusted in accor- dance with the battery charge state. A key to ultra fast charging is an intelligent controller that, given measured input parameters, can properly deter- mine the o

24、utput, i.e. the charging current (Ic). Apparently, the first task in designing the controller is to identify the control input parameters, which represent the battery sta- tus. Typically, the battery voltage and temperat

25、ure, together with their associated gradients are used as inputs to the controller [2,6]. Besides, the battery current may be used to estimate the amount of charge remaining in the battery [5]. One might expect that more

26、 parameters would provide better performance. From the implementation point of view, however, more parameters means more com- putational complexity and, hence, more complicated cir- cuitry. In our design, only the temper

27、ature and its associated gradient are employed as the control input parameters. In fact, only the battery temperature is mea- sured. The temperature gradient is easily computed using the past and current values of the me

28、asured temperature. Nonetheless, the voltage gradient is still necessary for ver- ification of the fully charged state. Based on extensive experimentations, the control inputs, i.e. the absolute tem- perature (T), temper

29、ature gradient (dT/dt) and charging current (Ic) were collected [12]. Fig. 2 depicts a surface plot that represents a relationship between the control inputs and corresponding optimal charging current. The figure reveals

30、 that the charger should supply a relatively high charge current during the initial phase of battery charging. Then, the charge current should be properly reduced dur- ing the later phases of charging depending on the tw

31、o con- trol parameters T and dT/dt. Following this strategy, anoptimal charging current within the safety limit will always be supplied without any battery damage, provided that the surface plot of Fig. 2 is well approxi

32、mated.3. Ultra fast charger using GRNN3.1. GRNNIn recent years, neural networks have received consider- able attention due to their efficiency in solving nonlinear, time varying problems in modern emerging applications.

33、As opposed to conventional mathematical model based techniques, neural networks do not require any models, which are often difficult, if not impossible, to obtain in practice. Although the traditional techniques can effi

34、ciently address linear, time invariant systems, they only perform linear approximations of nonlinear behavior. For some applications, the approximation may be sufficient. How- ever, better techniques are necessary when a

35、 higher degree of accuracy is mandatory. Neural networks are strongly recommended for such applications. In our context, the neural network is also a promising solution to realization of an ultra fast charger. This is be

36、cause charging the bat- tery naturally involves its inherent nonlinearities. There exist a large variety of models in the neural net- work paradigm. Among those is the GRNN [13], which is an emerging model of the neural

37、network paradigm. It possesses a relatively low computational burden while still offering a high degree of accuracy. Using training data obtained from Ref. [12], an ultra fast charger using the GRNN was introduced in [9]

38、. Once trained, the GRNN can appropriately determine the charging current (Ic), given the temperature (T) and temperature gradient (dT/ dt). In our system, the GRNN comprises three layers as depicted in Fig. 3. The first

39、 layer has two input nodes, par- ticularly for the temperature and its associated gradient. The second layer consists of neurons, each of which is rep- resented by a basis function. In Ref. [11], it was revealed that six

40、 neurons in the second layer yields the best tradeoff between mean square error (MSE) and computational complexity. This will be confirmed by simulations in Sec- tion 5. Note that, by error, we mean the difference betwee