外文翻譯--- 高精度溫度測量裝置的設計

1、<p><b>　　附錄A</b></p><p>　　Design of a High Precision Temperature Measurement System</p><p>　　1 Introduction</p><p>　　Sensors are one of the most important elements

2、used in many instrumentation circuits. They are used in many industrial applications and take a certain form of input (temperature, pressure, altitude, etc.) and convert it into readings that can be interpreted. Many typ

3、es of sensors are nonlinear in nature from which a linear output is desired. There are many different sensors for temperature measurement and thermocouples are the most commonly used . They are preferred in industrial ap

4、plications du</p><p>　　the second uses numerical methods that are computed by microprocessor or computer . Analog circuits are frequently used for improving the linearity of the sensor characteristics, which

5、 implies additional analog hardware and typical problems associated to analog circuits such as temperature drift, gain and offset error. Using the second method, sensor nonlinearities can be compensated by means of arith

6、metic operations, if an accurate sensor model is available (direct computation of the polynomial</p><p>　　2 System Hardware</p><p>　　A thermocouple generates a voltage proportional to the measur

7、ement junction temperature at mV levels while the cold junction temperature is constant. In order to make an accurate measurement the cold junction temperature must be known. Figure 1(a) shows the block diagram of the te

8、mperature measurement system designed via an ANN in the operation phase. It consists of a thermocouple (type E) exposed to a desired temperature, including signal conditioning circuit with 16-bit analog to digital conve&

9、lt;/p><p>　　3 Artificial Neural Network</p><p>　　ANNs are based on the mechanism of the biologically inspired brain model. ANNs are feed-forward networks and universal approximators. They are train

10、ed and learned through experience not from programming. They are formed by interconnections of simple processing elements, or neurons with adjustable weights, which constitute the neural structure and are organized in la

11、yers. Each artificial neuron has weighted inputs, summation and activation functions and output. The behaviour of the overall ANN de</p><p>　　MLPs are the simplest and most commonly used neural network archi

12、tectures . They consists of input, output and one or more hidden layers with a predefined number of neurons. The neurons in the input layer only act as buffers for distributing the input signals xi to neurons in the hidd

13、en layer. Each neuron j in the hidden layer sums up its input signals xi, after weighting them with the strengths of the respective connections wji from the input layer and computes its output yj as a function f of </

14、p><p>　　The experimental data taken from thermocouple data sheets are used in this investigation. These data sheets are prepared for a particular junction temperature (usually 0°C). The ANN is trained with

15、 80 thermocouple temperatures that is uniformly distributed between -200 and 1000°C which is obtained in the calibration phase. However the performance of the final network with the training set is not an unbiased e

16、stimate of its performance on the universe of possible inputs, and an independent test s</p><p>　　4 Results and Conclusion</p><p>　　The developed ANN models are trained and tested with the use o

17、f different learning algorithms called LM, BR, CGF, RP and BFGS to obtain better performance and faster convergence with simpler structure. Table 1 shows the errors from the complete learning algorithms used in the analy

18、sis for the same network configuration mentioned above. When the performances of the neural models are compared with each other, the best result for the training and the test are obtained from the model trained with th&l

19、t;/p><p>　　Figure 3 represents the percentage test error of the network trained with LM for type E thermocouple. As it is clearly seen from Figure 3, the maximum percentage error becomes lower than 0.3%. The av

20、erage percentage error is greater than 0.1% for temperatures between -200 and 200°C, the reason being that in this range the thermocouples are strongly nonlinear. However, it is obvious for best fit in the range - 2

21、00 to 200°C that the number of training data set must be increased. The normalized err</p><p><b>　　附錄B</b></p><p>　　高精度溫度測量裝置的設計</p><p><b>　　1 引言</b><

22、;/p><p>　　傳感器是許多設備電路中最重要的元素之一。許多工業(yè)都應用傳感器，傳感器采用某一形勢的輸入（例如溫度、壓力、幅度等），并轉換成能夠解釋的儀器指示數(shù)。在本質上許多傳感器都是非線性的，但輸出卻要求是線性的。有許多種傳感器可以進行溫度的測量，其中熱電偶的應用最廣泛。由于熱電偶具有低成本、寬操作范圍、響應迅速、精確度高的優(yōu)點，更適合工業(yè)應用。對于溫度，熱電偶也具有非線性輸出。因此，傳感器的仿真和線性化技術非常

23、必要。為了解決傳感器的線性化問題，提出了兩種方法。第一種方法需要非線性的模擬電路，第二種方法應用能夠用微處理器或計算機計算的數(shù)值方法。模擬電路經(jīng)常被用來提高傳感器的線性特性，但模擬電路需要額外的模擬硬件，還具有對于模擬電路的典型問題，如溫度漂移，增益和補償誤差。應用第二種方法，如果一個精確的傳感器是可用的（通過多項式直接計算），或應用查詢多維的表格，傳感器的非線性能夠通過操作算法得到補償。多項式方法的直接計算更加精確但需要較長的計算時間

24、，而查表的方法盡管快但并不十分精確。</p><p>　　最近幾年，在使用儀器和測量的領域，人工神經(jīng)網(wǎng)絡作為一種有前景的研究領域已經(jīng)興起。人工神經(jīng)網(wǎng)絡對于解決復雜問題特別是非線性系統(tǒng)的仿真提供了神經(jīng)計算方法，而網(wǎng)絡本身卻是一個非線性系統(tǒng)。當測量系統(tǒng)是非線性時包括用來訓練的試驗數(shù)據(jù)也是非線性時，人工神經(jīng)網(wǎng)絡是非常有用的。人工神經(jīng)網(wǎng)絡的最廣泛應用是數(shù)值逼近（曲線擬和）。與傳統(tǒng)的數(shù)值插補比較，基于人工神經(jīng)網(wǎng)絡的插補提供

25、低的插補誤差。在本文中，我們提出了基于人工神經(jīng)網(wǎng)絡方法的高精度溫度測量系統(tǒng)。校正數(shù)據(jù)是通過在人工神經(jīng)網(wǎng)絡的訓練和測試階段必須具有的調(diào)幅調(diào)頻信號源9100校正單元獲得的。系統(tǒng)的硬件和軟件部分被綜合在用于系統(tǒng)測量和校正的虛擬設備上。人工神經(jīng)網(wǎng)絡通過提供一個理想的最終誤差來校正數(shù)據(jù)。這就是按照神經(jīng)元的結構和層數(shù)和神經(jīng)元的數(shù)量通過軟件，將校正和人工神經(jīng)網(wǎng)絡仿真數(shù)據(jù)之間的均方誤差最小化。</p><p><b>

26、　　系統(tǒng)硬件</b></p><p>　　熱電偶產(chǎn)生一和測量溫度點成比例的在mV數(shù)量級的電壓值，，而在零點的溫度值是一個常數(shù)。為了精確的測量，我們必須知道零點的溫度值。通過人工神經(jīng)網(wǎng)絡的操作階段，圖1（a）顯示了溫度測量系統(tǒng)模塊。它組成了放置在理想溫度條件下的熱電偶（E型號），包括帶有16位的模擬數(shù)字轉換器和接入到計算機的輸入輸接口卡的信號調(diào)節(jié)電路，設計的信號調(diào)節(jié)電路具有可設計增益的放（PGA204B

27、P），它的增益是1，10，100，1000倍，16位的A/D轉換器（AD976A），帶有零點溫度補償?shù)腁D595單片電路熱電偶放大器，并把它作為攝氏溫度計的標準，用來選擇熱電偶和攝氏溫度計的輸出的4路模擬多路器（ADG529A）。AD976A具有高速度、低功耗、16位A/D轉換器的特點，采用5V工作電壓。這一部分提供連續(xù)的逼近，轉換電容ADC，間隔的2.5V干擾和一高速度的平行界面。系統(tǒng)的精度直接依靠ADC每步的大小。具有的輸入，AD9

28、76A的LSB是。當AD595作為攝氏溫度應用時，熱電偶被忽略，微分的輸入被匯集到一起。在這種模式中，AD595產(chǎn)生一個比例因子為10mV/°C，它的輸出應用在編寫軟件的零點溫度數(shù)據(jù)中。AD595的</p><p><b>　　3 人工神經(jīng)網(wǎng)絡</b></p><p>　　人工神經(jīng)網(wǎng)絡是基于生物機理的人腦的模擬。人工神經(jīng)網(wǎng)絡是前向反饋網(wǎng)絡和全局逼近網(wǎng)絡。它

29、是通過經(jīng)驗而不是程序來訓練和學習的。它們是通過簡單的進程元素相互連接而形成的，或是可調(diào)節(jié)的閾值相互連接而形成的，閾值組成神經(jīng)的結構組成層。每一個人工神經(jīng)都有閾值輸入，求和單元、功能函數(shù)和輸出。整個人工神經(jīng)網(wǎng)絡的特性依靠所提及的人工神經(jīng)元、學習算法和網(wǎng)絡的構造。在訓練的過程中，神經(jīng)元之間的閾值可以通過一些標準（對于所有的訓練集目標輸出值和測量值之間的均方誤差達到預先設定好的極限值）進行調(diào)整，或達到最大的允許步數(shù)。雖然訓練過程非常耗費時間，

30、但這能提前完成，并能脫機運行。訓練完的神經(jīng)網(wǎng)絡用來測試數(shù)據(jù)但這一過程是看不見的。機器語言程序是構建神經(jīng)網(wǎng)絡中最簡單的和應用最廣泛的。機器語言程序由輸入、輸出、事先定義好神經(jīng)元的數(shù)量的一層或多個隱層。輸入層的神經(jīng)元只是作為緩沖器，用來運輸輸入信號到神經(jīng)元的隱層。在應用分別的連接強度對輸入信號進行加權后，隱層的每一個神經(jīng)元加和他的輸入信號,有求和函數(shù)計算輸出值，可以表示為此式中是人工神經(jīng)網(wǎng)絡的功能函數(shù)之一。訓練的神經(jīng)網(wǎng)絡由應用不同的學習算法

31、可調(diào)整的閾值構成。學習算法給出了在時</p><p><b>　　結果和結論</b></p><p>　　人工神經(jīng)網(wǎng)絡模型可以通過不同的學習算法如LM算法，BR算法，CGF算法，BP算法，BFGS算法進行訓練和測試，簡單的結構就可以獲得更高的特性和更快的收斂速度。對于以上所提及的相同的網(wǎng)絡構造，表1顯示了應用學習算法進行分析的誤差。當我們將神經(jīng)模型的性質相互進行比較時

32、，我們發(fā)現(xiàn)應用LM算法訓練的模型可以獲得最好的實驗結果。采用LM算法進行訓練的網(wǎng)絡他的訓練和測試誤差（最小均方誤差）分別為和。從表1我們可以清楚地看到，更接近LM的解可以從BR算法得到。在所提及的神經(jīng)網(wǎng)絡模型中，對于特殊應用的RP的實驗效果最差。他所強調(diào)的學習算法的精確度依靠寓所選擇恰當?shù)膶W習參數(shù)、網(wǎng)絡的構造和初始化值。圖3代表了對于E類型的熱電偶采用LM算法訓練網(wǎng)絡的誤差百比。從圖3可以看出，誤差的最大百分比低于0.3%。在-200&