- 學位論文
應用類神經網路於滾珠螺桿之定位與行程分析
The Application of Neural Network in Positioning and Stroke Analysis for the Ball Screw
摘要
本研究為藉由滾珠螺桿在3000rpm高速中，以不同的負載所產生的螺帽溫度與螺桿溫度對定位及行程精準度影響之數據，應用類神經網路以建構定位學習與預測機制。實驗上藉由光學尺即時量測螺帽移動的位置，並在螺帽黏貼熱電偶量測螺帽溫度變化，且利用熱影像儀量測螺桿溫度，將溫度變化與馬達加速度值使用類神經網路中的倒傳遞演算法，進行滾珠螺桿之定位及行程的誤差。在類神經網路建構中，輸入參數包含加速度、螺桿溫度、螺帽溫度及時間，輸出值為定位誤差值與行程誤差值。經過訓練和驗證之後，尋找出網路架構之隱藏層神經元數目與網路學習速率的最佳網路參數設定，再以相同操作條件之另一筆測試資料進行網路模擬分析預測情形。與本研究誤差探討，結果顯示適當之學習誤差量選擇，有助於得到較佳之預測結果曲線。所建構之倒傳遞類神經網路模型，可作為滾珠螺桿之定位控制參數修正運用。 關鍵字:滾珠螺桿、類神經網路、倒傳遞演算法、定位、行程
並列摘要
The paper is through a ball screw system operating at high rotational speed 3000 rpm with the different of normal loads. High speed and applied load generates nut temperature and screw temperature, they will affect accuracy of positioning and stroke, and an artificial neural network is applied to construct positioning learning and prediction mechanism. In the experiment, position of nut is measured by a linear ruler, and thermocouples is also used for detecting temperature variation of nut. Infrared thermal imaging camera is used for screw temperature measuring. A neural network of the back propagation algorithm has created for analyzing positioning of ball screw and stroke. Temperature variation and acceleration of motor are input factors for the method. In the construct artificial neural networks, input parameters including acceleration, temperature of screw, nut temperature and time, the output values are the positioning error value and stroke error value. After training and validation, the best network parameter settings of the hidden layer neuron number and network learning rate are found in the network architecture, and the another positioning and stroke data with the same operating conditions were used to comparing with well trained network simulation analyzing result in order to confirm its prediction ability. Error is explored with the study; the results show that the appropriate selection of learning error can help to get a better prediction curve. The constructed back propagation neural network model can be used for correction of control parameter in a ball screw system. Keywords: ball screw; neural networks; back propagation algorithm; positioning; stroke