作者：北京和利時(shí)智能技術(shù)有限公司，寧波和利時(shí)智能科技有限公司 范瑩，李天輝，劉宗福，姜百寧

摘要：為解決機器人故障預測問(wèn)題，提出一種使用機器人軸溫預測的方法，實(shí)現機器人異常發(fā)熱的提前預警，避免故障的發(fā)生。本文使用生產(chǎn)中機器人的歷史數據作為數據源，分析對機器人軸溫產(chǎn)生影響的測點(diǎn)，結合機器人機理，對數據進(jìn)行衍生轉換等操作，生成可以用于機器人軸溫預測的特征數據。鑒于沒(méi)有免費午餐（NFL，No Free Lunch）定理，對所有數據可用回歸算法進(jìn)行自動(dòng)化選擇，然后使用hyperopt進(jìn)行優(yōu)化調參，最終生成可用的機器人軸溫預測模型。本文中所述數據獲取、分析建模、上線(xiàn)部署及在線(xiàn)預測全部過(guò)程在HolliCube工業(yè)互聯(lián)網(wǎng)平臺上進(jìn)行。

關(guān)鍵詞：溫度預測；故障診斷；機器人；工業(yè)互聯(lián)網(wǎng)平臺；運行周期

Abstract: Aiming at fault diagnosis for robots, a method which predicts robot's axes temperature is proposed to realize abnormal robot heating in advance and avoid the occurrence of faults. In this paper, the historical data of the robot in production are used as data source to analyze the measuring points which affect the axle temperature of the robot. Combined with the mechanism of the robot, the data are derived and converted to generate the characteristic data which can be used to predict the axle temperature of the robot. Then select regression algorithms automatically caused by the NFL theory, and then optimize the hyper-parameters by hyperopt and finally we got the model. The whole process of data acquisition, analysis and modeling, online deployment and online prediction described in this paper are carried out on HolliCube industrial Internet platform.

Key words: Temperature prediction; Fault diagnosis; Robot;Industrial internet platform; Running cycle

在線(xiàn)預覽：基于數據驅動(dòng)的機器人軸溫預測建模與應用

摘自《自動(dòng)化博覽》2019年10月刊