馬達系統被廣泛運用在現今工廠許多精密且複雜的機器中，一旦馬達發生異常或故障導致生產機台無預警的停機，將造成重大的損失，而其中的軸承扮演著重要的角色，許多故障原因與軸承均具有高度相關。近年來，數位分身越來越被廣泛應用於工廠機台之運作，藉由建置之虛擬模型，可以讓使用者能透過感測器或軟體收集的數據，即時了解目標對象或系統的運作情況。儘管許多研究表明數位分身的諸多優點且有針對馬達軸承來建構數位分身，目前仍缺乏以虛擬模型的模擬資料結合真實資料以及預警系統來提升對馬達軸承之異常偵測結果。本研究目標為針對馬達中的軸承，透過感測器收集到的資料，先使用線性回歸(linear regression)和決策樹回歸(decision tree regression)來建置其數位分身後，使用樣本合成技術(synthetic minority oversampling technique)平衡資料集並利用k鄰近演算法(k-nearest neighbors)、隨機森林(random forest)和單純貝氏分類(na?ve bayes)演算法以及警報機制，來分析軸承的異常偵測。結果顯示與未採用數位分身相比，使用數位分身後可以明顯提升模型準確率。 ;Motor systems are widely used in many sophisticated and complex machines in factories nowadays. Once the anomaly or failure occurs on motor and causes unexpected shutdown, it will lead to significant losses. The bearing plays an important role in motor, and many failures and anomalies in motor are highly related to bearings. In recent years, digital twin (DT) has become more and more widely used in factory machines. By building virtual model, it allows users to understand the real situation of the target object or system in real time through the data collected by sensors or software. Although many studies have shown advantages of DT and the construction of DT for motor bearings, there is still a lack of using simulation data of DT models combined with real data and early warning systems to improve anomaly detection results of motor bearings. The goal of this study is to use the data collected by sensors, first apply linear regression and decision tree regression to build DT for the bearings in motor. Then, use synthetic minority oversampling technique to balance dataset and adopt k-nearest neighbors, random forest, na?ve bayes algorithm and alarm rules to analyze bearings anomaly detection. The results show that compared with not using DT, the accuracy of the model can be obviously improved by digital twin.
