The Application of Machine Learning Technique to the Weapon System for Critical Spare Parts Requirements Forecasting

• Main Authors: Chang,Ying-Chen, 張瑩貞
• Other Authors: S. Deng
• Format: Others
• Language: zh-TW
• Published: 2012
• Online Access: http://ndltd.ncl.edu.tw/handle/86240607784241412833

碩士 === 國防大學理工學院 === 兵器系統工程碩士班 === 100 === M60A3 combat tank is one of the major weapon systems of armoured ground force, the availability of power system and components is directly influence fighting capacity. Because the spare parts take long led time and budget is decreasing each year, the accurate to forecast the spare parts is important. In order to improve the spare parts forecasting, this thesis applied the Machine Learning Methods, Support Vector Machines (SVM) and Back Propagation Neural Network (BPNN), plus Poisson distribution function, perform the requirements forecasting for machinery critical spare parts. This research collecting five years (4 for each year) of 20 spare parts replacement information and separated into two parts: training and forecasting stage. The Poisson distribution function is used to get Poisson probability. The first 14 requirements amount were used for training, and last 6 requirements amount were used for forecasting. When forecasting the requirement, the Poisson probability must be calculated first, then to feed it to machine learning model and get the forecasting requirement. The forecasting requirement then compares with actual requirement. The results show, regardless that Poisson probability and forecasting requirements, both SVM and BPN have good forecasting accuracy is more than 80%. The SVM predicts accuracy is relatively superior to BPN because less sampling rate. This research provided a concept that how to forecast the requirements when only the number of spare parts replacement were provided. This research will also help the defense logistics unit perform its annual plan for forecasting requirement of spare parts.