運用趨勢特徵擷取法建構預測性維修決策模型

Abstract

Although numerous attempts have been made on the study of predictive maintenance, little is known about to develop a predictive model that can detect the failure for the machines that run at different conditions. One of the major reasons is that a machine’s failure mode strongly depends on the customer’s machine parameter setting. Some customers run machine at tight specifications while others run different specifications for their operation of purposes. It is difficult to develop a golden predictive model that can applicable for all machines the run under different condition settings. Moreover, the ‘failure data’ for model building is unexplained or difficult to acquire. The poor quality of ‘failure data’ will increase the difficulty of analysis on the predictive model, and could lead to wrong decision. A new un-supervised modeling approach - Trend Attribute Extraction based Predictive Modeling (TAE-PM) is proposed to detect the potential machine failures. Compared with traditional predictive maintenance modeling, the presented methodology shows its adaptability and effectiveness to a gradually deteriorating system.

Although numerous attempts have been made on the study of predictive maintenance, little is known about to develop a predictive model that can detect the failure for the machines that run at different conditions. One of the major reasons is that a machine’s failure mode strongly depends on the customer’s machine parameter setting. Some customers run machine at tight specifications while others run different specifications for their operation of purposes. It is difficult to develop a golden predictive model that can applicable for all machines the run under different condition settings. Moreover, the ‘failure data’ for model building is unexplained or difficult to acquire. The poor quality of ‘failure data’ will increase the difficulty of analysis on the predictive model, and could lead to wrong decision. A new un-supervised modeling approach - Trend Attribute Extraction based Predictive Modeling (TAE-PM) is proposed to detect the potential machine failures. Compared with traditional predictive maintenance modeling, the presented methodology shows its adaptability and effectiveness to a gradually deteriorating system.