Integrated Group-based Valuable Sensor Selection Approach for Remaining Machinery Life Estimation in the Future Industry 4.0 Era

Abstract

Industry 4.0 is the evolution trend for current manufacturing technology. By analyzing the real-time sensing data, the health status of each machinery is usually monitored to reduce the risk of suddenly machine failure. Although massive sensors allocation can leverage the Remaining Useful Life (RUL) estimation for each machinery, the cost for the sensor network construction will become expensive. Hence, it is necessary to have an approach to remove the redundant sensors under a certain constraint of RUL estimation. On the other hand, due to the attractive performance on the object classification, many researches apply Artificial Neural Network (ANN) to decide which allocated sensor should be removed during the training process. However, the current researches aim to remove the redundant sensors based on the sensing data at a specific time, which lacks the intrinsic feature of time-series sensing data. Therefore, the current researches suffer from the problem of sensor under-killing due to the worst-case consideration. In this paper, we consider the information of time-series sensing data to propose an integrated group-based valuable sensor selection algorithm. Because the proposed approach considers the historical data during the redundant sensor removing process, we can reduce the number of involved allocated sensors precisely and significantly. In order to verify the proposed method, we use the Commercial Modular Aero-Propulsion System Simulation (CMAPSS) dataset and adopt Prognostics and Health Management (PHM) score to evaluate the RUL estimation performance. Compared with the conventional approach, the proposed approach can reduce 86% average PHM score and employ fewer sensors to fit the strict constraint of PHM score with less computing overhead.