A Probabilistic Approach to Maintenance Cost Analysis Considering Time-dependent False Alarms for PHM Design Mr. Joung Taek YOON, Minji YOO, Yunhan KIM and Byeng D. YOUN (Seoul National University)
Prognostics and health management (PHM) aims at predicting system failures in advance and timely conducting maintenance so as to minimize operation and maintenance costs. One of the challenges in PHM is how to deal with a false alarm. A false alarm in PHM implies incorrect estimation about a health state of an engineered system. This could incur unnecessary overhaul and maintenance or unexpected costly system failures. That is, the incurred costs due to false alarms should be analyzed prior to deciding the implementation of PHM. This paper presents the framework of life-cycle maintenance cost analysis considering false alarms. First, the framework estimates false alarm rates for a given PHM algorithm. In this study, two types of false alarms (i.e., false healthy and false faulty) are taken into account and health diagnostics is considered for a PHM algorithm. The false alarm rates are calculated according to their probabilistic definition while considering the dynamic (or time-dependent) property of false alarm rates. Then, three maintenance probabilities (unnecessary, corrective and preventive) are calculated by multiplying the probability of a true health state (reliability or failure rate) by the conditional probability from the detectability matrix. For instance, the probability of the unnecessary maintenance is calculated as the conditional probability of system faulty estimation given a true healthy system (false faulty) times the probability of a true healthy system (reliability). Lastly, the expected costs of three maintenances are calculated with consideration of the calculated maintenance probabilities and their costs. An electro-hydrostatic actuator simulation model is employed for demonstration. Its expected maintenance cost during life-cycle is estimated, to demonstrate the validity of the proposed framework.
Does PHM Make an Engineered System Resilient under Sensor Faults? Ms. Minji YOO, Joung Taek YOON, Yunhan KIM and Byeng D. YOUN (Seoul National University)
The use of Prognostics and Health Management (PHM) technology enables engineered systems resilient under adverse events. Adverse events include unexpected system failures, anomaly operation, manufacturing defects, etc. As engineered systems become more complex, resilience thus becomes an emerging engineering feature, which offers an attractive ability to resist and recover from adverse events. However, most resilience studies were conducted under the assumption of no false alarms. This study thus concerns how engineering resilience can be formulated while taking into account sensor faults. Sensor faults can be interpreted in various manners: sensor breakdown, calibration change, sensor noise, etc. This study proposes a new formulation of engineering resilience considering sensor faults. Sensor faults can be implemented by changing a sensor gain. The effectiveness of the proposed resilience measure is demonstrated by implementing the PHM into the electro-hydrostatic actuator (EHA) and assessing the resilience under sensor faults. This study is also conducted to see how the proposed resilience measure behaves in the course of the sensor degradation.
Online Temperature Estimation of Li-ion Battery Pack Using Principal Component Analysis Dr. Taejin KIM, Sunuwe KIM and Byeng D. YOUN (Seoul National University, Seoul National University / OnePredict Inc.)
Thermal management is one of the important function of the battery management system (BMS). The thermal management system monitors and equalizes the temperature distribution of the battery pack to prevent the different cell degradation rate and to keep the battery on its best performance. In this study, as a part of the thermal management system the in-situ temperature estimation method is developed based on the principal component analysis (PCA) reinforced with the measured temperatures. To begin with, the PCA is used for finding the basis vectors of the battery thermal system, which is the eigenvectors of the covariance matrix of the training data set. Then an arbitrary thermal map can be expressed as the linear combination of these basis vectors and their amplitudes. The amplitude for each basis vectors is estimated from the measured temperatures. The performance of the thermal map reconstruction depends on the accuracy of this amplitude estimation which again is related to the temperature measurement locations. The measured locations are determined considering two aspects: the prediction accuracy and the robustness of the sensor network. To find the sensor location satisfying both criteria, the sensor network optimization problem is accordingly formulated, and solved by the genetic algorithm. The proposed study is validated for various operating conditions including the distributed heat generation condition and different cooling conditions.