Chair: Dr. Kyung Ho SUN (Korea Institute of Machinery & Materials)
Fault Diagnosis of Tilting-pad Type Fluid Film Bearings in High-speed Rotating Machinery Dr. Kyung Ho SUN, Dong Hyuk KANG, Byeong Ock KIM, Jai-Kyung LEE and Donghyun LEE (Korea Institute of Machinery & Materials, Chungnam National University)
Of all fluid film bearing types, tilting pad bearings offer the optimal rotordynamic stability due to their excellent stiffness and damping characteristics. Due to the reduction in cross-coupling stiffness, tilting pad bearings are ideal for use in high-speed and high-load turbomachinery applications, such as industrial compressors and steam/gas turbines. In this investigation, we propose the method to diagnose the main failure mode of tilting pad journal bearings such as babbitt metal wear and pivot wear in bearing pads. Various experimental cases for bearing wear were precisely implemented through a high-speed rotating test rig operating at 10,200 rpm. The time, frequency and temperature data were obtained through a condition monitoring system and various features from shaft vibration, bearing metal and lubrication temperature data were extracted for the fault diagnosis. In this study, we have showned that even defects due to the abrasive wear of fluid film bearings including general rotor faults, such as unbalance, misalignment and rubbing, can be accurately diagnosed.
Feature Selection Method for Life Prediction in Multiple Degradation Unit: Generalized Rank Mutual Information Mr. Taewan HWANG, Keunsu KIM, Su J. KIM, Byungjoo JEON and Byeng D. YOUN (Seoul National University)
Many industries are making efforts to minimize the losses caused by shutdown of manufacturing facilities and to set an optimal maintenance schedule. In this context, prognostics, which predict remaining useful life (RUL) based on information extracted from sensory signals, have attracted attention. There are three methods to perform life prediction: physics-based, data-based, and hybrid. However, data-driven methods are the only way to apply them to a complex industrial facility. By assuming multiple degradation unit data, we can extract various features from the data and select the best feature to create a health index(HI). In this study, we propose a new method for the feature selection step that greatly determines the performance of RUL prediction. Proposed algorithm can automatically select features that are monotonic and have a consistent level of value in normal and failure zone. We validate our method using real degradation data acquired from bearing life testbed.
RPM Independent Fault Diagnosis of Rolling Element Bearing Mr. Su J. KIM, Keunsu KIM, Taewan HWANG, Byungjoo JEON and Byeng D. YOUN (Seoul National University)
Rolling element bearings are the most widely used and the most frequently broken mechanical components in rotating electrical machines. Therefore, many studies of diagnose the bearing health conditions have been conducted. To apply the diagnosis technology in the actual industrial field, there are two problems. For one thing, as the system operates at various rotational speeds, the behavior of the features change which indicate the health of the bearings making hard to diagnose. For the other thing, as the failure criteria is different for each system, bearing sometimes does not replaced even under localized fault. Continuous operation in such a condition propagates the defect up to distributed fault changing health feature. Thus, in this paper, we investigate the bearing diagnosis of localized and distributed faults under different rotational speeds. A deep groove ball bearing was used for this study and three health condition cases were considered, normal case, localized fault case, and distributed fault case. Experimental data under different rotational speed were acquired from the laboratory-level bearing life test-bed. The acquired data were only the vibration signals, which were gathered using an accelerometer sensor. Then, a number of signal features for bearing fault diagnosis were extracted after preprocessing. We proposed a new feature, which base on residual energy density of defect frequency, and then compared with the traditional time and frequency domain diagnosis feature including root mean square (RMS), Kurtosis, fundamental defect frequency, and so on. Finally, our new feature was selected to best features that could better diagnose bearing fault levels irrelevant to rotational speeds.
Four-Stage Degradation Physics of Rolling Element Bearings Mr. Keunsu KIM, Taewan HWANG, Su J. KIM, Byungjoo JEON and Byeng D. YOUN (Seoul National University, Seoul National University / OnePredict Inc.)
Rolling element bearings are a critical component of rotating machinery. Timely prediction of bearing faults become of great importance to minimizing unscheduled machine downtime. Most of the bearings experience gradual condition degradation due to repeated mechanical loads. Vibration signals are often used for bearing diagnosis and prognosis with a predefined threshold. However, false (positive/negative) alarms are often observed, thus leading to unnecessary downtime and expensive corrective maintenance. This is mainly because the thresholds are defined without accounting for bearing physics and a great deal of uncertainty in manufacturing and operation condition. To resolve this difficulty, this study aims at investigating the degradation physics of rolling element bearings using a vibration signal, while accounting for bearing physics and a substantial amount of uncertainty in manufacturing and operation condition. First, bearing feature engineering is thoroughly studied through time domain and frequency domain analyses. This study proposes the features that are most sensitive to the change in bearing physics. Second, bearing degradation physics is investigated so that the bearing degradation process can be modeled into four degradation stages. To the end, the proposed idea is demonstrated with vibration data measured from rolling element bearings, which experience accelerated life tested to simulate naturally induced degradation. This study will benefit to enhance physical understanding for bearing faults in various engineering applications.
Deep Neural Network for Fault Diagnosis of Power Transformers using Dissolved Gas Analysis Mr. Sunuwe KIM, Beom Chan JANG, Byeng D. YOUN, Daeil KWON and Byeong-Cheol PARK (Seoul National University, Seoul National University/OnePredict Inc., Ulsan National Institute of Science and Technology, Korea Electronic Power Corporation)
The dissolved gas analysis, produced by deterioration of insulating oil, is the most popular diagnostic tool to detect various incipient faults in power transformers. So far, the handcrafted DGA features, such as DGA composition ratios (i.e., C2H2/C2H4, C2H4/C2H6, CH4/H2), have been often used as the input features of shallow learning or used to identify diagnostic criteria (i.e., Dornenburg Ratio, Rogers Ratio, IEC ratio) for the fault diagnosis of power transformers. However, a false alarm rate is relatively large due to the limitations of the handcrafted features because they are made up of two or three gas combinations that can classify the fault types in a low dimensional space that can be analyzed by the human inspection. To enhance DGA-based diagnostic accuracy, a novel method using deep neural network (DNN) is proposed to determine high-level features without relying on the handcrafted features. Specifically, many layers of nonlinear transforms in a DNN convert the raw DGA data into a highly invariant and discriminative representation without losing high-dimensional information that human cannot analyze in high dimensional space. This makes health classification more effective. A proposed method is validated from the reference database of IEC TC 10, which is the visual inspection data of transformer faults. The results indicate that the proposed DNN approach achieves higher accuracy than the existing methods based on shallow learning with the handcrafted features