Integrated Approach to Density-Based Spatial Clustering of Applications with Noise and Dynamic Time Warping for Breakout
- PDF / 2,669,388 Bytes
- 11 Pages / 593.972 x 792 pts Page_size
- 50 Downloads / 147 Views
NTRODUCTION
MOLD breakout is a catastrophic accident that can occur during continuous casting. A breakout not only affects smooth production and slab quality, but it seriously damages the casting machine, causing significant safety risks and huge economic losses.[1] The development of an accurate and efficient impending breakout prediction method is of great significance to ensure continuity of steel production.[2] To prevent and avoid breakout, the usual solution is to install thermocouples (TCs) embedded in the mold copper plates to monitor and evaluate, according to the temperature variation, whether breakout is occurring between the shell and copper plates.[3,4] Current methods of predicting breakout based on temperature of TCs
HAIYANG DUAN, XUDONG WANG, YU BAI, and MAN YAO are with the School of Materials Science and Engineering, Dalian University of Technology, Dalian 116024, P.R. China and also with the Key Laboratory of Solidification Control and Digital Preparation Technology (Liaoning Province), Dalian University of Technology, Dalian 116024, P.R. China. Contact e-mail: [email protected] QINGTAO GUO is with the State Key Laboratory of Metal Material for Marine Equipment and Application, Anshan 114009, P.R. China. Manuscript submitted November 14, 2018. Article published online July 8, 2019. METALLURGICAL AND MATERIALS TRANSACTIONS B
are roughly classified into two categories: logical judgment and artificial intelligence.[5] The logical judgment methods[6] have strong dependence on caster equipment, processes, and casting parameters. When adjusting operating parameters or casting speed, the thresholds of the logical judgment methods will change significantly, resulting in a substantial increase in the number of false alarms.[7] The artificial intelligence methods mainly use neural networks[8–10] and support vector machines and other algorithms to identify the temperature variation modes of single and multiple TCs when breakout occurs. These methods have strict requirements for learning and training samples. If samples are incomplete or invalid, the prediction results are severely affected. Hence, the migration ability of these models is weak. As a typical method employed in machine learning, clustering is widely applied in data mining[11] to identify data with similar characteristics. It is a process of grouping objects into clusters that have higher similarity in the same cluster and lower similarity in different clusters. Density-based spatial clustering of applications with noise[12] (DBSCAN) is a momentous branch of clustering algorithms. First, DBSCAN uses a rule formed by the parameters Epsilon (Eps) and minimum number of points (MinPts) to capture the samples. The rule is whether the number of samples in the neighborhood with a radius of Eps reaches MinPts. Then, the VOLUME 50B, OCTOBER 2019—2343
captured samples are gathered together to form high-density connected clusters[13]; meanwhile, the samples that do not satisfy the rule are filtered as noise.[14] Finally, all the samples are divided into clusters an
Data Loading...
