A Centrifugal Process Monitoring With The Consideration Of Both Batch And Continuous Process Feature

Ethylene is one of important petrochemical product which is produced in the industrial furnace tube. Industrial furnace tube always suffers harsh reaction condition, which is easy to penetrate and fracture and even cause serious production accident, so the quality of furnace tube is very important for industrial production. Centrifugal casting technology as one kind of casting technology is widely used in the production of casting. Under the action of centrifugal force, the metal molten steel is solidified on the surface of the cylinder. The industrial furnace tube is produced by centrifugal casting. By this method, the structure of furnace tubes will be compact, the crystal will be integrity, and few impurities can be mixed into the furnace tube.The traditional methods used to test the quality of industrial furnace tube are appearance checking and penetration checking. There are great workload consisting in these methods, in which a large amount of capital is invested. In order to efficiently monitor the production process of industrial furnace tube, multivariate statistical analysis techniques are used for process analysis, monitoring and fault diagnosis.Large amount of historical process data are collected in this process.Data are transformed and represented as a priori knowledge to a diagnostic system. This is also known as feature extraction.The main work of this paper includes the following aspects:(1) The collected data consist of three parts. The materials related data collected from the raw material melting process, the process related data collected from the centrifugal casting process and test results on both appearance and penetration check. The data are divided into different categories according to the inner diameter (outer diameter) and the content ratios of chromium and nickel in the materials. The number of data sampling for each batch is determined. If the data is seriously lost, the data should be deleted. If not there are many methods to solve this uneven length problem.(2) The materials related data is the chemical element content. The value of this kind of data is not change with time and its characteristic is similar to the continue process data. The process related data mainly contain the temperature of the cylinder, the rotational speed of the cylinder and the vibration amplitude of cylinder. Those variables are changing with time. The data collected from this process is a three dimensional matrix. These two types of data need to be divided again. One is the process related data. The other is the single measurements that contain the materials related data and the statistics indices extracted from the process related data. These two kinds of data are modeled by different methods. A diagram of quality monitoring model for industrial furnace tube is built.(3) The characteristic of continues process is extracted by PCA method and the feature of batch process is extracted by MPCA. The combination of PCA and MPCA can fully extracted the characteristic of process contains continue process and batch process. Three different types of furnace tubes are used to build models, the modeling results are as follows:The first kind of furnace tube:the rate of correct identification is 100%, and the rate of false alarm is 0%. The second kind of furnace tube:the rate of correct identification is 100%, and the rate of false alarm is 0%. The third kind of furnace tube:the rate of correct identification is 80%, and the rate of false alarm is 22%. The contribution plots are given to analyze forecasting results and identify root cause for failure batch, which can provide a reference for the improvement of pipe manufacturing and saving the cost on penetration testing.