(654e) Application of Just-in-Time Statistical Process Control to Vinyl Acetate Monomer Process
AIChE Annual Meeting
Thursday, November 11, 2010 - 1:50pm to 2:10pm
In order to operate manufacturing processes efficiently, it is crucial in any industry to detect malfunction of equipments and abnormal quality of products as early as possible, to identify the cause of such faults, and to take appropriate measures. Although a lot of researches have been performed in the area of fault detection and identification / isolation / classification / diagnosis, the industry is not satisfied with the presently available technologies and still looks for practical methods. In general, the development of a fault management system depends on characteristics of a process and needs a long period of trial and error. Even if a good fault management system can be developed, it is more difficult to cope with changes in process characteristics and to maintain the system. In other words, a major difficulty of fault management is the system maintenance. Therefore, it is really important to develop a method to easily design a high-performance, flexible, and adaptive fault management system.
In the present work, a new fault detection and identification method is proposed. The proposed method focuses on the distance from the current operation data to the normal operation data stored in the database. The distance is calculated and checked every time when fault detection should be conducted. The judgment is based on the statistical test concept in the same way as statistical process control. Thus, the proposed method is called Just-In-Time Statistical Process Control (JIT-SPC).
The off-line procedure for building a fault detection system is as follows: 1) Store normal operation data in a database. An appropriate distinction between faulty data and normal data is crucial to realize the desired performance. 2) Set i=1. 3) Calculate distance between the i-th sample and all the other samples stored in the database. 4) Estimate the cumulative distribution function of the distance. 5) Calculate the distance at which the cumulative distribution function reaches the threshold determined in advance. This distance is called sparse distance hereafter. 6) Set i=i+1 and go to step 3 till i reaches the number of samples. 7) Estimate the cumulative distribution function of the sparse distance. 8) Set a control limit of the sparse distance on the basis of the cumulative distribution function.
The control limit in the above-mentioned procedure is updated easily because statistical modeling such as principal component analysis (PCA) is not necessary in the proposed method.
The on-line fault detection procedure is as follows: 1) Calculate distance between the query and all the samples stored in the database. 2) Estimate the cumulative distribution function of the distance, and calculate the sparse distance. 3) The query is abnormal if the sparse distance is beyond its control limit.
In addition to fault detection, the concept of a contribution plot, which is well-known in the area of multivariate SPC, can be used to identify measured variables that contribute to out-of-control signals. In the proposed system, contribution is determined as follows: 1) Select normal samples that exist within the sparse distance from the query. 2) Decompose the distance between the query and selected normal samples along each measured variable. 3) Calculate the sum of the decomposed distance for each measured variable. This sum is defined as contribution.
The JIT-SPC is applied to a vinyl acetate monomer production process. The results clearly show that the proposed JIT-SPC can cope with changes in process characteristics as well as multiple operation modes and also it is superior to the conventional MSPC.
This paper has an Extended Abstract file available; you must purchase the conference proceedings to access it.
Do you already own this?
Log In for instructions on accessing this content.
|AIChE Pro Members||$150.00|
|AIChE Graduate Student Members||Free|
|AIChE Undergraduate Student Members||Free|
|AIChE Explorer Members||$225.00|