Abstract: In the field of process industry such as metallurgy and chemical industry, the manufacturing process of the product involves several continuous related processes. Each process needs to determine the control range of the process parameters, that is, the process specification is formulated. At present, the process industry mainly uses the 6-Sigma method or the production experience to determine the range of key process parameters in each process. However, the 6-Sigma method based on the assumption of independent distribution is lacking in the ability to deal with multiple correlation and non-normal distribution, so the process specification is easy to cause the instability of product quality in industrial application. In order to solve the problem existing in the process specification of the process industry, a method for formulating process specifications based on multivariate statistical analysis is proposed. First, the concept of soft hyper-sphere is introduced, and the method of nonlinear kernel function is used to determine the boundary of the soft hyper-sphere in the high dimensional space, that is the process parameter boundary of the product quality-controlled zone. Then, the largest inscribed form of the hyper-sphere is found out, and the process specification is formulated according to the largest inscribed rectangle. Finally, the example of IF steel production is conducted, and the formulation process of the process specification is discussed, so the effectiveness of the new method is verified.

Key words: multivariate statistical process control, process industries, process specifications, soft hyper-sphere