Computer Assisted Fat Blend Recognition Using Regression Analysis and Mathematical Programming

✍ Scribed by De Jong, S. ;De Jonge, T. H. J. R.

• Publisher: John Wiley and Sons
• Year: 1991
• Weight: 477 KB
• Volume: 93
• Category: Article
• ISSN: 0931-5985
• DOI: 10.1002/lipi.19910931312

✦ Synopsis

The fatty acid profile of a fat blend depends on the blend composition and the corresponding profiles of the constituent raw materials. In principle, the composition of a blend can then be determined via these fatty acid profies. Two approaches have been applied to tackle this blend estimation problem. In the regression analysis approach the identification part of the problem is handled in a brute force manner: all possible combinations of raw materials are considered. Combinations giving suficiently close fits to the fatty acid profile with meaningful, positive regression coefficients are retained. The acceptable solutions are sorted in order of increasing complexity and decreasing precision yielding a survey that may help the expert in reaching a final decision. In the alternative Mixed Integer Programming approach the identification problem is solved more efficiently using a branch-and-bounds search strategy. This allows much larger problems to be considered with up to 40 different raw materials. The two methods are compared and directions for further improvements are indicated.