Several techniques to detect and identify systematic biases when process constraints are bilinear
Is Version Of
This work develops and evaluates several new approaches that detect and identify biased measured variables when physical constraints (material and energy balances) are bilinear. The objective of each of these techniques is to develop [alpha]-level hypothesis tests for the detection and identification of measurement bias. Constraints are bilinear when two measured variables (each with a power of 1) are multiplied by one another. Bilinear Constraints are statistically more complex than linear constraints because the normality assumption is no longer valid. A study is presented to illustrate important strengths and weaknesses of each approach under a variety of process conditions and compares these new approaches to the Two Stage Approach and the Linearization Approach. In particular, this study involves varying the significance level, the size and location of the measurement biases, the number of biased variables, and the sample size.