Comparison of four parameter selection techniques [force sensors]

The objective of this research was to find the best combination of nine temperature sensors to model the output from a force sensor caused by temperature variations. The technical approach was to monitor the force sensor output under a variety of temperature gradient conditions but with no other external inputs. From this data, an empirical formulation could be made of the best combination of sensors to use to compensate for the force output due to temperature gradients. We investigated four different techniques to find the best combination of temperature sensors: Principal components (Karhunen-Loeve) decomposition, two traditional optimization techniques (a gradient-based technique and a simplex search technique), and a genetic algorithm (GA). This paper gives a brief description of each technique, describes the results from the four techniques and offers a comparison of the advantages and disadvantages of each