Speaker:  evin McLaughlin - Controls Engineering Consultant and WSU Alumnus

Title:  Practical Application of Singular Value Decomposition

Abstract: Singular Value Decomposition (SVD) is a method that provides great insight into the workings of complex models. The easiest systems to understand are ones where single inputs link to single outputs: turn the volume knob clockwise on the car radio and the sound is louder; press the brake pedal and the car slows down. Models of dynamic systems, however, usually have multiple inputs that couple to multiple outputs. SVD can be used to understand interactions because it sorts information from most to least important using orthogonal matrices. The matrices and associated singular values contain a wealth of information on how combinations of inputs affect the outputs. For example, consider a model for a permanent magnet motor. What combination of parameters has the greatest affect on the torque/speed curve? Is there a combination of parameters that has little affect on the torque/speed curve? Proper interpretation of the SVD matrices answers these questions. An image processing example is used to demonstrate SVD and its built-in informationsorting capability. Examples are provided using SVD to help estimate parameters in motor models and vehicle models along with the automatic control of complex systems.