The current engineer atmosphere is buzzing about Neural Networks. The downside is that Neural Networks do not have a structure that is easily understandable and interpretable. When we can understand the physic mechanisms, Neural Networks should not be your first choice, in my opinion.
The links below take you directly to the start of the pertinent question(s). Among the two controls authors, they agree we can do the same or better than Neural Networks. Gene Morelli is from NASA. Stephen Boyd is from Standford.
The counterpoint to my viewpoint is Chris Rackauckas’s Automated Discovery of Mechanistic Models via Universal Differential Equations. Sometime in the future, Neural Networks may dominate. For now, I don’t see it this way and think controls engineers should use other methods if they can understand their system.
Soderstrom and P. Stoica, System Identification, Prentice Hall. 1989 (available for download from the author’s website)
Brunton, Steven L., and J. Nathan Kutz. Data-driven science and engineering: Machine learning, dynamical systems, and control. Cambridge University Press, 2022. Available here: https://www.databookuw.com/
I have created a list of system identification books via WorldCat. WorldCat allows you to search libraries across the wold. WorldCat Book List.
System Identification at Linköping University. Not an online course but gives a recommended schedule of lectures and readings. To preserve this information for the future, a screenshot is below of the schedule.
Technical Seminar: ‘Quest for Aircraft Stability and Control’ by Eugene Morelli. Testing of full-scale aircraft in flight to validate or improve predictions obtained through wind tunnel testing or CFD calculations is expensive and time-consuming. Being able to test aircraft stability and control using real-time onboard computations is now within reach and has far-reaching implications for efficient flight testing, control system design, aircraft health monitoring, pilot training, aircraft fleet maintenance, and safety. Aired May 18, 2007.
Introduction to System Identification. An introduction to System Identification with the System Identification Toolbox in MATLAB. Lennart Ljung is the presenter.
Linear system identification video lecture – YouTube playlist. The video course “System identification – linear theory” introduces the student to linear system identification techniques based on data-driven modeling techniques. The focus lies on experiment design, understanding and selecting model structures, and computing the best model within the model set using prediction error methods and model validation.
System Identification. Videos by Brian Douglas. Throughout the series, see the system identification workflow through several different examples that highlight the importance of data collection, model selection, model fitting, and model validation in MATLAB with the System Identification Toolbox.
System/Observer/Controller Identification Toolbox (SOCIT) This one is written at least initially by Jer-Nan Juang; his book is linked above. This is free but export-controlled. SOCIT is a collection of functions, written in MATLAB language and expressed in M-files, that implements a variety of modern system identification techniques. For an open-loop system, it features functions for identification of a system model and corresponding forward and backward observers directly from input and output data. For a closed-loop system, SOCIT identifies an open-loop model, an observer, and a corresponding controller gained directly from input and output data.
System IDentification Programs for AirCraft (SIDPAC) This one is written and maintained by Gene Morelli. His book is linked above. It is free and allowed for foreign use (not export controlled). Written in MATLAB, SIDPAC is a collection of over 300 programs that perform a wide variety of tasks related to system identification applied to aircraft. SIDPAC includes tools for experiment design, data analysis, kinematic consistency checking, static and dynamic modeling, simulation, numerical integration and differentiation, smoothing, filtering, finite Fourier transformation, statistical modeling and evaluation, optimization, parameter estimation, model accuracy quantification, model validation, and more.
Github Repositories
Mataveid System identification toolbox for GNU Octave and MATLAB was written by Daniel Mårtensson. It is mostly based on Applied System Identification.
Transmission Line Method (TLM) or Transmission Line Matrix method is an elegant technique used to solve lossless or lossy wave equations. Here I am gathering together resources on the topic since I have found it is not well studied in the US.
This is the book I found easiest to read from a Mechanical Engineering perspective. The book starts by looking at the wave equation in 1D. It uses the ideas of boundary conditions to explain the scattering effect that occurs in TLM. It is an easy read for an engineer who has studied the wave equation before.
Papers
Webpages
http://dandadec.com/what-is-tlm/ – This page is Donard De Cogan’s page. Donard’s website has lots of information on TLM such as conferences and papers.
