Selected Papers by the Author, Arranged by Subject
General Bio/Bibliography
Home Page
 E.H. Bristol, "EHB II at Foxboro", An umpublished personal history written at the request of the author of a Foxboro 100th anniversary history, Feb. 17, '09.
 E.H. Bristol, "A Control Retrospective; What We Should Have Done", ISA2002, Chicago IL, Oct. 2124, '02.
Unpublished ICL Discussion Notes:
 E.H. Bristol, "ICL and Model Predictive Control and its Variants", November 24, '01.
 E.H. Bristol, "ICL Extensions for Distributed and Duplicated Unit Controls", July 18, '02.
 E.H. Bristol, "After a Fully Defined Control Practice; What?", February 1, '05.
Multivariable (e.g. RGA) Control Related Papers:
 E.H. Bristol, "On a New Measure of Interaction for Multivariable Control", IEEE PTGAC, Vol. AC11, No. 1, January 1966. The RGA is a Schur (or Hadamard) matrix product now more common in engineering matrix references. There are many more recent RGA references in the ChemEng literature. But there is also a least one Mathematical reference as well! I occasionally get questions of how the RGA came into being, included here with the RGA bibliography. There have always been questions about the RGA "theorem" that negative RGA corresponds to a Right Half Plane Zero or Pole. There were many proofs, but I am not sure that the graphic proof here is generally known.
 E.H. Bristol, "The Right Half Plane'll Get You If You Don't Watch Out ", A Favorite Paper: Edited from the earlier version given at the 1981 JACC, University of Virginia, Charlottesville, Virginia, June 1719, 1981, Paper TA7A. This paper addresses the fundamental affect of right half plane zeros in defining the limits to Process Control type control, following these effects from single loop, to multiloop transmission zeros, and into possible nonlinear theory [Unrequited Love #3]. It fleshes out these conclusions with a simple idealized design procedure.
There is also a site note describing recent work addressing old loose ends.
PID Controller and Automatic Control Tuning or Adaptation Related Papers:

E.H. Bristol, "Understanding the PID and Its Tuning", IFAC PID01 Workshop, Terrassa, Spain, April 2000. The PID is the center of traditional Process Control. But I would argue that it also has a special position in fundamental control understanding being based on the first three terms in the Taylor series of any single variable control function. This is basic since all real control is implemented as approximation in terms of processes that are never fully treated analytically. In this sense the PID has not been accurately addressed in any theoretical treatment. Such a treatment is almost certainly going to turn out to be as complex as most real approximation theories. Also, check "PID Controller Tuning"

E.H. Bristol, "Adaptive Control Odyssey", ISA Conference, Philadelphia, PA, October 1970, Paper 56170. This paper has two aspects of interest: It briefly outlines early (1964), otherwise unpublished, identification and adaptation work roughly following the same modeling form as the more recently favored DMC and Model Predictive Control (This work is, to my knowledge, the only adaptive work using this form.) and on a sensitivity calculation which allows the successful adaptation. And it shows why this successful approach is unnecessary overkill, leading to the Pattern Recognition Adaptation of EXACT®.

E.H. Bristol, "Pattern Recognition: An Alternative to Parameter Identification in Adaptive Control", with minor clarifying modifications, from Automatica, Vol. 13, March 1977, pp. 197202, inself a revised version of "Pattern Recognition as an Alternative to Parameter Identification in Adaptive Control", IFAC Sixth Triennial World Congress, Cambridge, Massachusetts, August 2430, 1975. This paper includes a number of interestng variants on the Pattern theme including a concept called Simplex Modeling for addressing nonlinear adaptation generally.

E.H. Bristol, "Experimental Analysis of Adaptive Controllers", 3rd Yale Workshop on Applications of Adaptive Systems Theory, New Haven Connecticut, June 1517, 1983, pp. 249253. This and the above paper introduce the fundamental reasons for the preference for EXACT (Pattern Recognition Adaptation) over standard model based adaptation, using experimental arguments. (No, I'm not against math modeling; see the next paper.) Unfortunately because EXACT was an "Expert System", i.e. magic, people didn't feel called upon to understand its basic thinking. This paper (particularly Figure 5) shows the fundamental problem behind the usual modeling approaches. These usually give only low order approximations to the open loop model not necessarily corresponding to the actual closed loop behavior.

E.H. Bristol, "Moment Projection Feedforward Control Adaptation" (with P.D. Hansen), 1987 ACC, Minneapolis, MN, June 1012, 1987, FA10. This paper describes an extremely effective multivariable adaptive extension to EXACT. It is interesting for combining two methods generally held in bad taste: the method of moments and the projection method. The methods mesh especially well in capturing the low frequency model needed for feed forward control (as contrasted with feed back control). Also interesting is the mesh as a model based design with the nonmodel based EXACT design. I call the design a proof that practicing engineers could spell (A A^{1})^{T} properly! Despite its effectiveness and commercial presence the system has not had the application attention needed to get wide use.

E.H. Bristol, "Process System Adaptation Beyond Math Models", International Workshop on Adaptive Control Strategies for Industrial Use, Lodge Kankanakis, Alberta, Canada, June 2022, 1988. This includes additional discussion relevant to the previous three papers, particularly to Robustness with Nonlinear Processes.
Swinging Door (Patents on: http://www.uspto.gov/; Figures separate under "Images".):
 E.H. Bristol, "Swinging Door Trending: Adaptive Trend Recording?", ISA90, New Orleans, Louisiana, October 1418, 1990, Paper #90493, pp. 749754 in Proceedings. Swinging Door Compression has been of recent interest due to new commercialization of the algorithm, and concern for the impact of the compression on follow on least squares based* computations. Rumor suggests prior commercial use of the algorithm, despite its patents.
 Basic Patent: "Data Compression for Display and Storage", U.S. Patent No. 4,669,097 (Serial No. 789531), May 26, 1987.
 Related Patent on improvement in the form of a convex hull based compression: "Method and Apparatus for Data Compression", U.S. Patent No. 5,774,385, June 30, 1998.
* With respect to this concern, any compression forgetting data points, destroys some information, in this case to achieve the efficient operational representation of data excursions. However, it is also possible that experimental work may have missed a point: The compression was intended to store its results in a way permitting use of the swinging door limiting data points (B and the interpolated D' in the figure) in any further analysis rather than using just the ends (X and X') of the approximating line.
(Data storage capable of restoring both points, and the end points, is only slightly greater than one point per straight line in the compression process.) This further compression and processing is then based on selected real data points not modified values. [In particular, recompressing the compressed data to unchanged limits results in unchanged compression, as would be desired.] The references address how to reoptimize compression to save other kinds of data for special applications.
Alarm Analysis (also related to Abnormal Situation Operation)
The Alarm Analysis papers and patent describe extensions of earlier work (1976 with Harold Wade). Of special connection to the site's ICL subject, the described analysis tools are similarly implemented to ICL logic computations. Efficient computation is facilitated, in this case, by converting all alarm data into a very large bit string, that allows high speed massed boolean operations. In particular, the system allows sorting of all of the alarm data by applying user defined masks prior to tool operations.
The papers also describe a demo of the tools applied to a Styrene Plant example used in other ICL examples. The demo shows the following tools in action:
 Categories, which are lists of alarm tags which allow selecting all alarms whose tags are in such lists to be displayed or analyzed.
 Categorizations, hierarchically ordered lists of subCategories designed so that the subCategories in any such listed Categorization includes all the alarms within its corresponding Category. Groups of alarms can be selected, as being jointly in several Categories, by ANDing corresponding bit masks. In this context, Categories can be viewed as "coordinates" for locating alarms or groups of alarms, each on its corresponding Categorization "axis". The masking allows easy, flexible operator display control.
 History logs and "Battle Ribbon" Trends, the first including the listing of alarms in order of occurrence, the latter showing the history of all alarm state changes, for a given alarm tag, in a color coded "battle ribbon". Thus the different trends of a set of related process variable can be displayed in a single table of trends.
 Patterns, which are Categories augmented with an assumed time ordering and causality among the included alarm tagged variables. A dynamic programming algorithm automatically detects partially matching real life sequences of events, allowing symptom pattern recognition and the anticipation of any consequential problem.
 One Word and Priority Summaries, which allow one to display the alarm state of a given Category in terms of the smallest included Category (including a single alarm tag) which includes all active (or priority active) alarms. This selective display control within a single alarm annunciator space represents one of the most powerful of the tools.
The demo shows the use of Categories and other tools to allow the operator selection and control over his displays without information overloading. I intend to add the demo (implemented as a Visual Basic® interface over a CProgrammed engine) to the site when possible.
In a different way than ICL, the Alarm Analysis Tools involve compilation into complex data structures. Both would require high quality software detailed design and implementation efforts. The demo implementation shows that complexity. The industry seems to believe that it is incapable of this kind of effort. A linked note discusses the situation.
 E.H. Bristol, "Improved Process Control Alarm Operation", INTERKAMA99, Duesseldorf, October 1999 (Also given in the same trip at a London conference on Alarm Analysis)..
 E.H. Bristol, "Models and Support for Alarm Operation; Batch vs. Continuous", ISA99, Philadelphia, Oct. 59, 1999.
 Patent: "Alarm Analysis Tools Method and Apparatus" with Figures, U.S. Patent No. 6,690,274, Feb. 10, 2004.
Teaching Papers/Notes:
 E.H. Bristol, "A Practical Problem". First created as a Chemical Engineering Doctoral Exam Example at LSU in 1977. This single sheet is a distillation of my reaction to university teaching of frequency domain to chemical engineers. The difficulties that students found have since been confirmed by others of my teaching friends. One of the things that I would do if I taught the introductory control course again (apart from Idioms) is work with a lot of "real" data examples like this one (though perhaps less extreme). I would even like to expose the students to spectrum analysis using visual data of "real" noise.
To my knowledge, noone has solved this particular example problem, although it was the only control problem that one of the LSU candidates attempted. The difficulty is its unfamiliarity (And the PhD candidate, though one of the more selfconfident, didn't take my course! Of course the actual course students saw much more transparent examples!). But the issues are basic and simple; to check this, look at the hints and solutions. In any case, my original intended student population not withstanding, a PhD intending to teach prospective working engineers needs enough understanding of the practical implications of the material to solve such a problem!
 Site Page, "Introductory Process Control Teaching with Idioms". This note goes well with the "Practical Problem" discussion. Idioms introduce the big control picture, where the "Problem" refines the detailed frequency teaching methods. I intend to use the material in a "How I Wish I Had Taught the Control Course" paper.
Other Control Papers
 E.H. Bristol, "Basic Control Algorithms", Update of the earlier "Control Algorithms" for D.M. Considine, Process/Industrial Instruments and Control Handbook, 5th Edition. This paper covers all elements of digital implementation of continuous control algorithms for process control, particularly Idiom and ICL related issues.
Control History and Philosophy
 E.H. Bristol, "History of Process Control", Control Handbook, edited by M. Singh, UMIST, Manchester, England, subeditor, Stuart Bennett, University of Sheffield, England, Pergamon, Last date October 29, 1987.
 E.H. Bristol, Special ISA Paper for Historical Session on "The Contributions of the Bristols to the US Instrument Industry", ISA85, Philadelphia, Oct 20, 1985.
 "Riding the Silent Sky"; an old Foxboro Company internal magazine employee relations article. Fun is philosophy in addition to being history!
Earliest Significant Papers:
 E.H. Bristol, "Use of Relaxation Oscillator as a Voltage to Frequency Converter", IEEE PTGCT Vol. CT10, No. 4, December 1963.
 E.H. Bristol, "On a New Measure of Interaction for Multivariable Control", IEEE PTGAC, Vol. AC11, No. 1, January 1966. This is covered in detail above.
 E.H. Bristol, "On a Philosophy of Interaction in a Multiloop World", ISA Chemical & Petroleum Division Chempid Symposium, St. Louis, Missouri, May 1967.
 E.H. Bristol, "A Simple Adaptive System for Industrial Control", Instrumentation Technology, June 1967, pp. 7074. This is the first EXACT related reference.
 E.H. Bristol, "Compensation for Truncation Errors in Accumulative Addition Using Random Methods", IEEE Transactions on Electronic Computers, Vol. EC16, No. 4, August 1967. p.518. This originated as a random method to support integral control without double precision storage of the integrated value. In those days this represented a sizable memory savings in a large system. It worked great in that context but we didn't have the guts to commercialize it.
 E.H. Bristol, "On the RealTime Supervisory Operating System of the Future", IFAC Real Time Workshop, Boston, June, 1975. On top of the general operating system discussion, this paper cites Petri nets and anticipates the Sequential Function Chart; a French engineer came up, at the meeting, to say that they were thinking the same way! The diagram is in fact more general than the SFC in that it incorporates both logic and sequencing. It even allows (graphic) "subroutine" implementation of ladder logic! At the time there was concern for ladder logic "loops" when representing sequencing. I didn't see why it wouldn't be better to design a sequence language directly rather than "misuse" the ladder logic in such a way.