About an old paper, victim of peer review and editors choice, and about these

Glauco Masotti

Preface
This document is two sided, or perhaps three sided. On one side I am presenting you a paper, which I wrote, that was submitted to CAD for publication, in December 1993. Regrettably, the paper was rejected instead, and no other attempts to publish this work were made later, thus, since then, it rested in a drawer of mine, before it come into my hands again, in recent times. What I am proposing you, its thus material which have been seen hitherto only by a very limited number of persons. From my point of view its like a piece of jewelry lost for a long time under the sand, which come to light again. Well, this perhaps is not your opinion. From your point of view it may be relevant the fact that, after almost 19 years, this material, most likely, doesnt represent anymore the state of the art in this topic. As I abandoned this field many years ago, I dont really know how much the technology have evolved and in which direction, however, what I can tell you is that what is described in the paper worked quite well, and constituted a remarkable improvement, in terms of easiness of use and performances offered, over competitor commercial systems. Thus I think you could still find valuable content in it, which could also be applied in other contexts (e.g. in nonlinear optimization). In particular, certain general principles, which are highlighted in the paper, should maintain their validity across time. The second side, developed in this preface, is the main reason for this publication, and it is a reflection, which I am asking also to you to do, on the process which brought to rejection of the paper, which ends up with the questions was that decision right?, is this really the best that we can do?, and do we really need peer review and editors?. There could be also a third side, as I said, which could be entitled my life rowing countercurrent, but this most likely wont interest you, and thus it will not be developed here. The paper, object of discussion here, presented a software module for the modeling of assemblies, in which the various parts were positioned via the interactive, sequential specification of mating constraints. The paper describes the capabilities of the module, as well as the problems encountered in its development and the solutions found. It must be said that what is presented in the following pages is a greatly revised version of a previous document, which was submitted for publication to the same journal (CAD) in September 1993. The fate of this first version was determined by just two referees! One of them lamented that the paper was confused and unclear, and questioned many parts of it. Some observations were reasonable, and contributed to make the second revision clearer, but others made me think that he/she didnt understood much of what I wrote, perhaps really my fault, but a referee should make some efforts to understand the point of view of the author, isnt it? The other referee observed that the author (me) didnt discuss the vast literature on part mating and didnt explain how his work improves on it. I still think that this should come out also from an attentive reading of the first version, but probably I hadnt emphasized enough certain aspects, I must admit instead that the first remark was justified, too little space was dedicated to a discussion of known, related works. He/she also complained that the algorithm was demonstrated only with simple examples. Also this observation, was reasonable and acceptable, and contributed to improve the second version. After all it was not my business to write papers, working for industry and not for academy, thus inexperience may have accounted for some ingenuity in preparing the first version. However the referee continued observing that similar algorithms already appeared in the literature and that Just using an off the shelf algorithm from Press does not count as publishable work. I cannot describe what I felt reading this. A sensation of real anger, like when you are victim of a profound injustice! What!? Months of work, of studies and trials, several new ideas developed and thousands of lines of code implemented and I just used an off the shelf algorithm? If this, presumed expert in the field, and honored peer, made such an observation, it means that he did not go into the details of my work! And, in this case, the details are what makes the difference. In fact, if its true that in the first version I omitted an adequate discussion of known approaches to the problem, I think to have filled this lacuna in the second version, so that you can read what the state of the art was. At that time, all known works on the subject reported applications to more or less simple examples, generally even simpler of those used by me in the first version, but none reported of a truly spatial (3D)
July 2012 Page 1 of 38

kinematism, like the last example that you can see in the second version of the paper. This result was possible also because an adequate set of mating constraints was used, several of them introduced for the first time. In my paper I also made a clear distinction between one-object and n-objects positioning problems, as well as between committed and step-by-step approaches in defining the position of parts. These concepts, were not clearly defined in previous published works. The referees may have considered these distinctions as irrelevant, but taking one approach or another, changes the user perspective a lot. Also the mathematical problems to be solved change, as in the various cases we may have underconditioned or overconditioned problems, however the algorithm described can solve both. Known works typically used linearized algorithms, none reported, as I did, a detailed study on exploitation of at least second order terms, which instead is essential to guarantee convergence in all possible cases. None of these papers discussed in detail the difficult cases which may arise in practice and how to overcome these, providing in all cases the desired solutions. The use of second order terms or the computation of a direction of negative curvature to escape local maxima or saddle points, in this context, was an original contribution, as well as the particular exploitation of a polynomial approximation for the object function. Moreover, none of the known papers emphasized (or just reported) the importance of transforming the space of variables in a uniform domain, and of rescaling the equations to guarantee a uniform and balanced contribution of each term to the objective function. Just this detail accounts for at least an order of magnitude in speed of convergence! The algorithm which I described in the paper is peculiar also for other contrivances, which, altogether contribute to the final result, e.g. the temporary relaxation of some constraints, to escape local minima, the use of temporary attractor points (a newly introduced concept), to escape unwanted solutions, the use of random perturbations to determine the residual degrees of freedom. Saying that my algorithm was similar to those already published, IMHO is thus like saying that horses are similar to dogs! (After all they both have four legs, isnt it?). I also must emphasize that, considering all these aspects, saying that I just used an off the shelf algorithm from press, is absolutely scandalous! Like saying that Mr. Sacher, in conceiving and crafting his famous cake, did nothing more than just taking the ingredients off the shelf! Anyway, as I said, the two referees also made reasonable remarks, which were taken into account for preparing the 2nd version of the paper, which I carried out with the support of my head of division. This second version resulted in practice in a quite different paper. The text size doubled, stepping from 11 to 22 pages, an adequate discussion of known approaches was provided, details of the algorithm were explained further, recurring to more complicated and meaningful examples, illustrated by a completely different set of figures. A final demonstration of the capabilities of the software module was shown, with the realization and subsequent animation of a spatial kinematism. It is so that, in December 1993, we submitted the new paper for reconsideration of publication. But all our efforts resulted vain. This time in fact it was the Editor who directly rejected the paper, without submitting it to a new process of evaluation! He justified this decision saying that he couldn't reconsider a rejected paper for publication after such a short time! Therefore he didn't take in any account the fact that the paper was so much different from the previous one, and also the fact that we were so quick in preparing the second version, was paradoxically valued negatively rather than positively! He also said that he had a long queue of accepted paper waiting for publication, and this certainly played against us. He also suggested sending the paper to another journal, or retrying with CAD in about a year. My consternation, in reading these words was great, and this was the sentiment also of people of my group. For a private company, the interest in publishing some of its work, through one of its members, is in gaining visibility and recognition, CAD played a dominant role at that time, thus publishing elsewhere would have not been the same. Moreover those were times of rapid developments in the field, thus the appeal of the paper one year later could have been much lower. Therefore, we end up not pursuing further our attempts for publishing the work. Despite the rebuff received by the community of experts, that piece of software proved in the field its efficacy. It received constant appreciation from the users, who evaluated it more powerful and simpler to use than corresponding modules of the competitors. It was used also to model very complex machinery (like diesel motors) and remained in service, almost unchanged, for several years.

July 2012

Page 2 of 38

At this point, considering that it was not the first time that I had problems with peer review 1, but also that I am in good company, with the people who had similar experiences 2, I think that it makes sense to ask ourselves the questions which I anticipated above: was that decision right?, is this really the best that we can do?, and do we really need peer review and editors?. I leave to you the answer to the first question; the last two questions instead, deserve some considerations. What could be the alternatives to the usual process of submitting papers to peer review and editors approval? And would these be better? The limits of the traditional approach have been highlighted by several authors, and the debate on this issue is open since several years 2,3,4. The two papers cited are both worth reading, but I want in particular to explicitly cite the enlightening words by Horrobin: The history of many innovations, both in medicine and in other areas of endeavor, indicates that the innovators are often erratic, unsystematic, and difficult to deal with. The quality controllers often regard the work as of poor quality and not worth publishing or noting. Editors must be conscious that, despite public protestations to the contrary, many scientist-reviewers are against innovation unless it is their innovation. Innovation from others may be a threat because it diminishes the importance of the scientist's own work. I don't want to say that all of the considerations above apply to the paper which follows, but I think that they are valid in general and we should be aware of these flaws in the traditional process of paper selection. Well, I do believe that with current technology we are in condition to bypass these procedure and that the benefits in doing so would surpass the inconveniences. In fact we have now available a virtually unlimited space for storing publications. Passing papers through a process of selection makes sense if we have a limited space available, like for traditional journals or conferences, but if the space is unlimited? We also have sophisticated and efficient means for retrieving the contents which may interest us. Thus the solution could be as simple as lets publish everything on the web, or at least everything which passes some minimal, formal criteria for acceptance. This does not mean that papers will not be evaluated! Papers could receive votes, and thus be evaluated by a much larger public than a couple of referees, plus we can count the number of citations. Evaluation and critical comments from a wide public should also contribute, better than a couple of reviewers can do, to revise and improve reports, but, most of all, in this way we can avoid the censorship of peers or editors! So that heterodox, perhaps too ahead of time, or controversial works will still have a chance for appearing. This could be the main road to go. Websites like Scribd or arXiv are interesting experiments in this sense, but I think they are still not up to the task. I thus hope that there will be an evolution in the sense I advocated, and that, sooner or later, one or more websites, with all the required functionality, will emerge, gaining wide acceptance in the scientific community. Returning to the paper, which is presented in the following pages, I am posting the original master, digitized via a scanner, because the electronic version, which I was able to recover from my old Mac, would require extensive editing of mathematical formulas to be restored properly. I only made one correction of a wrong word, plus I edited the figures, to enhance their visual representation, and I associated to each figure the related caption, for your convenience. The Editor of CAD, suggested me to change the title, in case of submission to another journal or to a conference, because Placing mating condition at work was not meaningful to him. Perhaps he was right, the title of the first version was Interactive assembly of parts by means of mating constraints, but, in changing the title for the 2nd version, we intended to stress the fact that this was not an academic, theoretical work, but the description of a fully functional part, actually employed for production in industry. Perhaps another title can be found, capturing better this meaning, but we had not occasion to think about it again.

Glauco Masotti, "Floating-Point Numbers with Error Estimates (revised)", Jan 2012, http://arxiv.org/abs/1201.5975
2

C. W. McCutchen, "Peer review: treacherous servant, disastrous master", TECHNOLOGY REVIEW, Oct. 1991, http://www.technologyreview.com/magazine/magpdf.aspx?id=486 Horrobin, D. F., The philosophical basis of peer review and the suppression of innovation, J. Amer. Med. Assoc., 263(10), 1438-1441, 1990, http://jama.jamanetwork.com/data/Journals/JAMA/9237/jama_263_10_024.pdf
3 4

Peer-review (refereeing), http://www.iva.dk/bh/core concepts in lis/articles a-z/peer%C2%AD_review.htm

Page 3 of 38

July 2012

July 2012

Page 4 of 38

July 2012

Page 5 of 38

July 2012

Page 6 of 38

July 2012

Page 7 of 38

July 2012

Page 8 of 38

July 2012

Page 9 of 38

July 2012

Page 10 of 38

July 2012

Page 11 of 38

July 2012

Page 12 of 38

July 2012

Page 13 of 38

July 2012

Page 14 of 38

July 2012

Page 15 of 38

July 2012

Page 16 of 38

July 2012

Page 17 of 38

July 2012

Page 18 of 38

July 2012

Page 19 of 38

July 2012

Page 20 of 38

July 2012

Page 21 of 38

July 2012

Page 22 of 38

July 2012

Page 23 of 38

July 2012

Page 24 of 38

July 2012

Page 25 of 38

July 2012

Page 26 of 38

July 2012

Page 27 of 38

July 2012

Page 28 of 38

Figure 1. Assembled slider crank mechanism.

Figure 2. Parts in unassembled initial position.

July 2012

Page 29 of 38

Figure 3. Assembled configuration after sequential positioning of parts (without inserting the pins), before closing the kinematic loop.

Figure 4. A configuration which correspond to a local maximum of the objective function.

July 2012

Page 30 of 38

Figure 5. Here a saddle point will be encountered.

Figure 6. This configuration will end up in a local minimum.

July 2012

Page 31 of 38

Figure 7. An unwanted optimal solution.

Figure 8. Trying to flip the parts by means of a side-of constraint will trap the solver in this local minimum.

July 2012

Page 32 of 38

Figure 9. Desired 4-bar mechanism configuration.

Figure 10. Intermediate configuration of the four bars during the assembly sequence.

July 2012

Page 33 of 38

Figure 11. Collapsed solution upon imposition of the last fits constraint.

Figure 12. Flipping the configuration using an "attractor point".

July 2012

Page 34 of 38

Figure 13. Examining residual possibility of motion of parts.

July 2012

Page 35 of 38

July 2012

Page 36 of 38

Figure 15. Animating the crank slider mechanism, searching for maximum drive angle allowed.

Figure 16. A two-stage cardanic joint example.

July 2012

Page 37 of 38

Figure 17. Changing the configuration parameters requires the simultaneous repositioning of seven objects.

Figure 18. Animation of parts displacements.

July 2012 Page 38 of 38