The last post reflected on the passing of David E. Jakstis who was a friend of the concept of truth engineering (focus of this blog). In that post, there was some description of David's project which dealt with using metals to make critical parts for commercial airliners. In the parlance of the systems approach, KBE (below), David was the domain expert. He knew metals, their uses, manufacturing requirement specification and a lot more. The author of this post was the systems expert applying the KBE methods, in particular, and handling development of the modeling and algorithms behind the "intelligent" decisions. The particular project was RFD (below) that applied the KBE methodology which can be used to explain the motivation for the "truth engineering" as well as to describe its development.
After a brief pause to acknowledge the past year, we will look a little at KBE and RFD. Then, we will show two forging examples. The first is a large part and was of the type usually handled by RFD. The second is more recent and was done with a modern development method and illustrates the end goal which is a part. The example, also, provides a look at the result of improving a process. For those interested, 3D printing came into play in this new way. We were looking at that three decades ago.
Aside
Last year, we saw xNN/LLM systems appear in the world. An example would be ChatGPT, but there are others. With this exposure, we will be able to (can) start to summarize the impact of those systems and how they fit into the total scheme of AI which would include past modes. One of these modes that continues to today is the general knowledge based systems work, sometimes referred to as expert systems. In short, as a consequence of looking at this work, we expect to cover the history of AI in depth. Many others have a similar goal, so we will be able to reference these looks at AI.
Our continuing theme will be integrative. As we look at the motivations for approaches to software and consider details of a particular focus, we always note that tradeoffs had been made. Our goal is to see how these pertain to limits which can be identified and which, once known, need to be respected.
What is KBE?
Knowledge Based Engineering (KBE) came out of early AI and has an engineering focus. There are many varieties to the discipline which looks to raise the level of sophistication of support that an engineer gets from a computer. The variety addressed in this case applies constraint satisfaction to facilitate resolution of difficult choices that come with complex systems development. In this case, we used a Lisp-based system called ICAD. The page on Wikipedia for this system, ICAD (software), like all of Wikipedia pages, has a "Talk" tab.
Aside: The author has been involved in developing both of these pages.
Since ICAD was bought and shelved so that a vendor could push their own product, material is not readily available to show details. We can discuss outputs and results. In this case, the "Talk" tab has a section titled "Real example needed" with a photo showing parts done by the forging process. Let's use this photo next.
What is a forging? RFD?
The photo that was placed on Wikipedia was derived from photos on the site,
The machines that made the Jet Age. In purviewing the site's page, one can appreciate how the old technique of forging metal has kept up with advancement in technology.
Aside: At the site, consider the size of the machinery that is involved. Growth in demand for increased pressure during the forging process is one factor.
David's, and my interest, with RFD (Rapid Forging Design - below) was to support this work with proper modeling, so the focus of our work was the computer and its ways. As the photo shows, one forges to get to a near-net situation. Then, machining, like one sees with the work of a sculptor, gets the part to the desired condition. In modern manufacturing, CNC machines do this work.
With respect to the photo, the top shows the part after the forge step. The net part is in the lower part. The approach reduces waste since the final step has to remove less metal. Too, the properties can be controlled by the design of the forging die (RFD, next sentence). Testing, even destructive types, could be done by adding in tabs at critical points.
For the most part, we had the metals expert, David. We also had an engineer who was familiar with forging science and design. His parametric approach helped define a computer system that allowed views from the design model (CAD and the database controlling the design) to be marked up with values that transformed into instructions to guide the RFD's building of the die.
KBE and RFD
This approach was not accomplished by explicit invocation of rules. Rather, we used model-based reasoning with constraint satisfaction (CSat) as the primary control mechanism. The modelling handled the transforms between the CAD part and the operational views. Then, construction occured which wrapped the CAD "net" part with the envelope of the forging die. A requirement? That "envelope" which represented the form of the die had, additionally, to meet constraints of the solid modeler.
In this type of process, CSat was the adminstrator, not unlike the OS of the computer. But, as well as control, it handled relationships and resolved the explicit and implicit conflicts through resolution which was similar to that used by rule systems. We will provide examples, as we go along. As well as the model and constraints, ICAD acted as a geometric modeler.
That is where my work came in which was keeping representional conflicts at bay. That was a mathematical/modeling problem which can be difficult to solve in a heterogenous environment. We did local modifications of the die geometry to effect agreement (not unlike lining kids up in formation in the early grades as they learn boundaries about themselves and others). The fixup could be done in the background as the approach was applied generally for several reasons, including handling complaints by the solid modeler.
Aside: Since that time, interest in stability of these types of processes has switched attention to more homogeneous modes. But, at what cost? (Aside: several which I will discuss under the guise of truth engineering) In this case, both the exterior and the interior of the forging die were modified; the interior was the boundary of the near-net condition that was expected to result from the forging operation. The project doing this representational work was titled Multiple Surface Join and Offset (MSJO) which encompasses the general problem set that remains full of open issues when one is dealing with natural objects (which are heterogeneous). Hence, truth engineering deals with the issues, known resolutions, uncertainties, tradeoff discussion, and overall management of expectations.
Aside: One of my favorite books deals with open issues in topology. It's hundreds of pages and dense. The motive for the book was to identify possible projects for PhD students. As well, I have a book that merely looks at some of the hugely believed aspects of topology. Look everywhere, and you'll see reliance of understandings of topology that do not necessarily hold up. Doubt me? See messes. I have a litany that I have done from watching industry types run down some perdition-laden path. Anyway, that little book provides examples of counter examples with regard to the decision-supporting notions of continuity, completeness, and more. AIn't developers are culpable of this oversight. So what? Well, I saw this over three decades ago being a mathematical economist working in engineering support from the perspective of advanced computing. There has been progress that is noticable. For any of those, let me come look at what you might have done incorrectly which is a potential disaster waiting to happen. Of course, others are aware, too. Thankfully, the internet will allow proper discussion.
What is the structural part of the above example? It is not identified, but, in terms of the application of KBE, many parts were designed or had their design enhanced by the method. Here is a site showing definitions:
Basic aircraft structures.
We have to ask, what is the future of the forging method? A forging expert provides an
appropriate view:
- "Forging continues to be recognized as the premiere thermomechanical process. Not only to shape metals, metal matrix and metal composite materials, but to refine and transform the metallurgical structure as well. Forging achieves both durable, reliable component shapes and the need for engineered metallurgy to meet specific product requirements."
We can look at another approach that has been offered to replace forging. But, first, let's consider the major claimant of the day who really is problematic at its core (one might reasonably say: fakery factory). One of our goals? Explain what is the problem, why it exists, and what ought we do. And, metal modeling is a great framework to discuss (and to demonstrate - as science in the past did with small experiments) the associated issues.
What is AI?
One thing ought to be clear, AI is not that which relates solely to machine learning. This can be seen by reviewing those earlier projects more closely. This post deals with a problem of major scope which is handling AI (huge, multifaceted affair) going forward by bringing into the discussion insights from past accomplishments which need attention due to their success in performing (resolving intractable problems). They never got attention since they were not seen and were managed in the non-academic environments that are everywhere (doing the marvels that we all expect in our comfortable present).
There is another motive. Looking at the technical aspects from another view. Applications like RFD had their own value even if the scope was local and specific to geometric modelling. Lots of effort goes into building and using systems, in general, both on and by computers. This will not stop. However, much of the work (say Computer Science) is academic. This series will look at commercial efforts that successfully resolved complexity problems much like we see facing and being, somewhat, handled by machine learning (xNN/LLM). But, these were never really made known.
Again, truth engineering will be more widely discussed. Tradeoffs are broadly demanded; that does not mean cutting corners and cheating.
An example of a forging replacement
In the example for ICAD (see Wikipedia "Talk" page), a critical part was used with photos of parts after the forging step and when finished. See this article:
Norsk Supplying FAA-Approved 3DP Ti Parts to Boeing | New Equipment Digest
This photo is a composite of the slides (at Norsk's site). One thing to notice is that this is a much smaller part than the ones shown in the above example of major structural pieces. This smaller part still carries a structural responsibility. Basically, it ties together structural pieces that are fabricated with a "composite" construction. For the larger part, the forging die does one part. In this case, one can put together several of the parts with a die. These parts would be separated and finished as seen in the lower part of the picture.
A major benefit of forging was control of part properties to meet critical needs. But, each part then needs to be freed from the excess material. One constraint in RFD was to minimize this excess. In the below example, the smaller part went to a near-net state using a modern approach, 3D printing. One advancement which allowed this was the "plasma" assisted fusion of metal a layer at a time where the material was extruded with sufficient quantity to accumulate quickly.
Mostly we think of 3D printing, even with
plasma technology, as forming with a smaller increment of material and by providing the net part. In critical parts, though, years of experience has helped establish processes that go to near net with the proviso that known machining steps will do the finishing. This part was a demonstration of obtaining part properties without forging and encourages further work.
So what?
Does the change represented by this example bear on the future of truth engineering? Of course. This example represents the unceasing striving by humans for improvement, albeit there are many factors to bring to judgment in this regard. And, truth engineering was formulated in the time when computational systems were becoming more mature, sophisticated and effective. It framed itself within the interactive aspects that continue to today, even to the situation of the "cloud" and its nebulous state of affairs. Metals and their handling continues to be focal to progress.
All around are many possible avenues for advancement. Yet, what the situation that founded truth engineering allowed us to see still exists, albeit with a more complicated nature. Truth engineering is one factor in a multi-pronged effort at riding the one beast or the several that technology has thrown our way. There are others factors and approaches. Our interest is to get the details expressed for review as well as to foster the necessary discussions and operational choices going forward. An advantage that has accrued? Being non-academic in nature will allow aspects that have more nuance than generalization allows us to consider to be given their due attention.
Remarks: Modified: 01/15/2024
11/13/2023 -- Restatements for clarity.
11/24/2023 -- Spelling (typos), couple of words.
01/15/2024 -- To follow the work, see the TruthEng blog.
