Skip to main content
"Just give us an answer, please." This phrase gave us a clue to the development of a groundbreaking system called the "assembly-navigation system."
The system automatically generates assembling methods for products that incorporate several thousands of components. It can also effectively convey skilled craftsmanship knowhow, which has been fostered over many years, by using animation and three-dimensional (3D) images.
Developed through trial and error over 4 years, the system should link designing to manufacturing in a seamless manner.
ENOMOTOIn short, it is a system that automatically formulates methods to assemble components using 3D CAD (computer-aided design) data. The generated assembly methods are output in the form of animations and 3D work-instruction sheets.
At Hitachi, the assembly-navigation system is used mainly for directing production of industrial equipment, such as power generators, power controlling equipment and other devices used at power plants, for example. Such products may have several thousands of components incorporated.
When producing such equipment at manufacturing plants, we need to accurately instruct workers as to which components are used and how they should be assembled. As production is often conducted at overseas plants nowadays, such instructions must be given in a manner in which any person can understand and conduct assembly accordingly. We have know-how in a variety of aspects of assembly, and it can be very effectively conveyed through animations and 3D images.
ENOMOTOYes, that's right. Previously, when a new manufacturing plant was constructed overseas, veterans in production were dispatched from Japan and stayed for nearly half a year to coach the workers there on assembling products. It required much time for the veterans to convey know-how to the workers verbally. By utilizing animations for the coaching of assembly, however, the time required for instruction has been reduced considerably.
As for such animations, it used to take a lot of time to produce them because they were created manually using dedicated editing software. The same was true of the 3D work-instruction sheets, and it took quite a bit of work to prepare them, as it required the capturing of CAD images and pasting them to the documents as well as writing various instruction texts, among other things. We hear that, at one of our plants, they successfully received an order but providing instructions for the work occasionally caused a bottleneck to fulfilling them.
People at manufacturing plants wondered, "if you can create assembly methods automatically, why can't you create animations and work-instruction sheets automatically as well?" The assembly-navigation system was born from such wishes.
Figure 1: Application image of assembly-navigation system
ENOMOTOAn assembly method must follow certain rules. First, large components called "base components" are assembled. Then, assembly is conducted with components located lower than others, if there is any vertical order, or with components located backside, if there is any depth.
I believe you can understand this fairly easily by visualizing a scene in which you assemble something yourself. That is, if you assemble something by starting from the front side, what you've just built will interfere with your next movements, right?
Such basic rules are defined in advance as assembly rules. By applying these assembly rules, we first set out to find a method for disassembly.
ENOMOTOYes. To find the method for assembly method, the method for disassembly is first sought. This is a relatively popular way of thinking. That's because if disassembly is possible, assembly must also be possible.
Reverse the rule for assembly, and you get the rule for disassembly. The upper and front-located components are disassembled first, and the base components are disassembled last of all. The disassembly method is found based on this rule and then it is reversed, and the assembly method is attained. Since the 3D-CAD data has models of products that have been completed after assembly is finished ("assembly models"), we use them to disassemble the relevant equipment piece by piece.
ENOMOTONo, it is actually very difficult to identify the disassembly method. That is because there are a number of conceivable disassembly methods in proportion to the factorial of the number of components. This is what is called the "NP-hard" (non-deterministic polynomial times-hard) problem. When the number of involved components surpasses 100 or so, it becomes an enormous calculation that wouldn't be completed in our lifetimes.
Initially, we thought we would list all the disassembly methods and then evaluate them one by one. There were many other people who were researching automatic generation of assembly methods, but the idea of "listing all the disassembly methods" was shared by all. In practice, however, listing all the disassembly methods is not realistic in the actual workplace because the total number of components used in equipment is usually several thousands and can now be tens of thousands.
So we were having trouble coming up with a realistic idea. But our factory people gave us a suggestion by saying, "Wouldn't it be better to list the disassembly methods in the order of the disassembly rules? "We don't need all of the disassembly methods. Just give us an answer, please." Such comments gave us a clue to drastically changing our way of thinking: what we have to do is to present just a single idea—the one that we think is the closest to what they are asking for. So, at that point, we came up with the idea of the "dynamic exchange disassembly order algorithm."
Figure 2: Scheme of the assembly-navigation system
Figure 3: Steps of dynamic exchange disassembly order algorithm
ENOMOTOAs shown in Figure 3, the algorithm is largely composed of three steps.
The first step involves creating an "initial draft of the disassembly order." A single draft is prepared to suggest a tentative order of disassembly in accordance with the disassembly rules. As it is just a draft, it might not be a good one.
The second step involves "creating disassembly motions that avoid interference between components." Disassembly motions (motions to remove components) are created for the components planned to be removed first.
The third step involves "reviewing of the disassembly order." If a disassembly motion is generated in the second step, it means that the relevant component can be disassembled, and the order concerning that component is fixed as is. On the contrary, if no disassembly motion is generated, the component cannot be disassembled, and the order of the component is pushed back.
Let me explain this using the gear wheel in Figure 3 as an example. You can see a cylindrical axle inserted into the gear. As a result, the gear wheel can only move in the direction of the cylindrical axle. However, there are two other components (one at each end of the axle) for fixing the axle in place, and they interfere with the gear wheel, forbidding it to be removed. In other words, it is not possible to generate a disassembly motion for the gear wheel that avoids interference. Consequently, the order of the component disassembly must be switched so that the next component comes before the gear wheel.
By repeating these three steps, the disassembly order and disassembly motion of each component when they are disassembled can be determined. After that, the disassembly order and disassembly motions are reversed to create the assembly method with assembly order and assembly motions.
ENOMOTOWe make that judgment on the basis of geometric constraints. For example, take the case in which a planar surface of a component is in contact with a planar surface of another component. It is called a "planar constraint." In this case, the component can be moved in the normal direction to the surface (i.e., the direction moving away from the two surfaces at a right angle) or in an arbitrary direction along the planar surface.
Please view the movie on the right. In the case of this gear wheel, a "cylindrical constraint" is imposed. The gear wheel can be moved in both directions along the cylindrical axle. When a planar constraint is added, one of the disassembly motions interferes. In this manner, based on the geometric constraints, the directions of disassembly motions and their interference can be estimated. As such, we investigate how we can move components for each pair of contacting components one by one, and calculate the motions of components.
It was a challenge for us to devise a mechanism to extract necessary information, as information concerning "motion" is not explicitly defined in the data. Essentially, the assembly models are to be created within a CAD system by imposing geometric constraints between adjacent components. However, there are many assembly models that simply indicate component locations and do not have any geometric constraints as explicit information. For such models, our system analyzes the interference between all components to check which surfaces of components are in contact or in close proximity, to generate geometric constraints automatically.
ENOMOTOThe process of creating the animations also gave us a lot of trouble. The size of 3D-CAD data is huge, so we convert it into lightweight data called the "lightweight model" before making the animations in order to speed up drawing the graphics. As a result of this data conversion, it frequently became impossible to make a correspondence of components between the 3D-CAD model and the lightweight model when there are multiple components with the same name (such as "screw"). In such cases, a certain component should have moved but instead a different component moved in the lightweight model.
Since there are many components with the same name (as in the case of screws), it is difficult to make a correspondence of components using their names. So we developed a method to make the correspondence of components in terms of their positions and forms. The method compares the data of relevant components before and after the conversion, and recognizes them as the same components if the position and size are the same. This enabled correct motions of components.
ENOMOTOWe started the research in 2006, and the system took shape in 2009. We were allowed to spend as much as four years to develop it, and I thank our factory people for their long-term perspective.
In the early part of development, the created animations were pretty awful; people on the manufacturing site laughed at them. The assembly order was a total mess, and components sprang up from unexpected places. So the animations were absolutely awful, though people just laughed since it was our first attempt.
We tested a number of approaches before we finalized the algorithm in the present form. We did it the hard way to reach our goal. For example, the mechanism of "creating assembling motions of components based on geometric constraints" that I explained earlier is completely different from what we initially conceived.
Originally, this research was based on the results of an international collaborative research project among industry, academia and government conducted in the 1990s, called the "Intelligent Manufacturing Systems (IMS)." However, the approach considered at that time did not take into account the interference between contacting components, and could not extract motions of components having no mutual interference.
Therefore, at the end of the day, we discarded all the approaches that had been tried before. Although we were somewhat reluctant to do so, achieving results was more important for us. If we failed to form something and get it used by people on the manufacturing site, the research would be a waste.
Photo 1: Assembly operation following 3D
ENOMOTOIn an overseas plant*1 of ours, the assembly-navigation system is used to create 3D work-instruction sheets for the on-site production site, where the sheets are displayed on the screens to instruct workers how to assemble products.
I heard that the local people have found the sheets interesting and proactively use them. To my surprise, I was told that the section where the work instructions are given by displaying the sheets has been included in the factory tour course of the plant.
I also hear that, at a plant*2 in Japan, they have a smooth workflow for creating and using 3D work-instruction sheets. After completing equipment designs, they quickly create the sheets using the system. The created sheets are transmitted to the production site, and workers assemble equipment by checking the sheets on monitors.
What is amazing about the domestic plant is that the prepared 3D work-instruction sheets are also used for inspecting equipment and supplying components. They arrange all components in advance based on the sheets, confirming which components are required where. They also use the sheets to keep a log of manufacturing and inspecting events, recording how many seconds workers spend on their individual actions. Apparently, the assembly-navigation system is being thoroughly used.
ENOMOTOThat's right and we are very thankful. We receive comments from people actually using the assembly-navigation system, and we are continually improving it based on such comments. We intend to steadily launch upgraded versions going forward.
For example, the present structure of the system recognizes of the existence of interference only for contacting components. We want to make such recognition possible also for non-contacting components. We want to be able to answer the question, "What is the best assembly method?"
Furthermore, although the assembly-navigation system is currently for products of a size that can be assembled manually, we are conducting research with a theme of saving labor for assembly operations of much larger products, such as the construction of plants and installation of large-scale machinery. The system will have to answer such questions as how to use conveyance equipment like cranes and jacks and what is the most-efficient route for bringing in components. preparations before assembly are indispensable for large products. It is very challenging and we are always struggling.
To sum up, our mission is to provide solutions for making products efficiently and with high quality. We will continue to put forth our utmost efforts to provide the best results by fully utilizing all mathematical algorithms and IT tools available.
(Publication: August 5, 2014)