Chapter 3.2 - Technical Concepts

In this chapter we turn to some existing technical concepts to use in the design process. They are drawn from various fields of engineering, the sciences, good design practice, computer networking, and a desire for sustainability. They are particularly relevant to the design and operation of self-improving systems, but the rest of those fields are still important. Knowledge from many areas is needed to design any technical system, whether self-improving or conventional.

1.0 - Conservation of Flows

In the sciences and engineering, Conserved Properties are ones which do not change as a closed system evolves. Matter and energy are conserved properties under ordinary conditions. Another way of putting this is they do not appear from or disappear to nothing. Since the value does not change over time, we can write equations where the total at one time is equal to the total at another time, although the components that make up the total may have changed.

Open systems allow inputs to enter and outputs to leave. We can also write equations for open systems, where the total now is equal to the total at a previous time, plus or minus any items that flowed into or out of the system in-between. These equations can then be used to solve problems or answer questions about the design and operation of the system.

We can apply these kinds of conservation rules beyond just matter and energy. They can apply to all the items which go into a system, including intangibles like data and human labor. When designing or operating a system, any changes in the internal amounts must come from a flow in or out, or by internal conversion of items into other types.

The conservation rules apply to individual parts of a system as well. As a consequence, internal flows within a system cannot have detached endpoints. That would imply the flow's contents appear from or disappear into nothing. Flows must therefore connect between parts of the system, or connect outside of it. So if a production process consumes more raw materials and energy than is embodied in the products, the conservation rule tells us there must also be some waste material and energy outputs. An efficient design will identify these waste flows, and try to make use of or minimize them rather than ignore them.

2.0 - The Systems Approach

Figure 3.2-1. Open system with boundary and input and output flows across the boundary.

In Physics and engineering, an open system is simply a bounded area which has flows into and out of it, and a variable amount of contents inside at any point in time. For design purposes a System is defined as a functionally, physically, and/or behaviorally related group of elements that regularly interact or are interdependent.

Systems are distinguished from their surroundings by a System Boundary (Figure 3.2-1). The boundary is a mental construct drawn around the collection of elements for the purpose of design and analysis. It is not a physical boundary like a fence. Instead it defines certain elements as being part of the system, and everything else as being outside the system.

A system may have parts that are physically separated, such as a home satellite TV network where the ground stations, satellites, and home equipment are very far apart. These parts, however, are functionally related and interact, so their design is usefully done as a whole, rather than as unrelated items. A system also has a time dimension during its existence, and will evolve over that time.

The Systems Approach tries to optimize the design, not just at a single point in time, but across the whole time it exists. So it considers everything from the design stage to final disposal when the system is obsolete. Flows across the system boundary therefore are distributed in time as well as location.

From the conservation of flows rule, any matter, energy, data, or other item which crosses the system boundary results in an equal change in the quantity inside the system. Flows going in increase the amount, and flows going out reduce the amount. An example of a very simple system is a personal checking account. Deposits increase your balance, and payments reduce them. While flows do not appear from nothing or vanish into nothing, parts of a system known as Functional Elements can transform inputs into different outputs. For example, a machine tool can convert metal bar stock and electricity into a finished part, metal chips, and waste heat.

Each part of a system, when considered individually, also must obey the conservation rules, and can be considered as a smaller system, or Subsystem. By breaking a system down into smaller and simpler ones, then applying the conservation rules to each, you can trace and account for everything moving in and out of a system, and all of the flows and operations that happen inside.

Complex systems are too hard for people to design as a whole. Dividing them repeatedly into simpler subsystems will reach a point where each can be individually designed, then combined to make the whole. A common example of this is designing a house. The various parts like carpentry, electrical, and plumbing can be designed individually by people with different skills, then combined.

3.0 - Modular Design

In general we don't know ahead of time everything an evolving system will need, or all the things we will want it to do. We would like to easily make changes as the need arises. A traditional factory with large fixed pieces of equipment, like a blast furnace, can produce a lot of output, but isn't flexible. It is most efficient when run continuously at full speed. If demand is lower, it gets shut down periodically and then sits idle.

If the blast furnace is used to 80% of capacity, 20% is unused. If demand grows to 120% of capacity, a second furnace is needed, again leaving unused capacity. For US industry as a whole, Capacity Utilization has averaged 79.6% from 1972 to 2020, with a low of 66.6% in 2009 and a high of 85.2% in 1988-89. To better match capacity to need, we would prefer a design which can change in smaller increments, rather than large fixed steps that are expensive.

One way do this is by Modular Design. This sets envelopes and interfaces so that different elements can work together without special changes. It is commonly used in a number of fields. For example, building construction in the US uses standard increments of two feet for lumber, so that pieces will fit together with less cutting and waste. Personal computers use standard sockets for processors, memory, and add-in cards, so that a variety of parts can be added as desired, without changing the case or motherboard.

For an evolving factory, we can extend this idea to the building layout, with standard modular locations for equipment and their utilities. We can also apply the modular concept to vehicle and robot design. A vehicle would have standard chassis sizes, and standard locations to mount motors, wheels, construction implements, or robot arms. In both cases, this makes it easy to expand or modify things as needed.

If the volume of production or the mix of outputs changes, having the equipment in smaller modules allows you to better match the output to the demand. Alternately you can replace a given size machine with a larger one that occupies more modules on the factory floor. In addition to making equipment layout more flexible and efficient, modular design is easier. Rather than having to update the design each time a change is made, you only have to design the individual modules once, then use as many copies as needed.

Module Sizes

A defining characteristic of modules is using specific sizes and spacing so that items will fit together easily. A popular example of this is the Lego building toy (Figure 3.2-2). It uses an 8 mm horizontal by 9.6 mm vertical spacing, which is suitable for toy construction.

Other products need a wider range of sizes, from smaller to much larger. Industrial design often uses Preferred Numbers which are nearly equal multiples in a geometric series. An example is to use a base scale that starts with 1, 2, 5, and 10 mm, cm, tens of cm, and meters. This gives a range of scales that are 2 or 2.5 times larger than the last. The actual module sizes can then be 1-6 scale units. This gives overlapping size ranges, since 6 times a given scale is more than twice the next larger one. With a 5 cm base scale, then 5, 10, 15, 20, 25, and 30 cm would be module sizes in each dimension. A hardware component would then fit any number of modules of the appropriate size.

This approach can handle a wide range of sizes while still being modular. Mounting holes, for example, can follow such a system, being spaced some number of module sizes apart. A vehicle chassis, motor, and wheels would not require special parts if they all use the same size modules and hole spacing. In a factory layout you might use 1 meter modules, with two meter aisles to move items, a 1x1 space for support columns, utiliy lines every 5 meters, and the overall building being 20x50 meters. Different sized equipment is more likely to use the space efficiently because they are all even multiples or fractions of a module.

Standard Interfaces and Protocols

Paradoxical as it may sound, a standard interface can make things more flexible. Electrical outlets are standardized, but they allow plugging in almost any device at any location. There are already standards for connecting automated equipment, such as the Common Industrial Protocol, and for exchanging design data, such as STEP (ISO 10303 ). Such standards can be used as-is, and can take advantage of existing items that use them.

A full set of standards for a factory would apply to both physical and data items. Physical standards include placement and types of utility connectors (power, data, water, etc.) so each machine can be "plugged in" without custom design. It would also include standard floor loads and other building features. The PCI Express computer expansion bus standard is an example this sort of modular system. The physical, power, and data connectors are defined so any expansion card can fit any slot of matching type.

Modular Automation

Factories now now commonly use automated machines and robots, plus some amount of human labor, to make end products. Such equipment is inherently flexible. You can choose to make a single part or a whole production run, or change products by changing the machine instructions. For Modular Automation we can extend this idea to include the factory as part of the product. If the assembly of the building, and setting up of factory equipment locations, storage, and other items can be automated, then the entire factory can become more configurable and flexible according to changing needs.

We do not expect 100% automation of these tasks, especially at first. To the extent we can implement it, though, it can dramatically improve productivity and self-expansion. With standardized modules, the tasks would be repeated many times across different factories, and therefore be worth the extra effort to automate.

Transportation is already largely modular and standarized with Intermodal Shipping Containers. These fit as units on ships, rail cars, and trucks designed to handle them. So one way to automate factory changes is to use container-sized bases with modular mounting points built in. Shipping Pallets are also often standardized to fit within containers. So smaller equipment items can have pallet-like bases secured to the mounting points on the larger container-size bases.

With known and consistent locations, it should be easier to automate equipment shipments, rearrangement, and expansion. Containers are designed to interlock and be stacked on ships. This feature has sometimes been used for Container Architecture on land. In principle you can make a container-size frame to serve as a building wall, then fill it with roof truss, roofing sheet, and support column sections. If standardized, setting up or expanding a factory building could also be automated.

4.0 - Local Resources

The use of locally available resources was once a fact of life. High transportation costs made it too expensive to move a lot of goods, so people mostly used what was available nearby. New methods like railroads and large cargo ships dramatically lowered transportation costs. It became feasible to extract high-grade resources from the rare places they are found, then move them long distances to where they are wanted. A prominent example is petroleum, which is only found in abundance in certain locations, and is shipped around the world.

Such high-grade resources are by definition finite: they only make up the top end of the abundance distribution. They tend to be exploited first because they bring the highest return for the effort to obtain them. If the resource is still in demand, then people must necessarily turn to lower-grade sources. Counteracting this to some extent are that not all resources are discovered at once, and extraction methods improving over time. These can delay needing to turn to lower grades, but the Earth is finite. At some point the highest-grade sources run out.

It was efficient in the past to extract concentrated energy sources like wood and fossil fuels. The easiest such sources have been tapped, requiring more work to get what is left. Extraction can also cause local environmental damage, and burned fuels deposit excess CO₂ in the atmosphere, causing unwanted climate change. The extra work and problems encourage moving away from these sources. Petroleum products are currently the major source of power for modern transportation. Alternatives need to be low cost, or they will drive up the cost of everything that that needs to be moved. That includes fossil sources needed for other purposes besides burning for energy.

We see that past methods of extracting energy and materials are not sustainable. over the long term. Using lower-grade sources, that can be found in many locations, helps address this in two ways. First, since they are found in many locations, the average distance to move them is smaller, requiring less transportation energy. Second, the lower concentrations are more abundant and less prone to run out. For example, today only ores above 25% iron content are worth mining, and those only occur in certain places. However iron makes up 5% of the entire Earth's crust. Learning how to extract it from common minerals in would vastly increase the sources and decrease needed transport.

Renewable energy sources like solar and wind are widely dispersed across the Earth. So the average distance to the point of use is low. Since 2011 (wind) and 2016 (solar) they have also become the lowest cost utility-scale sources. So it makes sense to substitute more energy to extract lower-grade nearby ores than to keep extracting high-grade ones from far away. One such local source is waste materials. The energy to recycle them can be lower than new production, and recycling avoids the costs and space needed for disposal.

5.0 - Cyclic Flows

Even low grade ores are not ultimately sustainable if mass flows are linear. Linear means matter flows from a source such as a mine, is used one time within civilization, and then is disposed of as waste. So the resource, no matter how abundant, gradually gets depleted, and wastes accumulate over the long term.

The alternative is to use cyclic flows. These mimic how nature operates, where most materials get recycled many times. An example is the finite amount of fresh water available. People and natural processes convert waste water back to fresh by treatment or evaporation so it can be used again.

Used materials have some advantages over newly-extracted raw sources. They often occur near where new products are wanted, since that's where old products were used. So transportation distances can be short. Their quality as inputs can be quite high, and require less energy to process than newly-mined ores.

Recycling is enabled by sufficient energy, which, unfortunately cannot be reused. So-called renewable sources only arrive in usable form on human time scales, such as the Sun rising daily or a forest regrowing. The ultimate sources of solar, wind, hydroelectric, tidal, and geothermal energy are nuclear fusion, gravity, and radioactive decay, which are finite. But they will take millions or billions of years to run out. This is so much longer than human time scales that from an engineering standpoint we can treat them as unending sources.

Designing an item or an entire system for recycling should be more efficient than adding it after the fact. Production processes can be integrated to make use of wastes from other processes, and to use recycled materials. It should also bring some cost benefits by reducing the need for mining and processing of raw materials. For example, a rusty iron pipe is still a much higher grade of ore than most iron mines supply. So the market for iron and steel scrap, and the processes to use it in new products are well developed. The same has happened for materials like paper, glass, aluminum, and others.

Recycling also reduces the need for waste disposal. If recycling is efficient enough within an automated factory, there is even the possibility of taking past wastes from outside sources and converting them to useful products, thereby helping to clean the rest of the world. This would make economic sense if the wastes can be acquired cheaply enough.

6.0 - Distribution of Operations

Coordinated tasks and processes happening in more than one location are called Distributed. They are common in large organizations with multiple offices, stores, warehouses, and production sites. This concept can also be used by self-improving systems such as seed factories.

We can categorize distribution by where the buildings and equipment are, and by where the control is. Physical distribution is straightforward. A particular site either has all the buildings and equipment, or some subset and the rest is at other places. An integrated factory grown from a starter set can be the opposite of distributed relative to conventional factories. They can concentrate more production steps and equipment types than is usual into a Unified Operation.

At the other extreme you can have single machines spread all over the world, but coordinated electronically. Control of the physical items can be any combination of local or remote, and manual or automated. The four possibilities are therefore direct human operator at the equipment, remote operation by humans, local computer/automation at the equipment site, or remote automated control.

Remote Operation

Before the electronic age, only one of the options was practical: manual operation of tools and machines by people in the same place. The coordinating people at a distance was difficult, and transportation costs were relatively high. So it made sense to locate offices and production tasks in centralized locations. This is why large factories and office buildings were and are still common.

Electronic devices like smartphones, personal computers connected to broadband networks, and robots with vision and force feedback now exist. Such technologies enable operating in a more distributed way, with people separated from each other and from the equipment. It is not required that people and equipment be separated, but it is an option made possible by new technology. Modern electronic controls and software also add the option of doing some tasks without needing people at all, i.e. automation. Such controls and software can be either local or remote.

The economics of remote operation depend on the cost of using features like telepresence and remote control relative to having people physically on-site. Onsite workers typically add costs for items like parking, physical space to access equipment, cafeterias, and bathrooms. Travel to work has significant overhead in commuting time and costs.

Remote access has the potential to reduce the cost of supplying items people need and want. For example, instead of parking spaces for a large number of daily workers, the land area could be reduced to just what is needed for occasional maintenance staff. But the other reasons for putting things in one place, like efficient transfer between production steps, still apply. So the specific circumstances of a project will determine how distributed the design should be.

We can use a hypothetical solar panel factory in Libya to illustrate how these factors relate. Particular desert areas have both plentiful sunlight for power, and high quality silica sand (silicon dioxide) for making the silicon solar cells. Land costs are likely very low. The factory operators, though, might not want to live in the desert, and the costs to support them there would be relatively high. Working remotely in this case would be relatively attractive.

Another reason for remote work is if there are not enough local people with the right skills. Remote work expands the candidate pool to the whole world (or at least the part with fast enough broadband). A third reason is multiple people in different time zones can operate the same equipment by remote control, using it around the clock and making it more productive. A fourth is job flexibility. You may not need a full time person at a given location, but you can assign them to operate different places and switch between them electronically.

When moving to more difficult locations on Earth, and eventually into space, sending the equipment ahead and operating it remotely is a relatively more attractive option. At first, the new location is not able to support people from local resources or provide comfortable living space. Supplies and temporary quarters have to be delivered from elsewhere. Once local support capacity is in place at a given location, more people can then follow. Human presence is not an all or nothing situation. Temporary visits or a small permanent crew can supplement a larger operation which is mostly controlled remotely.

The difficulty of supporting people is why remote control has been used until now for all long term spacecraft beyond Low Earth Orbit. Remote operation is also used for military drones, deep sea vehicles, and some types of mining, where the environment is hazardous or supporting people is expensive. The more difficult and distant the location, the higher the incentive to operate remotely. Continuing improvements in electronics and network bandwidth will allow more effective use of remote operation for more tasks over time.

Self-Expanding Networks

Self-expanding networks are a particular kind distributed operation, consisting of multiple nodes that exchange data, physical resources, and products. What distinguishes such networks from the general background of modern commerce is the elements are designed to work together and make items for each other. The network also uses some level of automated software to coordinate tasks and payments.

Individual nodes vary in complexity (how many tasks they perform) and output capacity. A specialty node may only do one or a few related tasks, while a general-purpose one could perform many tasks. A general-purpose node may have enough flexibility to produce many of its own parts and materials. A local personal or commercial node would have smaller output capacity, while an industrial one would have a high output capacity. These are not sharp categories, but rather descriptions of types within a spectrum of different nodes.

The network as a whole can provide a higher level of self-expansion and self-production than an individual node, because it includes a wider range of processes and products, and more people with different skills. In particular, the products of the network can purposely include designs, parts, and complete elements to establish new nodes, helping the network to grow.

A large conventional factory, like one for assembling automobiles, is normally dedicated to one task. When demand for that particular item falls, the equipment is under-used and people get laid off. Separating production tasks into smaller, more flexible nodes can make it easier to change what they do as demand fluctuates. This requires more computers and software to re-direct tasks as needed, but computers are relatively cheap compared to re-purposing a big factory. A flexible network may therefore have higher utilization and be more efficient overall.