If complicated physical systems were able to be serviced and repaired completely automatically there would be many advantages. There would be higher productivity and efficiency without people in the loop - we tend to slow things down and are error prone; also people could be freed up to do something less menial; and the systems could scale quickly when more capacity is needed.
Machines today, such as industrial machinery, are designed to be looked after and serviced by people, and it would likely need artificial intelligence beyond our current capabilities to maintain or repair these systems completely autonomously. However it is feasible to design them from the outset to be maintained autonomously; designed in a modular fashion with components easily removed and replaced by another machine, and embedded wired or wireless sensors giving the ability to diagnose faults on all significant parts.
Many parameters can now be sensed with solid-state sensors, manufactured on tiny silicon chips, which can be embedded within functioning machines. If the signatures from multiple sensors relating to each machine function is known when operating within normal bounds, it provides a method for pin-pointing problems with great accuracy. Vibration, temperature, rotation, pressure, distance, voltage, acceleration and structural integrity as examples. This already happens to a certain extent on machines today such as vehicles but it only applies to a small sub-set of vital components.
What we are talking about here is having multiple micro-sensors within every single component and also scattered throughout structural parts. A little similar to the way humans and animals have pain receptors that let the brain know when a part is damaged. Unlike humans, however, the computer would have a complete blueprint for how it functions, which would allow precise diagnosis, and flawless new parts could be fabricated to replace broken ones,
Operations can be assessed in real-time and if there is a failure then the defective parts, or the relevant sub-assembly, can be replaced. In many instances it may be possible to know a failure is imminent before it actually happens due to an abnormal rise in temperature or vibration for example. With the system containing a full three-dimensional schematic with exact positions and extraction paths for every part, a repair machine can swap the part without requiring any human intervention.
Where would the replacement parts come from? Either held locally in a store, or shipped in via automated transport, or even manufactured on demand via additive fabrication or CNC milling.
So the physical aspects of the machines need to be designed with autonomous replacement in mind, with magnetic, RFID or optical cues that can easily be read by a repair robot, and highly modular design of components allowing them to easily be extracted and replaced. For instance, a gearbox that slots in or out as a single cartridge.
This same principle, by extension could allow these self-maintaining systems to become self-building too, which means that scaling up facilities becomes easy. More autonomous farming equipment required? Just increase the scale of the agricultural manufacturing facilities, and now you have a greater output of agricultural machinery. A little simplistic perhaps, but you get the idea.
- The shape of sensors to come
Hitachi's µ-Chip (shown in top image) is 0.4 x 0.4mm in size. With a small antennae (not shown) it is able to transmit a 128 bit ID number. However, in the lower image is a microscope image of Hitachi's new 'RFID powder', currently in development, shown next to a human hair. It is 64 times smaller by area than the original µ-Chip and has the same capability and can be embedded within a sheet of paper . It will not be long before various types of solid-state sensor can be made on a similar scale and embedded within every component of a machine giving unprecedented information relating to its operation.
A quite different approach to self-repairing machines would be to build them out of self-healing materials . It is likely that sophisticated self-repairing machines in the future will utilise both these methods.