I have decided to cease further work on my Elfdroid Sophie because I am unable to repair her servo motors and will need to replace them each time a MOSFET fries (which is likely to happen frequently). Therefore this is too costly an endeavour for me to sustain.
I have attempted repair using the correct solder and soldering-iron head with flux paste, but modern ICs are designed with very small and delicate electrical connections. Even using a head-mounted magnifier and correct lighting, I find it impossible to remove and replace such ICs without doing irreparable damage to the solder-pad connections.
This means that if I forget to change a 0 value in the program for one of her servos, or I input an incorrect movement/acceleration value, it is highly likely to be destroyed. This is very likely to happen because there are hundreds of values in each movement program run on the servo sequencer. Which means paying another £20-30 to get new ones each time I balls-up a program. Each time. This includes any testing and rehearsals.
Also, if I connect one 5V micro-servo to the wrong circuit then the whole lot (8 servos) are fried. Replacing all of these is relatively cheap, but the hassle of deconstructing the animatronic eye mechanism, unwiring, re-installing then re-constructing and re-wiring the eye mechanism is incredibly time-consuming and fiddly.
Long story short - it has become apparent I will need a Herculean amount of effort and a significant amount of money if I am to operate Sophie on a regular basis (on top of the effort and money I already put in to designing and building what I have so far).
This is certainly not efficient and not worth the effort when all I'll have at the end is a novelty singing animatronic chatbot. Far more advanced versions of which already exist and are operated and maintained by professional teams with corporate funding.
Call it excuses, call it quitting. I don't care. I've decided to focus on digital modelling instead, because I get more enjoyment from it and 3DCG is MUCH cheaper. Despite the steep learning curve, it is also significantly easier and less frustrating than actual robotics.
Therefore I have mothballed Sophie and I will leave the CAD of all her parts (both tested and untested) online just in case anyone else wants to use them for their robowaifu or some other 3D printing/electronics project in future.
Sophie is very fun in the half an hour or so that she's operating before I break something. So if I come back to her in future I will likely just scale her back to only a talking head and moving neck. This removes all of her large servo motors and most of the expense and complication. Since her hands had very poor grip and were only used for gesturing anyway.