...coal carrying railcar... That is indeed a high-G environment. Sounds like interesting work.
I also design boards and write firmware. There was always work for me. Even now I have some clients that want to keep me busy. It seems they can't find the hardware/software/system design in one place like me.
Lightning. My favorite.
I've used a bushel of uP's, lately I'm leaning heavily on ATSAM ARMs. I used ATMEGA devices for many years, sprinkle in PICs and some of the wonderful old stuff like Z80s, 6800s, 68000s, and 68HC11s along the way.
I used principally PIC18F devices on the trains. The last iteration used CAN bus to the devices to minimize wiring and improve signal to noise on the rail cars. I ran the clock speed down to 4 MHz to minimize power requirements as well. The research cars had a Timken generator bearing that included a 12 pole tachometer signal to estimate the speed. Wheels are 36" diameter. At 18 pulses per second, I was making enough power to charge the battery and sustain booting the PC104 stack. Initially it was QNX based, but we added an observer/subscriber package that was dirt cheap for Linux and outrageously expensive for QNX. I ported all of the code to Linux from QNX. At the Linux level I had a serial interfaced GPS and serially interface CAN bus controller. An ethernet connected to "access points" that were operating a mesh network based on OLSR. There was also a 16-bit A2D that allowed sampling 100 KSPS against the accelerometer on the bearing adapter. That picked up signal from the cup, cone, cage and roller bearings. I did some local DSP to extract signals from each component. Tri-axial accelerometers were on the bolster at each end with signals in vertical, lateral and longitudinal planes to assess ride quality. I sampled those with PIC18F devices and sent results back on the CAN bus. There was a brake piston position sensor, handbrake actuator, anglecock actuator and cut-lever actuator. It was capable of closing angle-cock values, setting the handbrake and break the car free under complete remote control. All actuators were my designs and implemented with 18F PIC. Back at the PC104, I had a serial connection to a 1xRTT cellar modem to send data collected back to the server in Fairfax, VA. LAT/LON/timestamp/battery condition were transmitted at least hourly when idle and once per minute in motion. What do you do with that data? You tell the customer about 55 kinds of defects that are detectable and allow for maintenance scheduling at the customer shop instead of a breakdown "on the road". There were also temperature sensors on each bearing adaptor in place of the old "stink bombs" so a bearing burn off could be detect immediately instead of waiting for a track-side detector to spot it on 20 mile intervals.
That's the short synopsis. I published papers at the 10th International CAN Conference in Rome and IEEE/ASME conference in Pueblo, CO in early 2005. The whole project was canceled about 2 hours after Obama was inaugurated.