Keep in mind, EN 50121-3-2 specifies immunity requirements, including 2kV EFT, 10V conducted (RMS, 0.15-80 MHz, 80% AM 1kHz), and up to 20V/m radiated (RMS, 80-1000MHz, 80% AM 1kHz). Even if your box isn't located in a normally troublesome location, it may still be subject to noise from occasional or proximate sources, like overhead wires (catenary arcing is powerfully noisy), or antennas in infrastructure (railside or station) or others. And who knows where the cables are routed, and by what.
The conducted emissions limits are very lax (99/93 dBµV (QP) for 150-500kHz / 0.5-30MHz respectively, so you're not likely to have problems there, following basic (appnote level) advice at least.
Radiated emissions is somewhat stricter, at 40/47 dBµV/m (QP, 10m) for 30-230 / 230-1000MHz respectively. This is comparable to commercial levels (e.g. CISPR 22 class A), so requires at least as much care as that kind of equipment.
I might be most worried about the IMU connector overlapping the area with the isolated signals. Whether the ground plane solves that, or if the CM capacitance is low or high enough to deal with it, I don't know. The fact that there's several inches of overlap (a resonant length for the upper test frequencies), and that radiated immunity could carry significant voltage (~10s of V) along these paths, is a matter of concern. Perhaps ferrite beads, on as many signals as you can afford (including power and ground, but not shield; and CMCs on differential pairs -- I see CAN there already), might be prudent.
The metal enclosure does give the option that, if the cable shields can be well bonded to it (doesn't have to be galvanically, but should involve very little -- preferably zero -- lead length on the shield connection, and preferably a wide connection made with multiple capacitors in parallel), most of the external noise can be excluded, and radiation needn't manifest as radiation as such, but will only be carried inside the box on those connections (hence the value in solidly [RF-]grounded shields), and can be filtered directly at the connectors, keeping noise away from the bulk of the circuit. In other words, using the enclosure as a ground plane, and all external connections act as ports with respect to it.
Remember that shielding is only as good as the weakest link in the chain. If the cable shields will not be grounded at both ends, then any incident energy upon the cable will find its way inside the shield and thus onto the signals within; it's almost as bad as nothing at all (unshielded).
Likewise, any gaps in the enclosure allow RF energy through; gaps of just an inch or two can transmit low ~GHz, as well as significant amounts (10s to 100s of V) of ESD when struck in the middle of the gap.
Aluminum can be difficult or inconsistent here, as painted and anodized enclosures are common, which will only make contact around the screw holes, if at all. Bare mating surfaces are preferred, or even better, nickel plating (not environmentally friendly, though), and EMI spring stock or gaskets filling the gap is best. There may be compromises including IP6x sealing (EMI-shielding IP65+ rated gasket material is available, remarkably enough).