EV LiDAR mounts custom CNC fabrication

• 1
  What a LiDAR mount actually has to hold
• 2
  Datum strategy decides more than tolerance does
• 3
  Thermal drift is a material problem before it is a machining problem
• 4
  Vibration control starts with the first bending mode
• 5
  Machining features that keep the mount stable
• 6
  From CAD to an inspected mount
• 7
  When custom CNC is the right call, and when it is not

A LiDAR mount is not a bracket. It fixes the sensor's optical reference frame to the vehicle body, and the sensor's point cloud is only as stable as that frame. A roof pod that rotates 0.1° under braking swings a return at 100 m by roughly 175 mm. That is the difference between one object and two, and the tracking software spends the next few frames deciding which.

So the mount carries three jobs at once: it locates the sensor repeatably, it resists the vibration spectrum of a moving car, and it moves heat away from the sensor housing without bending. These jobs fight each other. Stiffness wants mass and section depth. Thermal matching wants thin walls and a low expansion material. Vibration damping wants compliance, which is exactly what kills angular stability.

Custom CNC fabrication enters the picture because the mounting interface is rarely a catalog part. Sensor suppliers change bolt patterns between firmware generations, roof geometry changes between trims, and the bracket is usually the last part frozen before tooling. Cutting from billet gives you the geometry and the material choice while the rest of the vehicle is still moving.

• 1
  LocatingTwo dowel pins plus a machined flat beat four clearance holes every time.
• 2
  StiffnessFirst bending mode should sit well above the body's 20–200 Hz excitation band.
• 3
  ThermalSolar load on a roof pod can reach 80 °C surface temperature in summer testing.
• 4
  ServiceThe sensor must come off and go back on without losing calibration.

A drawing that says ±0.005 mm on every feature is a warning sign, not a good sign. Angular error at the sensor comes from the stack between the optical reference and the vehicle reference, and only a few features sit in that stack. Machining everything to the tightest grade raises cost and does not improve boresight. Machining the right three features does.

The practical datum scheme is a primary flat for the sensor foot, two dowel holes for rotation about the vertical axis, and a machined pad for pitch reference. If those features are cut in one 5-axis setup, the angular relationship between them is set by the machine's rotary accuracy, not by the sum of three separate fixtures.

Where teams get into trouble is mixing datum schemes. The CAD model is referenced to the vehicle body, the fixture is built from the sensor interface, and the inspection report is written against the drawing. Three different origins. The parts pass inspection and still point the wrong way in the car.

• 1
  One setupCut all angular features in a single 5-axis operation where size allows.
• 2
  Hard datumsPrefer reamed dowel holes over slots for the rotation reference.
• 3
  Inspection matchWrite the CMM program against the same origin as the CAD model.

Aluminum expands about 23 × 10⁻⁶ per °C. A 200 mm roof rail that sees a 40 °C swing grows roughly 0.18 mm. If one end is hard-bolted and the other is also hard-bolted, that growth goes into the rail as stress and lifts the sensor pad. The fix is not a tighter tolerance. It is one sliding or slotted interface with a defined preload direction.

Material choice sets the floor. Aluminum 6061-T6 is the default: easy to machine, good stiffness to weight, available in plate. Where the coefficient of thermal expansion matters more than weight, 4130 or 17-4PH stainless drops the growth to roughly half. The penalty is density and cutting time, so it is usually reserved for small sensor plates rather than long rails.

Invar and similar low-expansion alloys exist, but they machine poorly, cost a lot, and are rarely justified unless the sensor itself is thermally compensated to the same reference. Match the mount to the sensor's internal compensation, not to the lowest number in a handbook.

• 1
  6061-T6Default for sensor plates and long rails up to 4,000 mm.
• 2
  7075-T6Higher strength, adequate for thin stiffening ribs.
• 3
  17-4PHUse where CTE and corrosion both matter; expect longer cycle times.

Road input, motor whine and cabin acoustics all feed the roof structure. A mount whose first bending mode sits inside that band will amplify the input and blur the point cloud. The design target is to push the first mode above the excitation range, typically well above 200 Hz for a compact sensor pod, and to keep the sensor's own mass out of the resonance.

Rib layout does most of the work. Tall thin ribs in the bending direction add stiffness far more efficiently than thickening the base plate, and they can be machined from the same billet without adding joints. Pocket floors between ribs should stay thick enough to avoid drumming, usually 3–5 mm depending on span.

Damping is the second lever. A machined aluminum mount has very little internal damping, so any resonance that survives the stiffness work will ring. Elastomer pads under the sensor feet help, but they introduce a compliance that must be accounted for in the angular stack. Measure the assembly, not just the part.

• 1
  Rib directionRun ribs along the load path, not across it.
• 2
  Wall thicknessKeep pocket floors at 3–5 mm to avoid local modes.
• 3
  Fastener preloadSpecify torque values; slip at the interface reads as angular drift.

Long monolithic rails are the hardest parts in this family. On a 4,000 mm rail, thermal growth of the workpiece during roughing is often larger than the tolerance being chased. The practical approach is rough, stress-relieve, then finish in a second setup with the part at room temperature. Skipping the intermediate step is the most common cause of a rail that bows after two weeks in the car.

Fixturing matters as much as the cut. A long rail held only at the ends will deflect under its own weight and under cutting load. Support it at the same points the vehicle will support it, and use the same torque. When the fixture and the vehicle disagree, the finished part is accurate and the installed part is not.

Surface finish is functional here, not cosmetic. Sensor feet and dowel bores are usually specified between Ra 0.8 and 1.6 μm for stable contact. Cosmetic faces can stay as-machined. Anodizing adds a few micrometers and can round a dowel bore edge, so masking or post-machining of the bore after coating is worth writing into the drawing.

• 1
  Rough then finishLeave 0.3–0.5 mm for the finishing pass after stress relief.
• 2
  Bore after coatingReam dowel bores after anodizing to hold the fit.
• 3
  Contact facesSpecify Ra 0.8–1.6 μm on sensor and dowel interfaces only.