PYTHON · NUMPY

What "manufacturable" means

Given the sensor-mount CAD part, identify at least 3 manufacturability problems and state why each is a problem, each mapped to the correct defect class on a fixed grader seed.

01Challenge

This part passed every check in the Design course; it's geometrically valid and simulates fine. But a factory can't build it as drawn. Find at least 3 places this part will fight the tools that have to make it, tag the offending feature, and pick the reason. You haven't been taught what to look for, so inspect, guess, and tag. Distractors (wrong color, not symmetric) are in the list, so tagging isn't free.

02Model

In CAD you are a god: you draw a shape and it exists. A wall one molecule thick, a pocket sealed inside solid metal, the screen never says no. But a real part isn't drawn, it's reached: a spinning cutter, molten plastic, a jet of metal has to physically get to every surface, and physical things have size, direction, and limits.

Manufacturable means a real process can reach it, form it, and have it survive. Four ways geometry fights the tool: a wall too thin to survive being made, a side with no draft so a mold grips it forever, a sharp inside corner a round tool can't cut, a feature too deep and narrow for anything to reach.

Can a real tool actually make this feature?

03Guided practice

Pass A (worked): confirm the pre-placed tag on the barrel wall. Wall probe reads 0.80 mm; reason = too thin to survive the process.
Pass B (hint): Section the boss; its side walls rise at 90.000°. Ask 'how does this come OUT of a mold?' Tag no draft.
Pass C (independent): find the remaining two unaided: the sharp internal corner (FILLET_R = 0) and the deep narrow unreachable rib (RIB_GAP = 2.5, ~24 mm deep).

04Feedback

PASS WHEN At least 3 distinct true defects tagged, each with the reason class that matches its failed test (survive/form/reach), and zero distractors accepted on seed 1101.

If your run fails, check:

Tag on the barrel wall is right, but reason 'hole too small' doesn't fit: the probe reads wall = 0.80 mm; the problem is thickness (SURVIVE). Re-pick the reason.
'wrong color' isn't a manufacturability defect; color never changes whether a tool can reach or form the part. Remove it and check the side-wall angles instead.
2 of 3 found. One real defect is still untagged. Run Draft check on the barrel: a face at 0.00° off the pull direction can't release from a mold.
Reason 'no draft' is correct, but the draft problem is on the boss side-wall, not the base plate. Move the tag onto the vertical wall.

05Retrieve & space

From Design M1/M2: in the part script, name the variable that sets barrel wall thickness (WALL) and state its value.
From Design M2: which manufacturability test (reach/form/survive) does a minimum-wall constraint protect? → survive.
Within-lesson: for four mini-parts, pick which of reach/form/survive each most violates: discrimination, not pattern-matching one part.

06Mastery & project

In sim, tag ≥3 distinct true defects on the sensor-mount part, each mapped to the correct defect class, with zero distractor tags accepted, on held-out grader seed 1101.

Produces the defect ledger for the carried part: 1.2 chooses a process knowing these problems, 1.3 clears them, and the M5 capstone's yield is only achievable because the part was made manufacturable here.

← Manufacturing Physical AI Choosing a material and a process →

PYTHON · NUMPY

What "manufacturable" means

The full Bench runs the WebGL inspector; here you re-implement its deterministic geometry tests in numpy: measure each feature, classify the defect, and tag it.

You read

the arrays and values already in scope

You change

the code you write in each cell

Fixed

the dataset and the checks that grade you