An end-of-line test station for boards, from sequence to yield report.
YAML test sequences with Setup/Main/Cleanup, limits and retries, run against simulated boards through a simulated SCPI DAQ. Per-unit reports, yield, failure Pareto and Cpk.

Scope: Simulated boards and instruments; modelled on NI TestStand concepts, not TestStand itself. Personal portfolio project, not affiliated with any employer.

---

The problem

On a production line the question is not only pass or fail, but why units fail, how often a retest rescues them, and whether a rail is drifting towards its limit. The station answers those after every batch.

Overview

A small end-of-line production test sequencer in Python, modelled on the concepts of NI TestStand. It runs a YAML test sequence against simulated boards through a simulated SCPI DAQ/DMM, writes a report per unit and a station log, and summarises a batch with yield, retest rate, a failure Pareto and Cpk.

I built it to learn how a production test station is structured: sequence groups, step types, limits, retries, retest flow and the statistics a test engineer looks at after a shift.

What it does

• sequences/board_eol.yaml: example sequence with Setup / Main / Cleanup groups, station ID, fixture ID and a serial number pattern.
• eol/sequence.py: parser and validator (unknown types, calls, keys, missing or inverted limits, duplicate names and YAML errors are rejected with the step named), and limit evaluation.
• eol/station.py: PyVISA driver for the DAQ, the step calls, and the per-unit execution: retries per step with attempt count, stop-on-fail or continue per step, Cleanup always runs, an exception in a step becomes an Error result instead of stopping the station, simulated operator prompts and simulated barcode scan checked against serial_pattern.
• eol/dut.py: simulated board with three rails (5V0, 3V3, 1V8), idle current, three components and a firmware version. Unit-to-unit spread is about 0.5 % on the rails. Defect injection: rail out of tolerance, missing component, intermittent fixture contact (reads 0 V at random, usually fixed by a re-seat), wrong firmware, high current.
• eol/daqsim.py: SCPI DAQ simulator vendored unchanged from my daq-testbench. The DUT "drives" a DAQ input via SOUR:VOLT, the station reads it back with MEAS:VOLT?, so measurements carry the simulator's offset, gain error and noise. Current is read as a voltage across a 0.1 ohm shunt on the 0.1 V range.
• eol/__main__.py: batch runner. A failed unit is re-seated and retested once.
• Outputs in out/: units/<serial>_run<n>.json and .md, station_log.csv (one row per step per run), batch_summary.md and .json.

Run

pip install -r requirements.txt
pytest -q
python -m eol sequences/board_eol.yaml --units 200 --seed 7 --out out
python -m eol sequences/board_eol.yaml --units 50 --rates '{"missing_component": 0.2}'

Mapping to TestStand concepts

TestStand	Here
Sequence file, Setup / Main / Cleanup step groups	YAML file with setup, main, cleanup lists
Numeric Limit Test, Pass/Fail Test, String Value Test, Action	numeric, passfail, string, action step types
Comparison types (GELE, GTLT, GE, LT, EQ, ...)	same names in comp, with low, high, units
Step loop / retry on fail	retries: N, attempts recorded per step
Step failure causes sequence failure / skip	on_fail: stop (default) skips the remaining Main steps; continue keeps going
Code modules / adapters	Python functions in CALLS
Process model (serial number entry, UUT loop, report, database logging)	run_batch: serial scan, per-unit run, retest, report files, CSV station log
Station globals	station and fixture ID in the sequence file
Operator interface	not built; prompts are auto-acknowledged and logged

Sample batch summary (200 units, seed 7, default defect rates)

Metric	Value
Units	200
First-pass yield	88.5%
Final yield (after retest)	91.0%
Retest rate	11.5%
Step retries used (run 1)	36

Failure Pareto (first run, first failing step): Rail 5V0 5, Idle current 4, Connector present 3, Firmware version 3, Rail 1V8 3, Rail 3V3 3, ETH PHY present 1, PMIC present 1.

Step	n	Mean	Std	Low	High	Units	Cpk
Rail 5V0	195	4.999	0.0253	4.85	5.15	V	1.965
Rail 3V3	192	3.299	0.0174	3.2	3.4	V	1.9
Rail 1V8	189	1.801	0.00927	1.746	1.854	V	1.899
Idle current	185	180.6	5.7	140	230	mA	2.372

Definitions: first-pass yield = units passing on run 1 / units. Final yield = units whose last run passed / units. Retest rate = units run more than once / units. Cpk = min(USL - mean, mean - LSL) / (3 * s), sample standard deviation, one-sided when only one limit exists. Cpk here uses passing measurements only, because 0 V contact faults and defective units would otherwise dominate the spread; that makes it optimistic. The retry count shows intermittent contacts being absorbed by step retries, which is also how retries can hide a fixture problem on a real line.

Tests

52 pytest tests: every comparison type at and around its limits, non-numeric and NaN values, string and pass/fail strictness, 13 sequence parsing error cases, retry recovery and exhaustion, stop-on-fail vs continue with Cleanup still running, step exceptions, operator prompts and serial validation, yield / Pareto / Cpk on hand computed data, and end-to-end batches (all good units give 100 % yield; five injected defects give exactly 35/40 first pass and 36/40 final). CI runs them on Python 3.11 and 3.12 and uploads a sample batch as an artifact.

Limitations

• Simulated DUT and simulated instruments only. Nothing has run against real hardware or on a production line.
• This is not NI TestStand and does not read TestStand sequence files. It borrows the concepts to show I understand them.
• Limits and defect rates are illustrative, not from any real product.
• No operator UI, no database, no MES integration, no parallel or batch fixtures, no measurement uncertainty or guard banding, no Gage R&R.
• Not affiliated with Axis Communications or National Instruments.