# SYN-DOE-M01 Rev A / Continuous improvement and experiments **SYNTHETIC DEMONSTRATION.** All blanks, settings, loss records and responses are fictional. No real employer/customer study, implementation, machine qualification or approval is implied. Version improvement-1.0. ## Problem, scope and responsibility Investigate losses in SYN-OPS-01 and decide whether a tested machining setting warrants independent confirmation. This is an independently authored portfolio demonstration. Same component SYN-VS25, machine SYN-MC-01, station SYN-ST-M01. Experimental blanks SYN-EXP-V001–032 are separate from existing SPC lots and production-history SYN-RUN-01. ## Loss register and evidence Fixed eight-skid teaching window; independent of editable weekly demand. Window SYN-CI-OBS-01. All entries normalize person-minutes by the same eight skids. They describe different loss mechanisms, not a capacity calculation. Travel is model-derived; the other entries are fictional attended event segments. No total, elapsed lead time, causal attribution or realized savings is inferred. Editable weekly demand does not rescale this fixed teaching snapshot. | Loss | Person-min / 8 skids | Person-min / skid | Evidence kind | | --- | --- | --- | --- | | Loaded travel | 3.50 | 0.4375 | Modeled walking time / CI-E01 | | Setup / changeover | 62.00 | 7.7500 | Synthetic event log / CI-E02 | | Attended waiting | 48.00 | 6.0000 | Synthetic event log / CI-E03 | | Rework | 45.00 | 5.6250 | Synthetic event log / CI-E04 | ### CI-E01 / Loaded travel 210 loaded metres from the baseline layout at eight skids; 30 trips. Assumed 1 m/s loaded walking speed: 210 / 60 = 3.50 person-minutes. Empty return, pickup and congestion excluded. Question: Can a feasible layout reduce loaded travel? Compare equal-demand layouts in the material-flow workspace. Route distance is a modeled exposure, not a timed observation. ### CI-E02 / Setup / changeover SYN-SET-01: 34 attended person-minutes; tool/fixture preparation. SYN-SET-02: 28 attended person-minutes; setup verification. Two changeovers in the fixed window. Question: Which preparation can occur before the machine stops? Observe and separate internal/external tasks, trial staged tooling, then verify first-off quality. The machining experiment does not estimate setup savings. ### CI-E03 / Attended waiting SYN-WAIT-01–04: 12, 9, 16 and 11 person-minutes, each for one idle operator. Fictional fixture, kit and review readiness delays; exclusive logged segments. These are not the NCR chronology or elapsed lead time. Question: Does a readiness check remove avoidable waiting? Trace missing readiness information, test a pre-start checklist and measure idle time. Existing material holds still require independent NCR disposition. ### CI-E04 / Rework SYN-RWK-01–03: 18, 12 and 15 attended person-minutes for three fictional repair events. Inspection notes raise dimensional consistency as a hypothesis. No causal attribution, scrap count or link to existing SPC lots is established. Question: Can process settings improve dimensional margin without excessive cycle time or energy? Use the randomized machining experiment to test the setting mechanism. Confirm on new parts before claiming any reduction in historical rework. ## Controlled design Full factorial 2³: A feed (0.12/0.20 mm/rev), B cutting speed (120/180 m/min), C coolant (6/10 L/min). Four independent cuts per corner; 32 runs; eight fitted cell means. Negative/positive codes denote low/high. One independent blank per run; repeats are not repeated gauge readings on the same part. Every run is randomly ordered with a seeded Fisher–Yates shuffle. Settings are teaching assumptions, not machine operating advice. - Same fictional material grade, blank geometry, fixture, tool type and calibrated measurement method. - One independent blank and a reset/stabilized setup per run; four replications of all eight settings. - Completely randomized run order; no tool-wear trend, carryover, blocks or nuisance drift in this generator. - Cycle response is cut-only elapsed seconds; energy is machine-plus-coolant Wh during that cut. ## Reproducible synthetic generator Seed 41973. 32-bit LCG: state=(1664525×state+1013904223) modulo 2³² using Math.imul; U=(state+0.5)/2³². Fisher–Yates consumes uniforms first. Each independent standard-normal draw uses Box–Muller sqrt(-2 ln U1) cos(2πU2); no cached second draw. For each randomized cut, response noise is generated in error/cycle/energy order. Full precision is retained in JSON/CSV; UI rounds only for display. For A,B,C coded ±1: - Diameter error µm = 4(3+5A+4B−4C+3AB−4AC−3BC+ABC) + 6Z₁. - Cut cycle seconds = 82−9A−6B+2C+2AB + 1.2Z₂. - Cut energy Wh = 34−2A−3B+3C+AB + 0.6Z₃. - Diameter mm = 25 + error/1000. Nonconforming if outside [24.90,25.10] mm. Noise is independent and Gaussian, with constant variance separately within each response. The coefficients encode a teaching mechanism. This is not a learned ML model, actual tool physics, calibrated energy model or manufacturing dataset. No block, drift, carryover, curvature or extrapolation is modeled. ## Analysis and uncertainty Balanced ±1 factorial columns are orthogonal. Fit intercept plus all seven effects: A,B,C,AB,AC,BC,ABC. Grand mean is the intercept; coefficient βj=Σ(xj y)/32, effect=2βj. With a full factorial cell model, fitted corner means equal the four observed-cut means. Pure-error SSE=Σcells Σreplicates(y−cell mean)². df=32−8=24; MSE=SSE/24. t(0.975,24)=2.063898562. Pointwise two-sided 95% intervals: - Effect half-width: t×2√(MSE/32). - Corner-mean half-width: t√(MSE/4). - One new independent cut prediction half-width: t√(MSE(1+1/4)). The run plot shows independent cuts by randomized order, not a process time trend. The effect plot includes all seven terms and uncertainty. Two-factor interaction plots average over both levels of the third factor; each plotted point represents eight cuts. A three-factor interaction can remain, so an averaged plot is not a complete setting prediction. Replication supplies error variance without assuming higher interactions vanish. Confidence assumptions are deliberately met by the generator and must be investigated for real data. ## Engineering constraints and decision Default mean-error guard ±40 µm, cut-cycle maximum 80 s, energy maximum 36 Wh. Guard is a tighter study target around 25.00 mm, not the drawing specification. Accept the screen only when the entire mean-error interval lies within the guard, the new-cut prediction interval lies within drawing bounds, and mean upper bounds for cycle/energy meet selected limits. Quality guard editable 20–80 µm; cycle 60–110 s; energy 28–45 Wh. No eligibility is shown for a failed screen. Lowest observed mean cycle, then energy, ranks eligible tested corners. No continuous or global optimum is claimed. Pointwise intervals are not simultaneous confidence guarantees after eight-setting selection. Eligibility means worth confirming; it is not process release, guaranteed defect prevention or capability qualification. The experiment does not estimate historical rework reduction. Cut seconds cannot replace the existing 6 min machining-plus-inspection assumption or establish capacity. ## Proposed implementation and verification The visitor records a rationale and saves the selected eligible setting and study limits as a draft. Updated limits do not silently rewrite that proposal; the current screen is shown separately. Proposed SYN-SW-M01 Rev B and SYN-CP-M01 Rev B remain unapproved. Checks indicate review of the following future plan; execution stays **not run**, with empty evidence. - **Measurement and setup:** Verify gauge resolution/bias, fixture, tool condition, material and stabilized coolant. Record actual setup and calibration evidence. - **Independent confirmation:** Run 12 new blanks at the proposed setting, across three fresh setups. Check every diameter against 24.90–25.10 mm; compare mean error, cut time and cut energy against the saved study limits. Investigate any failure; no automatic approval. - **Sustain and control:** Collect at least 100 subsequent individual measurements across representative tool life and shifts. Assess stability before estimating capability or revising control limits; define a hold/investigate response to signals. - **Standard work and review:** Propose SYN-SW-M01 Rev B and SYN-CP-M01 Rev B with settings, setup checks, first-off inspection, reaction plan and traceable change approval. Keep current work and quality holds until reviewed. The SPC workspace remains a monitoring example, not confirmation data for this setting. Existing lots, control limits, NCR disposition and FAT acceptance are unchanged. ## Measurable study result and lessons At default limits, 2 settings pass the study screen; S8 has the lowest mean cut time. Exact results, intervals and runs are in improvement-manifest.json. The fastest raw setting fails the quality screen. Good engineering decisions combine multiple responses, experimental uncertainty and implementation evidence. Historical loss notes suggest a question; controlled comparisons test a specific mechanism; confirmation and sustainment test whether a proposal transfers beyond the synthetic study. ## Sources - [NIST: Two-level full factorial designs](https://www.itl.nist.gov/div898/handbook/pri/section3/pri3331.htm) - [NIST: Effect uncertainty from replication](https://www.itl.nist.gov/div898/handbook/pri/section5/pri5971.htm) - [NIST: Randomized run sequences](https://www.itl.nist.gov/div898/handbook/pri/section3/pri331.htm) ## Limits - Synthetic equations encode the mechanism. Experimental effects describe this generator, not validated real machining physics. - Only eight tested corners are compared; no interpolation, global optimum, production capability or machine recommendation. - Pointwise 95% intervals assume independent normal errors and equal variance within each response. They are not simultaneous selection guarantees. - No real implementation, confirmation trials, cost savings, customer results or approval evidence exists. - Cut seconds cannot replace the factory's six-minute machining-and-inspection work assumption or establish capacity.