Soft Goods Manufacturing — Istanbul

The estimate is the symptom. The system is the case.

I work with manufacturers and engineering teams in soft goods industries — garment, automotive interiors, medical, and technical textiles — to diagnose and resolve spreading, cutting-room, and production-planning problems before they invest in the wrong fix.

Send a production problem →

§ 00 — Where I work

Everything between design and assembly — between a finished pattern and a sewn garment — is the cutting room. Within it, attention concentrates at the two ends: nesting software and automatic cutting machines. Both attract investment, competition, and feature wars.

To many, the process is invisible — nesting produces a cut file; the cutter reads it. A clean handoff. But behind that transfer lies everything that determines whether the material — the largest single cost in any soft goods product — is used efficiently, cut correctly, and delivered on time.

Between the two ends: machines, people, and materials in constant conversation.
cut order planning · roll management · spreading · quality control · machine data
The signals are all there. Nobody is listening.

§ 01 — Problems I encounter

The spreading estimate says 40 minutes. The floor takes 65. Nobody can explain the gap.

Machine data exists in the controller but never reaches planning.

The cutting room is scheduled manually across machines with different fabric-weight and table-length limits.

The system performs in isolation but fails under roll changes, splices, and real operator conditions.

A new system or process change is about to be purchased — but the underlying problem has not been formally defined.

The planning model assumes linear throughput. The machine does not behave linearly.

§ 02 — Method

Observe

Collect facts, constraints, failures, use cases, and unknowns.

Isolate

Separate symptoms from causes.

Model

Turn the problem into a technical, commercial, or operational model.

Test

Validate assumptions with prototypes, calculations, or field evidence.

Decide

Choose the simplest path: stop, redesign, build, or scale.

Develop

Move from diagnosis to practical implementation.

The science behind these steps →

§ 03 — Case files

Case 01 / 2024–2025

Spreading time prediction under real factory conditions

Observed Planning estimates diverged from actual spreading cycles by an unacceptable margin. No reliable model existed to predict machine throughput across material types, speeds, or spreading modes.

Hidden cause Throughput is not a linear function of speed. End-loss geometry, acceleration ramps, and mode-dependent behavior introduce non-linearities that simple estimates ignore.

Intervention Physics-based calculation engine modeling trapezoidal motion profiles, end-loss geometry, and spreading mode logic.

Outcome RMSE 0.52 s validated against 2,500+ real production cycles.

0.52 s

Prediction error (RMSE) — accurate enough to replace manual estimates in production planning

2,500+

Real production cycles validated against the model

Run the model →

Case 02 / 2025–

Intelligent control for legacy automatic spreaders.

Observed Thousands of automatic spreaders still running — mechanically sound, operationally blind. The original controller was designed to get the job done: any material, any operator, any job. Generic by design. Optimized for nothing.

Hidden cause The machine cannot model what it is spreading, who is operating it, or what the job requires. Decisions offload to the operator. Nothing is recorded. A system that runs the machine but does not know it.

Intervention ReLay is a brain transplant. Existing mechanics — motors, drives, I/Os — stay intact. The controller — hardware, software, and HMI — is replaced with a purpose-built system that takes direct command of the machine's bones, muscles, and nerves. Two design mandates: run correctly without operator-dependent inputs; report what is happening. The transplant also delivers what the original never had — work order management, roll tracking, splice guidance, waste measurement, throughput analytics. An unskilled operator now runs the machine like a veteran with an industrial engineer overseeing every move.

Deployment Piloted at a garment manufacturer with production facilities in Turkey and Egypt.

Outcome A machine that knows what it is doing — and records it. The gap between planned fabric allowances and what actually hits the floor becomes a number, not a habit. Performance data sourced from the control layer: not reconstructed, not estimated, not operator-dependent.

Case 03 / 2023–

Cutting room scheduling as a combinatorial optimization problem

Observed Cutting rooms with multiple spreading machines plan job sequences manually — assigning hundreds of lays across machines with different fabric weight capacities, without systematic visibility into total makespan, machine utilization, or order continuity.

Problem class Parallel machine scheduling (Pm ‖ C_max). NP-hard. Constraints include machine-fabric compatibility (GSM), table length, order continuity (lays from the same order should not be fragmented across distant time slots), and realistic job durations that depend on marker geometry, ply count, spreading mode, and motion physics — not flat averages.

Measurement Agent software runs on each machine's on-board PC. It records actual cycle times, roll change durations, defect interventions, splice events, and setup transitions — per ply, per job, per machine. This produces a ground-truth performance dataset that feeds both schedule validation and model refinement.

Intervention GA Planning: a Genetic Algorithm scheduler that encodes the full cutting room schedule as a chromosome. Each gene represents a job-to-machine assignment. The algorithm evolves a population of complete schedules, evaluating each against a composite fitness function that balances makespan, idle time, and order continuity. An event-driven simulator computes realistic timelines — including roll changes and setup — rather than summing nominal durations.

Technical layer

Fitness

F = (T_ref / T_total) − (T_idle / T_ref × 0.1)
With order continuity: F = F_base × 0.6 + S_order × 0.4

Stagnation escape

Adaptive mutation rate — rises automatically when population fitness plateaus. Population restart preserves the elite cohort and regenerates the remainder to break out of local optima.

Machine constraints

Hard-encoded in the chromosome: fabric weight (GSM) determines eligible machine subset. Jobs violating machine capacity cannot appear in a valid schedule — infeasible assignments are rejected at gene level, not penalized post-hoc.

Cost function

Job duration is computed by a physics-based spreading time engine (OCalc) — trapezoidal motion profile, end-loss geometry, mode-dependent behavior — replacing historical averages. This makes the optimizer as accurate as the underlying machine model.

Outcome Near-optimal spreading sequences across a mixed-machine fleet, generated in seconds. Order integrity preserved. Machine utilization made visible and measurable for the first time.

Spreading machines (example config)

18 m

Table capacity modeled

5,000

Generations, parallel fitness calc

§ 04 — How I think

From factory symptoms to technical decisions.

My background spans industrial engineering, manufacturing systems, and apparel CAD/PLM — from developing and deploying pattern design and product lifecycle management software used in production environments, to the full cutting room: spreading and cutting technologies and processes, production planning, and process optimization across garment, automotive interiors, medical, and technical textile applications.

I work at the boundary between machine physics, production data, and planning decisions — where a technically correct answer is worthless if it cannot survive the factory floor, the supply chain, or the business model.

My standard for evidence is: would this hold under real operating conditions?

Cutting room end-to-end Spreading & cutting technologies Physics-based throughput modeling Production planning & scheduling IoT & machine data Apparel CAD & PLM development R&D strategy & decision support Industrial software development Manufacturing process diagnosis Technical product strategy

§ 05 — Submit a problem

Bring a case.

I take on a small number of consulting and advisory engagements. Send a short description of the problem, the system involved, what has already been tried, and what decision depends on the answer.

You will get a fit/no-fit reply and the first diagnostic questions.

Good fit: spreading and cutting room systems, production planning, machine data, throughput modeling, R&D decisions before tooling investment — across soft goods manufacturing: garment, automotive interiors, medical, and technical textiles.

hi@ziyk.io