A lean final assembly plant for a midsize business jet. The line moves at takt, the cells build to kanban, andon stops the line. Change the plan and watch it run.
The jet is our own midsize twin (24 m long, 23 m span, aft engines, T tail), not a real type. The plant is designed from a demand target: takt is the available shift time in a month divided by the jets the customers want. At 6 a month on two shifts that is 56 shift hours, 3.5 working days, and the moving line advances one 36 m station in that time, about 0.4 inches a minute.
The final assembly work (4,370 labour hours in 22 tasks, our own estimate) is balanced over the stations so the busiest carries as little as possible, and each crew is sized to finish in 88 % of the cycle. Upstream the wing and fuselage cells build to kanban, two sets of cards each; engines and interiors are bought on cards sized by demand times lead time. Paint has two bays, the flight line five stands.
The model is a discrete event simulation: it moves straight from one event to the next (a task done, a jet at the end of a station, a machine down or repaired, a kit done, a truck in, a shift change). Task times vary (log normal, 12 % spread, 3 % rework); machines fail at random. Automated checks: the line reaches a steady state, meets demand, keeps WIP flat, obeys Little's law, does not overproduce, and falls short in the cases it should.
The equipment is modelled after the real thing and moves by its own physics. The fuselage cell's robots are KUKA KR 500 class arms on linear tracks with a multi function end effector; every hole is reached by inverse kinematics inside the joint limits, and the one pass cycle (clamp, drill and countersink, probe, sealant, insert, release) takes 7.1 s, about 8.5 fasteners a minute. The wing cell's C frame riveters follow Electroimpact's wing panel machines at 12 slug rivets a minute. The 20 t bridge cranes run at the two speed figures of a 28 m span Hak crane (bridge 15/60 m/min, trolley 7.5/30, hoist 2.4/12) with acceleration limits, and the load swings on the ropes as a pendulum when the bridge starts and stops. Fuselage and wing joins are aligned with Leica AT960 class laser trackers (15 µm + 6 µm/m): the positioners move by most of what was measured, the trackers shoot again, until the residual reads inside 0.25 mm; the rounds and minutes it takes are random, as they are on the floor, and are charged to the station. Crews ride a 97 % learning curve. The test flight's takeoff roll is integrated from thrust, weight, lift and drag for a Praetor 600 class jet (two 33.5 kN engines, 21.5 m span) and lands within the published field length.
Take the buffers out so problems surface; pull instead of push; stabilise and standardise before levelling; make the state of the work visible from across the room (andon boards at every station, the plant board at the head of the line); put parts at the point of use (flow racks, supermarkets beside the stations that use them); automate the constraint, not the loudest problem.
Simplified: labour hours, crews, hand fastening rates, mean time between failures and lead times are our own scenario numbers; the robots drill a representative sample of their holes on screen; transporter speeds are assumed (vendors publish capacity, not speed); weekends are left out; only the takeoff roll is flown by a flight model, the rest of the sortie is scripted; no certification, cost or supplier risk. A design study, not a real plant.
Follow on