Home » Kaizen in a Machining Cell: A Practical Improvement Event
Kaizen in a Machining Cell: A Practical Improvement Event
Contents
Use a short, focused Kaizen event to improve one repeatable problem in a machining cell, such as long setup time, a recurring tool search, or parts waiting for inspection. Start with a measured baseline and a problem the team can influence. This page gives you an event plan and a worksheet; its sample figures are illustrative and do not describe a real Mechanitec customer.
A focused improvement cycle for a machining cell. Mechanitec instructional diagram.
A documented example from NIST MEP describes Lynn Welding using value-stream mapping, kaizen teams, 5S, visual management, smaller batches, and setup reduction. The public case supports those methods, but it does not substantiate the anonymous percentages previously presented in this article. Read the NIST MEP Lynn Welding case.
Choose one problem the cell can test
Write the problem as an observable gap, not a proposed solution. “Operators need a better tool board” prescribes an answer. “During 12 observed setups, the tool setter waited an average of 14 minutes for a verified cutter” states a condition the team can investigate.
Choose a part family and a machine with enough repeat work to observe. Include the operator, setup technician, planner, maintenance or quality support when the problem crosses those roles. The supervisor removes barriers and protects time; the people who perform the work help define and test the change.
Illustrative Kaizen problem statement and boundary
Field
Example to replace with observed data
Part family
Two aluminum valve-body variants on one machining center
Observed gap
Median changeover is 52 minutes across 12 qualifying changeovers
Desired condition
Reduce avoidable setup waiting while maintaining first-piece approval and safe machine guarding
Boundary
From last accepted part of the old job to first accepted part of the next job
Measures
Setup minutes, first-piece acceptance, schedule attainment, and safety or quality exceptions
The times in this worksheet are fictional examples. Do not use them as targets. Define the start and stop events before measuring so every shift counts the same work.
Observe the work before proposing a fix
Confirm the standard. Check the current setup sheet, tool list, fixture, program revision, and first-piece approval requirements with the authorized owners.
Watch representative changeovers. Record each task, who performs it, whether the spindle is stopped, and what causes waiting. Do not infer all setups from one unusual job.
Separate work from delay. Note internal setup tasks that require the machine to stop and external tasks that could be completed safely beforehand.
Verify causes with the team. Use a simple cause-and-effect list, then check the leading causes against observations and records.
Test one controlled countermeasure. Examples include a verified preset tool cart, a labeled fixture location, or an approved checklist. Keep engineering and quality approvals intact.
Do not remove guards, bypass interlocks, change a qualified process, or substitute an unapproved cutting tool to improve a time measure. A safety or quality concern stops the trial until the responsible owner resolves it.
Run the event and record the decision
Five-day event plan with accountable outputs
Day
Team action
Output
Owner
1
Confirm scope, baseline definitions, customer requirements, and safety constraints
Approved problem statement and observation plan
Facilitator with operator and supervisor
2
Observe normal jobs, setup steps, queues, and exceptions
Time-stamped task and delay log
Operator and setup technician
3
Validate causes; select a small countermeasure
Cause evidence and approved trial plan
Cross-functional team
4
Trial on a representative job while protecting existing controls
Trial record, actual times, and exceptions
Operator; supervisor protects schedule
5
Review results and decide to adopt, adjust, or stop
Decision, standard owner, and follow-up date
Process owner and quality/engineering as required
A short event does not prove that a change will work across products, operators, or shifts. Continue measuring for a defined follow-up period and include the conditions where the trial did not apply.
Build a baseline the team can reproduce
Use a defined observation window, such as ten consecutive production days, and record the conditions that affect the operation. A useful log identifies the part family, material, machine, shift, operator qualification, order quantity, tool condition, program revision, and whether the run was normal or interrupted. Record actual start and finish times. Separate planned breaks from unplanned waiting, and separate setup work from cycle time. If a shift has several product variants, keep the observations by variant instead of blending them into one average.
Before the event, agree on the exact meanings of each measure. Setup time can begin when the last accepted part from the prior order exits the machine and end when the first approved part from the next order is accepted. Queue time can begin when a job is ready for deburring and end when work starts. Rework should count pieces that require another operation, not every inspection touch. Keep the numerator, denominator, sample size, and exclusions with every percentage.
Ten-day baseline log for a machining or deburring cell
Date / shift
Part family and lot
Cycle or setup minutes
Queue / WIP
Quality result
Condition and source
Record each production day
Use a stable family and identify mix changes
Record observed values; mark downtime separately
Record queue age and physical quantity at the same time
Accepted, reworked, rejected, and inspected counts
Machine, shift, tool, revision, and reason for exclusions
Summarize the median and range along with the average. The median describes a typical observation when an unusual repair or rush job creates a long tail. The range shows whether the process is stable enough to compare. Do not remove a slow observation because it weakens the result; record its verified cause and show the result both with and without it only when there is a defensible reason.
Check that the sample represents the work you plan to change. If only one operator, material lot, or easy part variant appears in the baseline, the result cannot establish that the method works for other conditions. Add observations or narrow the event boundary. Ask the operator to validate the sequence and record any step the time study missed, such as retrieving a fixture or waiting for first-piece approval.
Use an A3 to connect causes to tests
An A3 is a compact problem-solving record. It does not replace detailed process, safety, or quality documents. Keep it focused on the condition, evidence, cause, test, result, and next decision. A 5 Whys discussion is useful when each answer can be checked against the work; stop asking “why” when the next answer is speculation or when several causes need separate evidence.
One-page Kaizen A3 fields for a cell improvement
A3 field
What to record
Evidence that supports the entry
Background and boundary
Part family, operation, customer or next-process effect, and start/stop points
Router, schedule, observation, or approved drawing
Current condition
Baseline values, sample count, variation, queue, and exceptions
Time-stamped log and physical count
Target condition
A measurable change with a due date and guardrails
Demand, capability, safety, and quality requirements
Cause and countermeasure
One verified cause and the smallest controlled test
Repeated observation, record check, or test result
Trial and follow-up
Owner, approval, jobs included, result, reaction plan, and review dates
Trial record and revised controlled standard if approved
Rank actions by the risk they address, the evidence for the cause, and how safely the team can test them. A low-cost idea is not automatically the best first test. If a change affects a qualified process, customer specification, machine guarding, inspection method, or engineering release, involve the authorized owner before the trial. Record the approval reference and the limit of the test.
Illustrative deburring-cell example
Illustrative example — not observed plant performance. A fabrication shop receives an average of 45 brackets per shift for deburring. The usable production window is 420 minutes after scheduled breaks. The demand takt is 420 ÷ 45 = 9.33 minutes per bracket. During a ten-day baseline, the team records a median deburring cycle of 9.8 minutes, 32 minutes for a family changeover, a median queue of 18 brackets, and 5 reworked pieces per 100 inspected. The cycle is slightly slower than takt, but that comparison alone does not prove the cell is the only constraint. The team checks downstream capacity, product mix, and first-piece approval delays before selecting the test.
Observation shows that the operator searches for a verified belt and fixture after the prior lot is complete. The team does not assume that tool searching is the only cause: it records the task sequence and confirms the search on several representative changeovers. With maintenance and quality approval, the trial stages an identified belt and approved fixture before the changeover, adds a point-of-use condition check, and retains the existing inspection and guard requirements. The standard does not authorize a different abrasive or a changed edge specification.
Across the next ten qualifying changeovers, the illustrative median setup falls from 32 to 21 minutes. The median cycle falls from 9.8 to 9.1 minutes after the team removes a repeated handling step and verifies that the new sequence does not damage the part. The cycle is now below the 9.33-minute demand takt by 0.23 minutes, or about 14 seconds per part. That small margin is not proof of sustainable capacity: variation, breaks, downtime, mix, and rework still affect output. The team also records 2 reworked pieces per 100 inspected and a median queue of 14 brackets. It reports the sample counts and conditions rather than claiming a permanent percentage gain.
For each lot, reconcile quantity released to accepted, reworked, rejected, and remaining WIP. If 100 pieces enter inspection, 96 are accepted, 2 need rework, and 2 are rejected, then 96 + 2 + 2 = 100. The counts must match the physical containers and traveler. Do not count the two rework pieces as accepted output until they pass the approved inspection. If the quantities do not reconcile, pause the comparison and resolve the record before drawing a conclusion.
The event result is a decision, not a victory statement. Adopt the staged-tool method only if operators can follow it on each shift, the tool identity remains controlled, the part meets its drawing requirements, and the measure remains stable during normal work. If the cycle improvement disappears on a difficult variant, separate the variant and investigate its setup requirements rather than applying the average to every job.
Plan sustainment and a safe reaction
Before the trial, name the process owner and agree who can revise each controlled document. The operator may own a point-of-use check; maintenance may own tool-condition limits; quality may own an inspection acceptance rule; engineering may own a drawing or program revision. A visual board can show the current status, but it does not replace the controlled record or authorize a production deviation.
Review the new method after one week, 30 days, and 60 days. At each check, sample the same part family and shift conditions, confirm that required setup tasks still occur, and ask operators where the method creates delay or risk. If a guard, fixture, tool, or product condition changes, use the site’s change-control process to determine whether the standard needs a new approval. If quality or safety deteriorates, stop or contain work under the applicable site procedure and notify the designated owner.
Kaizen sustainment checks and reaction rules
Check
Owner
Trigger
Response
At shift start
Operator or cell lead
Wrong, missing, or damaged setup item
Hold the setup; replace or verify the item through the approved process
Weekly
Supervisor and operator
Cycle or queue moves outside the agreed review band
Check job mix and conditions; observe the work before changing the standard
30-day review
Process owner with quality and maintenance as needed
Repeat defect, rework, safety concern, or loss of method adoption
Contain affected work and restart problem solving with verified evidence
60-day review
Department owner
Results hold across normal shifts and product mix
Close the event or define the next specific improvement problem
Use this start-now checklist:
Name the part family, machine, and operation the event will cover.
Write the problem as an observed gap and define when measurement starts and stops.
Assign an operator, facilitator, supervisor, and any needed quality, engineering, or maintenance owner.
Collect ten representative days of cycle, setup, queue, rework, and exception data.
Select one countermeasure that addresses a verified cause and preserves approved controls.
Define the result, sample count, guardrails, and adopt-adjust-stop decision before the trial.
Record the first result and the next review date in the controlled action register.
Use a one-page experiment record
Kaizen experiment record to copy into the team’s controlled form
Prompt
Record
What condition did we observe?
Describe the operation, job family, shift, and sample count.
What evidence supports the cause?
List observations, timestamps, records, and disagreements.
What change will we test?
State the single change, responsible owner, and approval needed.
What must not worsen?
Define safety, first-piece acceptance, defects, and delivery guardrails.
What happened?
Record results for each trial, including failures and conditions.
What is the decision?
Adopt, revise, or stop; name the owner and next review date.
For a repeatable five-step workplace check, use the 5S audit checklist. For mapping the full material and information path before choosing an event, see the value-stream mapping guide.
Check whether the change holds
Compare the same measure, boundary, part family, and operating conditions before and after the trial. Report the sample size and range as well as the average. A median can help when one unusual changeover would distort the average. Track quality and safety guardrails beside setup time so a faster setup does not hide a new risk.
Assign one process owner to maintain the approved standard and review the result after the change has run across normal shifts. If the result regresses, return to observation and find the changed condition. Do not label a one-time result as a permanent productivity gain.
