Gains:
- Ability to determine where to use AI in lean manufacturing and process improvement cycles
- Ability to accelerate value stream map, waste analysis and DMAIC steps with structured prompts
- Ability to apply the discipline of validating AI recommendations with field data and gemba observation
Industrial engineering is essentially a "doing better" discipline: producing more value with the same resources, making waste visible, reducing variability. Artificial intelligence enters this discipline not as a new calculator or magic solution, but as a powerful thinking and drafting partner. When used correctly, it reduces value stream mapping, waste classification, root-cause studies and improvement reports from hours to minutes. When used incorrectly, it produces suggestions that have never seen the field, that are "perfect on paper" but cannot be implemented in reality. In this unit, you will learn where to place AI in the lean manufacturing cycle and how to validate its output.
Where Does AI Stand in Process Improvement?
Let's divide an improvement project into three layers: data collection, analysis/decision, reporting/communication. AI helps with different weight in these layers.
layer
The role of AI
man's role
data collection
Designs the survey/form and organizes the data
Measures and observes in the field (gemba)
analysis/decision
Generates hypotheses, suggests methods, drafts calculations
Verifies with data, makes the decision
Reporting
A3, presentation, summary text draft writer
Controls accuracy and tone
The thing to note is that AI cannot generate the data itself. Only field measurement gives the actual cycle time of a machine, the minutes an operator waits, how long a die change takes. AI organizes, interprets and turns this raw data you bring into an improvement hypothesis.
Tip: I always ask AI “what data do I need?” Also ask: Establishing the correct measurement plan before starting improvement is much cheaper than performing analysis with incorrect data later.
Seven Wastes (Muda) and Classification by AI
Making waste (muda) visible lies at the heart of lean production. The classic seven wastes: overproduction, waiting, unnecessary transportation, overprocessing, inventory, unnecessary movement, and errors. Often "untapped human potential" is added as an eighth. AI is very fast at breaking down a process narrative into these categories.
Role: You are an industrial engineer experienced in lean manufacturing. Task: Divide the following process narrative into 7 waste (muda) categories. For each determination: (1) type of waste, (2) evidence sentence, (3) metric I need to measure, (4) possible rapid improvement. Process description: """On the assembly line, the operator walks 6 meters in each cycle to pick up the part. A queue forms in front of the paint booth for an average of 40 minutes. Approximately in one shift 12 parts are going to be reprocessed. 3 days of semi-finished products accumulate in the intermediate warehouse.
The power of this prompt is that it makes the output measurable. It is not enough to say "there is unnecessary movement"; It should be possible to say "the operator walks 6 meters per cycle, total Y meters in X cycles per day." The AI points out the evidence statement and the metric to measure, you measure it in the field and plug in the number.
Weak Prompt / Strong Prompt
Weak prompt:
Name the waste in this production line.
This produces a list without categories, without evidence, and without measure; Most likely, the common phrases "there may be excess stock, there may be a waiting period" are repeated.
Powerful prompt:
Divide this process data into 7 categories of waste, provide the evidence statement for each, the metric I should measure, and the estimated impact. Classify the impact as "low/medium/high" and write the rationale. Data: {{ ... }}
The difference is that the output can directly translate into an action plan. The second prompt gives a prioritized, measurable table.
Speeding up the DMAIC Cycle with AI
Six Sigma's DMAIC framework (Define-Measure-Analyze-Improve-Control) is a common language in improvement projects. AI helps differently at each stage:
- Define: Draft problem definition, project scope (SIPOC), and purpose statement.
- Measure: Measurement plan, data collection form, sample size logic.
- Analyze: List of hypotheses, fishbones (Ishikawa), possible root-causes.
- Improve: Solution alternatives, ECRS (Eliminate-Combine-Rearrange-Simplify) suggestions.
- Control: Draft control plan, standard work instruction and monitoring KPI.
Create an Ishikawa (herringbone) skeleton for the following problem. Problem: "Surface roughness on CNC machine is above target." Categories: Machine, Method, Material, Human, Measurement, Environment (6M). Write 3-4 possible causes under each category and indicate what data I should collect to test each cause.
The golden rule here remains constant: every root-cause suggested by AI is a hypothesis, not evidence. The cause of surface roughness may or may not be "cutting tool wear"; Only tool life data and experiment confirm this.
Caution: AI sometimes constructs chains of reasons that seem very plausible but are wrong on the ground. For example, it is easy to say "operator error" and is often unfair. Eliminate systemic causes (methods, equipment, measurements) with data before blaming the root cause on humans.
Mini Case: Shortening Mold Change Time (SMED)
Mold changes in a plastic injection workshop take an average of 55 minutes, and this creates great losses in small batches. Industrial engineer Deniz videotapes the changeover process and calculates step-by-step times (this data comes from the field). It then gives the AI the list of steps and their times:
Classify these mold change steps as "internal" (can be done while the machine is stopped) and "external" (can be done while the machine is running) according to the SMED method. Mark the inner steps that can be moved to the outer ones and write a preliminary preparation suggestion for each. Steps and durations: {{ list }}
The AI marks steps like “bringing the new mold to the machine” and “preparing the bolts” as portable to the external job, suggesting a makeshift cart. Deniz is testing this proposal in the field; it standardizes what is truly applicable. As a result, changeover time decreases from 55 minutes to 32 minutes. The critical point: AI sped up the plan, but it was field testing and standardization that really reduced the time.
Common Mistakes
- Bypassing Gemba: Implementing AI output without observing it in the field. It is based on the simple "go and see" principle.
- Relying on made-up numbers: AI sometimes produces non-existent numbers, like "cycle time is about 45 seconds." Every number must be supported by measurement.
- Blaming the root cause on humans: Accepting the first explanation that comes to mind is "carelessness"; Not investigating systemic causes.
- Locking in a single solution: Thinking that AI's first suggestion is the only option. Generate at least 2-3 alternatives and compare them according to the constraints.
- Forgetting the control phase: Making the improvement and leaving it without establishing standard business and monitoring KPIs; The process returns to its previous state.
In summary
- AI does not produce data in industrial engineering; It organizes the data you bring, interprets it and turns it into an improvement hypothesis.
- Make the output measurable when classifying the seven wastes: ask for evidence statement + metric to measure.
- Use AI as a draft generator at every stage of DMAIC, but always verify root-causes with data.
- In techniques like SMED, AI speeds up the plan; What really reduces time is field testing and standardization.
- Skipping Gemba, relying on made-up figures, and forgetting the checking phase are the most expensive mistakes.
Application task
Choose a process from your own workplace (or a realistic production/service process you know). First, observe the field for 5-10 minutes and note the actual steps and estimated times. Then have the AI analyze the process using the “7 waste classification” power prompt in this unit. For each detection of waste in the output: (a) compare to your observation whether the evidence is actually accurate, (b) identify at least one metric that should be measured, (c) select one quick improvement suggestion and evaluate it with an “inside/outside job” or “ECRS” logic. Finally, find an example where a determination produced by the AI turned out to be wrong (or could not be verified) in the field and write in one sentence why it was wrong.