Book summary “Lean Six Sigma”: summary

Introduction

You face a constant struggle when high-mix, low-volume manufacturing creates immense operational complexity and drains your profitability. Traditional tools fail because standard Lean systems target high-volume repetition while Six Sigma methods often ignore complex scheduling realities.

Michael L. George presents a breakthrough by combining traditional Lean Six Sigma principles with modern Artificial Intelligence and Neural Networks. This powerful synthesis solves low-volume job shop manufacturing complexity and unlocks massive EBITDA growth.

The book guides you through a clear operational framework across four core areas.

• How can you apply data mining and process metrics to rescue failing manufacturing operations?
• What methods control inventory and solve complex job shop scheduling problems?
• How does the Fourth Manufacturing Revolution eliminate setup waste without massive capital investment?
• Where should you apply AI readiness assessments and predictive maintenance across your organization?

1. The Turnaround Strategy and Data Mining

You can rescue a failing manufacturing operation by systematically applying ten process improvement metrics guided by artificial intelligence data mining. A structured turnaround strategy moves a struggling business from heavy losses to high profitability in months.

A failing aerospace manufacturer achieved a massive financial turnaround by systematically applying ten process improvement metrics guided by AI data mining.

An undisclosed aerospace manufacturer producing precision-machined components for Boeing 737 and Airbus A320 jet engines and landing gears improved its EBITDA from minus 3.6% in 2016 to plus 19.5% within 18 months.

EBITDA Turnaround in Aerospace

The company utilized Big Data spreadsheets covering all 1,000 part numbers from its enterprise resource planning system to find patterns, remove outliers via interquartile analysis, and correct erratic pricing structures that previously omitted consumable tooling costs. You must obtain management commitment and track ten specific performance metrics across the organization.

• 1. Management team: Obtain management team's engagement in process improvement. Team members must actively engage in project selection and progress reviews.
• 2. AI data mining: Remove outliers using interquartile analysis. Create a Big Data spreadsheet that provides all accounting cost detail on all 1,000 part numbers produced per year regardless of volume.
• 3. Cash flow improvement: Increase to 20% of revenue using Pull systems.
• 4. Labor efficiency: Measure actual labor cost vs. verified and corrected accounting standards; react daily to adverse cost variances. Investigate outlier data source and correct.
• 5. Scrap cost: Reduce from 10% to less than 3% of revenue.
• 6. Total Productive Maintenance: Increase machine up time from 88% to 99%.
• 7. Setup time reduction: Reduce the waste due to setup time by greater than 25%.
• 8. Employee morale: Reduce negative perception of company from 62% to less than 5%. Engage all employees in process improvement activities.
• 9. EBITDA: Increase from -3.6% to +20% of revenue.
• 10. Work-in-process inventory turns: Increase by 50%. Reduce cycle time by 50% and improve on-time delivery to more than 90%.

AI Data Mining and Outliers

Treating every defect as a treasure allows you to implement a structured hierarchy of countermeasures that eliminates recurring scrap and secures vital quality certifications. Rather than blaming operators, you must evaluate the method of manufacturing, apply mistake-proofing or poka-yoke, fix poor gauging, and retrain staff.

Treating every defect as a treasure and implementing a structured hierarchy of countermeasures eliminates recurring scrap and secures vital quality certifications.

The company's scrap rate dropped from 7-10% to 3% of revenue, and their AS9100 quality audit score rose from 6 out of 15 to 14 out of 15. You must execute these exact steps when defects occur:

• 1. Supervision and engineering evaluate the Method of Manufacturing with the operator of the part number to eliminate and improve error-prone processes to make repeatably perfect parts possible.
• 2. Preferable countermeasure method is mistake-proofing, or poka-yoke, so that it is impossible for the error to recur.
• 3. Check for poor gauging, bad incoming material, poor work instructions: fix as needed.
• 4. Determine when the defective part number will be run again and prepare prevention documents for the operator.
• 5. Retrain operator as needed when the next production run of the part number is about to commence.

Achieving high equipment effectiveness requires raising machine uptime through Total Productive Maintenance to prevent single-point failures and hidden downtime losses. Machine downtime directly drains labor efficiency and forces either excessive overtime or larger batch sizes. Implementing Total Productive Maintenance requires operator, engineering, and quality control engagement to systematically eliminate mechanical failure points from your operations.

On an Okuma lathe, a TPM event uncovered a massive hole in the upper bellows allowing chips into the ball screw, and a ruined bearing pack that required a two-week parts wait and a two-week rebuild. You must address these hidden losses to reach the target machine uptime percentage of 99% from a previous uptime of less than 88%.

2. Inventory Control and Job Shop Scheduling

Traditional push manufacturing floods the shop floor with excess material regardless of machine capacity. Instead of creating massive queues, you can implement a Pull system that limits each machine to a work-in-process cap of less than 10 days of work.

Implementing WIP caps and AI Neural Networks solves the complex job shop scheduling problem and reduces cycle times.

This approach eliminates month-end frantic rushes, prevents overtime and premium freight costs, and ensures production linearity. According to Little's Law, cutting work-in-process in half while keeping output constant achieves a 50 percent reduction in manufacturing cycle time.

The aerospace company capped work-in-process, which reduced cycle time, dropped customer status calls, and improved on-time delivery from 50 percent up to 96 percent within 12 months.

Pull Systems and Little's Law

Decomposing factory work-in-process into generic AI Pull Groups of like machines bypasses the computational limits of job shop scheduling. You can track the number of WIP units and exit rate per day at each Pull Group to calculate average cycle time delay using Little's Law.

You must follow specific operational mechanics to maintain flow:

• Cap the work-in-process in front of each functional department of like machines to a low level, such as 10 days of work or about 2 jobs per machine.
• Track the number of WIP units and exit rate per day at each Pull Group to calculate average cycle time delay using Little's Law.
• Route completed jobs to the next functional AI Pull Group rather than a specific machine.
• Determine the appropriate WIP unit of measure.
• Establish the optimum size for the WIP cap.
• Establish a queue management process to control the sequence of starts.
• Maintain the WIP cap by ensuring new starts are triggered by exits from the process.
• Reduce process lead time if required by the customer.
• Resize the WIP cap to meet the new process lead time.
• Use statistics to help decide on a unit of measure.

Neural Networks and Sequencing

Finding the optimal sequence of part numbers to minimize setup time is an NP-hard combinatorial optimization problem where computation time grows exponentially. Cloud platforms like Amazon Web Services provide massive parallel computing power to train Neural Networks using thousands of offline Branch and Bound solutions.

You can execute precise sequencing workflows at each station:

• Supply the Neural Network with a list of all active jobs and part numbers at each Pull Group due within 30 days.
• Maintain a matrix of setup times from any part number to any other part number.
• Wait until the 3-hour window before a machine finishes its task to pull current data and respond to emerging shortages.
• Examine all jobs at the Pull station and find a sequence of 4 jobs that minimizes total setup time.
• Calculate the remaining process lead time down the router of each selected job to verify on-time completion.
• Alert the tool crib at the 3-hour point to prepare and deliver a cart with all needed tools and instructions.
• Reduce sequence length or batch sizes if delivery validation fails.
• Identify about 50 candidate part numbers for the next available machine.
• Use Cloud computing to run offline Branch and Bound or TSP Heuristics training examples.
• Train the Neural Network using back-propagation and gradient descent to learn the mapping of optimal part sequences.
• Validate customer promise dates for the selected sequence using simulation before releasing production.

3. Historical Revolutions and Setup Reduction

Manufacturing progress has advanced through four distinct historical revolutions. Each new phase builds upon the previous one to radically increase productivity and responsiveness.

The Fourth Manufacturing Revolution substitutes data for capital investment to eliminate setup waste in low-volume production.

Four Manufacturing Revolutions

The First Revolution replaced muscle power with steam power. Andrew Carnegie's father used a hand loom in Scotland before steam-driven Jacquard looms revolutionized fabric production.

The Second Revolution utilized electricity and assembly lines for high-volume identical production. Henry Ford's Model T mass production exemplified this second revolution.

In 1908, the price of a Ford Model T was 850 dollars. By 1920, Ford's production reached 2 million cars per year. Production efficiency drove the price down to 245 dollars by 1922.

The Second Revolution required an annual volume of 250,000 units to generate good investment returns. The Third Revolution introduced Lean Six Sigma and the Toyota Production System to reduce inventory and cycle time.

The Fourth Revolution uses Artificial Intelligence and data to optimize low-volume and complex production without massive capital investment. The 18th century marked the period when the First Manufacturing Revolution began.

Generic Setup Reduction

Global Big Data analysis reveals that the vast majority of setup time waste is concentrated in low-volume, non-repetitive part numbers previously ignored by traditional Pareto analysis. Traditional Lean Six Sigma focused on the repetitive 20 percent of part numbers that drove 80 percent of revenue.

However, global AI data mining shows that low-volume parts consume a disproportionate amount of setup time per dollar of revenue. This results in massive inventory write-offs when companies build large batch sizes to amortize setup.

Analysis of Okuma LB lathes showed that 73 percent of total setup time waste was consumed by low-volume part numbers that generated only 20 percent of revenue. Machine tools in the study had a typical setup time of 4 hours.

Yearly revenue was previously lost to multimillion-dollar write-offs of unsold finished goods inventory, accounting for 20 percent of losses. Generic Setup Reduction eliminates universal fumble factors and shared setup tasks across low-volume parts without requiring expensive part-specific engineering.

Developed by John Smith, Generic Setup captures about 75 percent of the setup time reduction of the Toyota method at a fraction of the cost. The method focuses on organizational and externalizable tasks that apply to all part numbers.

The method allows the Aerospace company to economically reduce setup time on 800 low-volume parts without spending capital on expensive tooling for every single part. You achieve 75 percent as effective results as Toyota's method without part-specific engineering investment.

4. Readiness Assessments and Non-Manufacturing Applications

AI readiness assessments and predictive maintenance extend the benefits of Lean Six Sigma to project management and complex process industries.

AI Readiness Assessments

An AI readiness assessment identifies existing process waste and determines the training, software, and management effort required to eliminate it. The assessment aims to quantify process waste, specify the resources needed to eliminate over 50 percent of that waste, and target real-time factory floor decisions that yield greater than 10 percent higher capacity with no extra labor, overhead, or capital equipment, while raising on-time delivery above 95 percent. Initial EBITDA of 20 percent of revenue can approach 30 percent, which increases shareholder equity by 50 percent.

At the Aerospace company, leadership conducted a readiness assessment to target specific actionable projects and correct data collection failures.

To prepare your organization, you follow specific practices:

• Determine approximately how many dollars of waste exists in your process
• Determine what additional training, implementation effort, and AI software are needed to eliminate greater than 50% of the waste
• Target greater than 10% higher capacity with no additional cost in labor, manufacturing overhead, or capital equipment
• Increase on-time delivery to greater than 95%

A readiness assessment evaluates current business performance across four main areas: financial, effectiveness, efficiency, and management and culture. By reviewing metrics in financial health, customer delivery, resource optimization, and team alignment, organizations uncover performance gaps and misalignments that must be resolved prior to deploying AI and Lean Six Sigma. You monitor financial performance to track the health of the business. You measure effectiveness performance by delivering to customers what they need when they need it at the quality they expect. You evaluate efficiency performance by optimizing all of the company resources to drive profit. Finally, you assess management and culture performance by enabling high performance through team alignment and cohesiveness, employee morale, talent, and opportunity using a value driver tree and economic profit waterfall.

Predictive Maintenance

Predictive maintenance uses vibration frequency data to train Neural Networks to estimate residual bearing life and prevent unexpected machine downtime. High pressure creates subsurface defects that lead to spalling and acoustical frequencies as bearings rotate. By feeding vibration amplitude data into a Neural Network, manufacturers can accurately predict remaining bearing life before failure occurs. More than 80 percent of predictions are within 20 percent of the actual bearing life.

The jet engine oil separator was machined at 2,000 RPM on a Multus versus 400 RPM on an Okuma LT. The Multus machine milling motor peak power increased from 5 HP to 25 HP.

You implement Total Productive Maintenance by following specific steps:

• Verify that the Lean Six Sigma TPM process is in place and generating valid data including vibration monitoring
• Supply data such as the amplitude of acoustical noise, fundamental, and harmonics to form the training set of the Neural Network
• Correct input weights to the Neural Network to reduce the error between predicted and actual operating time

Scheduling project teams near 100 percent utilization dramatically inflates cycle times due to task variation and queuing delays.

Variation between scheduled and actual task times is roughly plus or minus 70 percent. When utilization approaches 95 percent, queue times surge, turning a three-day cycle time into a two-week cycle time. A manager at a jet engine manufacturer noted during a presentation that their engineering teams routinely scheduled near 100 percent utilization and were always late.

You avoid these delays by applying the Pollaczek-Khintchine equation and managing slack time:

• Schedule project team task estimates at approximately 80% to 85% of a 40-hour work week to maintain a flat cycle time on the Pollaczek-Khintchine curve.
• Provide 15% slack time to absorb unforeseeable problems, cover rework and absenteeism allowances, and improve on-time delivery performance.

Neural Networks should be deployed when an exact mathematical equation cannot be solved or an algorithm cannot be derived to predict outputs from inputs. Neural Networks use AI to find similar past patterns of work across teams and subprojects, eliminating duplicated efforts and reducing cost and cycle time. You use a Neural Network if you cannot solve the equation that provides the desired output. You also use a Neural Network if you cannot derive an algorithm to predict an output based on inputs.

Enabling cross-project collaboration and consolidating common software tasks can dramatically reduce lead time, cost, and life cycle expenses. Engineering teams often resist sharing code across projects because they view it as an unrewarded distraction from their individual deadlines. Overcoming this requires management support, matrix management models, and entrepreneurial initiative. At Xerox PARC, 10 separate product teams were independently writing controller/scheduler software for copiers. Mark Webster and about 20 engineers used their vacation time to design a common platform. Their meetings eventually gathered 100 engineers, proving that 90 percent to 95 percent of the code could be reused across all machines.

You organize software development by adopting specific practices:

• Form a single expert software or task team to write common code or execute shared deliverables across multiple project teams.
• Adopt a matrix management model where individual project teams act as customers of the centralized expert team.

Data mining ERP accounting records allows organizations to compare scheduled versus actual task times and apply learning curves to project management. Historical project labor costs and task variances reveal learning rates. While manufacturing labor costs historically decreased by 20 percent for each doubling of production quantity, project management tasks can achieve much steeper learning curves when commonality is high, appearing as a quantized step-function due to smaller quantities. In the 1930s aviation industry, aircraft production rose from 3,000 units in 1939 to 84,000 units in 1943 as labor decreased by 20 percent for each doubling of quantity across five production doublings. The total remaining labor cost factor after five doublings dropped to 0.32, which is a 67 percent reduction. An example with 90 percent commonality results in subsequent unit costs dropping to 10 percent of the original cost.

You apply learning curves by using concrete steps:

• Mine past project records in ERP systems to identify scheduled time versus actual recorded time variances.
• Group tasks with common subtasks within specific time buckets to be produced by an expert team.
• Apply learning curves based on the percentage of commonality among tasks to project scheduling and costing.

10 Key Ideas

1. An aerospace manufacturer improved its EBITDA from minus 3.6% in 2016 to plus 19.5% within 18 months by systematically applying ten process improvement metrics guided by artificial intelligence data mining.
2. Treating every defect as a treasure and executing a structured hierarchy of countermeasures eliminates recurring scrap and secures vital quality certifications.
3. Implementing Work-In-Process caps limits each machine to less than 10 days of work to eliminate bottlenecks, reduce cycle times, and improve on-time delivery.
4. Cloud platforms provide massive parallel computing power to train Neural Networks using thousands of offline Branch and Bound solutions to solve complex job shop scheduling problems.
5. The Fourth Manufacturing Revolution substitutes data for capital investment to eliminate setup waste in low-volume production.
6. Generic Setup Reduction captures about 75 percent of the setup time reduction of the Toyota method at a fraction of the cost without requiring part-specific engineering investments.
7. An AI readiness assessment quantifies existing process waste, specifies needed resources, and targets greater than 10 percent higher capacity with no extra labor, overhead, or capital equipment.
8. Predictive maintenance uses vibration frequency data to train Neural Networks to estimate residual bearing life and prevent unexpected machine downtime.
9. Scheduling project teams at approximately 80% to 85% of capacity provides necessary slack time to absorb task variation and maintain predictable cycle times based on the Pollaczek-Khintchine equation.
10. Data mining enterprise resource planning accounting records allows organizations to compare scheduled versus actual task times and apply learning curves to project management and costing.