Production optimization - methods, examples and automation of production processes

1. What is production optimization?

⚡ In one sentence

Production optimization means removing downtime, long changeovers, defective parts (scrap), excess inventory and unnecessary movement from a plant, so that the same machines and people produce more good products, faster and at lower cost.

💡 In plain terms

Optimizing a production process improves three areas at once:

• Output - how many good parts the line produces in the available time.
• Quality - how many products have to be reworked or thrown away.
• Flow - how long material sits between operations and how much money is tied up in inventory.

Each of these areas can be improved without new hires or new machines. A shorter changeover gets more parts out of the same line, removing the cause of defects reduces rework, and smaller batches cut the time semi-finished goods sit on the shop floor.

🔧 Deep dive

The Toyota Production System, from which Lean grew, divides losses into seven categories (muda), to which an eighth was added later:

• overproduction,
• waiting,
• transport,
• over-processing (more processing than the customer requires),
• inventory,
• unnecessary motion,
• defects and rework,
• unused human potential.

We describe Lean and Six Sigma in more detail, including outside the shop floor, in the article What is business process optimization?

2. Where to start? Measurement and OEE

⚡ In one sentence

OEE (Overall Equipment Effectiveness) shows what part of a machine's planned working time turns into good products.

💡 In plain terms

OEE breaks lost time down into three causes:

• Availability - did the machine run during the planned time? Breakdowns, changeovers and missing material reduce it.
• Performance - did it run at full speed? Micro-stoppages (short stops nobody records) and running at reduced speed reduce it.
• Quality - how many of the parts produced were good the first time?

Example: a machine is planned to run 900 minutes a day and stands still for 135 minutes (changeovers, breakdowns). Availability is 85%. In the remaining time it reaches 90% of full speed, and 97% of parts pass inspection. OEE is the product of these three figures: 0.85 × 0.90 × 0.97 ≈ 74%. This means that more than a quarter of the planned time does not produce good products.

The result has to be broken down into its three components. When most is lost on availability, shorter changeovers and better maintenance (keeping machines in good working order) help. With poor quality, you need to look for the causes of defects. Without this breakdown, optimizing production processes turns into guesswork.

OEE, however, describes machines only. When the bottleneck is a supplier, the organization of work or the shop-floor layout, other things are measured: time spent waiting for material, the length of queues in front of workstations and the distance forklifts and people travel.

🔧 Deep dive

OEE was introduced by Seiichi Nakajima as part of TPM (Total Productive Maintenance), in which operators and maintenance staff work together to prevent breakdowns. The 85% level referred to as "world class" comes from his work. Following the trend on your own line matters more than comparisons with other factories.

OEE is easy to "improve" on paper, e.g. by excluding changeovers from planned time. That is why the definition has to be fixed once, in writing, and data collected automatically from the machines or from an MES (Manufacturing Execution System) instead of from handwritten notes filled in at the end of a shift.

3. Production optimization methods

⚡ In one sentence

The method is chosen to match the loss that dominates on a given line: long changeovers, breakdowns, defects or excess inventory call for different tools.

💡 In plain terms

The starting point is a value stream map. It shows the biggest loss, and the tool is chosen to fit it. We describe how to draw such a map in the article Process mapping.

🔧 Deep dive

The SMED method was described by Shigeo Shingo. "Single-Minute" in the name means a changeover in a single-digit number of minutes, i.e. under 10 minutes. The method divides changeover activities into internal ones (which require stopping the machine) and external ones (which can be done while it is still running), and then moves as many as possible into the second group.

Eliyahu Goldratt's Theory of Constraints (the book "The Goal", 1984) holds that every hour gained at the bottleneck increases the output of the whole plant, while time saved at an operation before it only adds to the queue. Improvements therefore start at the stage that sets the pace.

4. Optimizing a production process step by step

⚡ In one sentence

Choose one line or a group of similar products, map the value stream, measure OEE, find the bottleneck, remove the biggest loss, standardize the new way of working and only then automate.

💡 In plain terms

1. Narrow the scope. One line or a group of similar products, ideally the one with the highest output or the longest list of delays.
2. Map the value stream. Backwards from shipping to raw material: operations, inventory, cycle times (how long it takes to process one part), information flow.
3. Measure. OEE broken down into availability, performance and quality; a list of stoppages with their causes; the scrap rate.
4. Find the bottleneck. The operation whose cycle time exceeds takt time (the pace at which the customer takes products), or in front of which inventory keeps growing. It may be a machine, a manual workstation, quality control or the material supply.
5. Remove the biggest loss. SMED for changeovers, TPM for breakdowns, root cause analysis for defects.
6. Standardize. Work instructions, changeover checklists, 5S. Without a standard, the improvement disappears after a few weeks.
7. Automate and keep measuring. Choose only stable processes for this; the indicators go onto a permanent chart dashboard.

On a single line the whole cycle takes a few weeks. You will see the effect of a SMED workshop at the very next product change.

🔧 Deep dive: a worked example

A line runs two shifts, with a planned time of 900 minutes a day. The bottleneck is a press with three changeovers a day of 90 minutes each. Changeovers alone therefore take 270 minutes, and the availability of the press (ignoring other stoppages) is 70%.

After a SMED workshop, some activities are moved into the machine's running time: tools, material and documents are prepared before the press stops. One changeover now takes 30 minutes, three a day take 90 minutes in total, and availability rises to 90%.

At the same speed and quality, the press therefore delivers about 29% more output (90 / 70 ≈ 1.29), without a new machine and without an extra shift. Because it is the bottleneck, the gain translates into the output of the whole line. The figures are illustrative.

5. Automating production processes - what is worth automating?

⚡ In one sentence

Automating production processes covers robots and machines as well as the flow of information around them: collecting data from the line, reporting, quality control, planning and maintenance.

💡 In plain terms

Buying a robot or a new line is a large investment with a long lead time. The flow of information on the shop floor, on the other hand, can be automated in small steps, without such large outlays:

• Collecting data from the line - part counts, stoppages and their causes read from the machines or an MES instead of from sheets filled in by hand during the shift.
• Production reporting - plan fulfilment, OEE and scrap go into a summary that updates automatically.
• Quality control - a camera and an image recognition model (computer vision) check every part instead of a sample and record the type and location of each defect.
• Planning - a schedule based on the demand forecast and machine availability rather than built by hand in a spreadsheet (an example from our projects).
• Maintenance - servicing based on the condition of the equipment and its breakdown history, not only on the calendar.

The principle is the same as in the office: first tidy up, then automate. An unstable process with robots added only produces scrap faster. We write more about choosing tasks for automation in the article What is business process automation?

🔧 Deep dive

In industry, the layers of software are described by the so-called automation pyramid (in line with the ISA-95 standard). At the bottom are sensors and PLCs (Programmable Logic Controllers), small computers that run the machines. Above them are SCADA systems (Supervisory Control and Data Acquisition), whose screens show the operation of the line, then the MES, and at the top the ERP (Enterprise Resource Planning), which connects sales, warehouse and finance. The problem arises when the layers are not connected: the machines have the data, the ERP has the orders, and in between there is a spreadsheet and a telephone.

In such a plant, the first automation project may be systems integration, which makes data from the shop floor flow by itself to where decisions are made.

6. AI in production - where does it deliver a measurable effect?

⚡ In one sentence

Artificial intelligence works best in a plant where there is data and a recurring decision: in visual quality inspection, demand forecasting, failure prediction and scheduling.

💡 In plain terms

• Visual quality inspection - a model recognizes defects in camera images. It needs images of good and defective products and stable lighting.
• Demand forecasting - a model learns from sales history, seasonality and external factors. Matching production more closely to sales means lower inventory and fewer urgent changes to the plan.
• Failure prediction - analysis of vibration, temperature or power consumption makes it possible to schedule servicing before the machine stops. Prerequisite: sensors and a record of past faults.
• Scheduling - optimization algorithms arrange the order of jobs so as to reduce changeovers and meet deadlines.

Each of these applications starts with the question of data. If the plant does not record the causes of stoppages or the types of defects, the first step is to collect this information systematically, and the AI model comes later. More on the conditions for implementation in the article What is artificial intelligence?

🔧 Deep dive

With AI on the shop floor, you have to plan what happens to the model's output. A camera that detects a defect but neither stops the line nor removes the part only produces statistics. A working implementation therefore consists of three parts: the model, the connection to the process (a signal to the machine, a system or an operator) and a confidence threshold below which a person makes the decision.

Setting the threshold is a trade-off. A high threshold sends more parts for manual assessment and adds work for the inspectors. A low threshold leaves most decisions to the model, and with them the number of mistakes grows: defects let through or good products rejected. The right value is found by comparing the cost of such errors with the cost of manual inspection.

The model also has to be monitored after go-live. A new raw material supplier, different lighting or a product change reduce recognition accuracy, so the results are compared regularly with the inspectors' assessment and, if necessary, the model is retrained on new images.

7. Production optimization mistakes

⚡ In one sentence

The most expensive mistakes are hiring people or buying machines and systems before measuring where the time goes, and improving an operation that is not the bottleneck.

💡 In plain terms

• Adding people. Extra staff outside the bottleneck raise costs while output stays the same.
• Buying instead of diagnosing. If an old machine has 60% OEE because of long changeovers, a new one with the same work organization will achieve a similar result.
• Improving in the wrong place. A faster operation in front of the bottleneck only increases inventory.
• Data from memory. Stoppage causes entered "from memory" at the end of a shift are useless for analysis.
• No standard. A shortened changeover goes back to the old time when the crew changes.
• A project without people from the shop floor. Operators and supervisors know why the line stops. Without them, the solution is designed at a desk and then bypassed.
• Automating an unstable process. A robot or a camera will not fix work that runs differently every time.

🔧 Deep dive

When assessing results, look at three figures together: the OEE of the bottleneck, the lead time from raw material to shipment and the inventory level. OEE is easy to raise by producing large batches: fewer changeovers mean the machine runs longer. But the products are made to stock before any customer orders them, and the other variants have to wait their turn. The indicator rises, while the plant ties up more money in the warehouse and lengthens delivery times. That is why OEE cannot be the only goal. Batch sizes are set to match actual orders, and changeovers are shortened with SMED so that frequent product changes do not drag the indicator down. If OEE rises together with inventory and lead time, the improvement is only apparent.

At cm-opti we combine shop-floor and logistics experience with data analysis and AI. We start with the challenge the company comes to us with: in conversations with the team and using figures from its systems, we check whether the problem lies where management thinks it does and how much it costs today. Only then do we propose a solution, for example software integration, automatic capture of machine data, reporting or an AI model in a specific process step, together with a calculated return on investment.

- The cm-opti perspective

Frequently asked questions (FAQ)

What is production optimization?

It means removing downtime, long changeovers, defective parts, excess inventory and unnecessary movement from a plant, so that the same machines and people produce more good products, faster and at lower cost.

What are the methods of production optimization?

The basic methods are value stream mapping (VSM), SMED for shortening changeovers, TPM for preventing breakdowns, 5S for order and standards at the workstation, Kanban for producing to actual demand, Six Sigma for reducing defects and the Theory of Constraints for working on the bottleneck.

How do you calculate OEE?

OEE is the product of three figures: availability (how much of the planned time the machine ran), performance (what percentage of full speed it reached) and quality (the share of parts that were good the first time). Example: 0.85 × 0.90 × 0.97 ≈ 74%.

Where should you start optimizing production processes?

With one line or group of similar products: a value stream map, measuring OEE broken down by loss and finding the bottleneck. Improvements start with the operation that sets the pace of the whole line.

What can be automated in production?

Apart from robotizing the operations themselves, it is worth automating the flow of information: collecting data from the line, production reporting, quality control with cameras and image recognition, planning based on demand forecasts and condition-based maintenance.

Does production optimization require investment in new machines?

Not at the beginning. First, measure how much time is lost to changeovers, micro-stoppages and rework. In the example in this article, a SMED workshop at the bottleneck delivers almost 30% more output without a new machine.

First step

Choose the line with the highest output or the longest list of delays. For two weeks, collect the causes of stoppages and the changeover times, calculate OEE (broken down into availability, performance and quality), and then draw a value stream map. The bottleneck and the biggest loss will then be visible in the figures.

Would you like to check whether the problem lies where you think it does, and how much it costs you? Let's talk - the first conversation is free of charge.

Related articles in the cm-opti Knowledge base

• Process mapping - how to create a process map step by step
• What is business process optimization?
• What is business process automation?
• What is computer vision?
• Data analysis and Business Intelligence
• What is systems integration?
• Case: demand forecasting and distribution optimization

Concepts explained in this article → Glossary

Production optimization, OEE, availability, performance, quality, muda (waste), VSM, SMED, TPM, 5S, Kanban, Six Sigma, SPC, Theory of Constraints (TOC), bottleneck, takt time, lead time, MES, SCADA, PLC, ISA-95, computer vision, demand forecasting, predictive maintenance

Sources and references

• Taiichi Ohno, "Toyota Production System: Beyond Large-Scale Production", Productivity Press, 1988
• Seiichi Nakajima, "Introduction to TPM: Total Productive Maintenance", Productivity Press, 1988 - Wikipedia: Overall equipment effectiveness
• Shigeo Shingo, "A Revolution in Manufacturing: The SMED System", Productivity Press, 1985 - Wikipedia: SMED
• Eliyahu M. Goldratt, "The Goal", 1984 - Wikipedia: Theory of constraints
• Mike Rother, John Shook, "Learning to See", Lean Enterprise Institute, 1998
• ANSI/ISA-95 (IEC 62264), enterprise-control system integration - Wikipedia: ANSI/ISA-95