1. What is OEE — Overall Equipment Effectiveness?
OEE (Overall Equipment Effectiveness) is true plant availability. It covers all losses in three dimensions:
Availability level: Is the system standing even though it should be running? Causes can be downtimes, faults, set-up or cleaning times. The level of availability expresses how much of the planned run time the plant actually produces.
Performance level: How much does the plant produce in relation to the maximum possible speed? Employees often turn the system down to 95 or 90 cycles out of good faith — this results in 5—10% performance losses that are at least as significant as downtimes.
Quality level: NIO parts (out-of-spec) tie up machine capacity that is irretrievably lost. From a plant perspective, every scrap part is also an OEE loss.
Practical example: The coffee machine. Using a coffee machine, the principle can be clearly explained: Change the pomace = downtime (availability). Conversation between two cups = 10 seconds waiting time (performance). Cup not correctly placed = coffee puddle (quality).
OEE is particularly valuable because it is free of manipulation: If availability losses are not logged correctly, they automatically appear in other areas. It creates a common, unquestionable truth for all parties involved.
2. OEE recording — manual vs. automated
Manual entry: Suitable as a start, as all employees are actively working with the system. The disadvantages are inaccuracies in the assignment, a lack of confidence in the recording (e.g. due to manual Excel entries) and lack of evaluation in real time.
Automated capture: For use on multiple systems, automated signal capture directly from the machine is recommended. An operator interface allows real-time display of quantities, downtimes, Microstop frequencies and OEE processes in minute resolution — divided by shift.
Decisive: The data must be reflected back to employees. Only when operators see the same data as management do they create a common truth — without employees being able to manipulate data.
3. Classic OEE management & continuous improvement
Based on the OEE data collected, causes of loss are prioritized in Pareto analyses. Typical follow-up actions in the lean program include: weekly and monthly OEE talks, set-up workshops (SMED), coordination of production sequences with scheduling, and workshops to reduce cleaning times.
4. AI to increase OEE — basic principle & approach
Artificial intelligence is used in two variants:
- Machine Learning (ML): Supervised and unsupervised learning for pattern recognition in time series data.
- Large Language Models (LLM): Language-based AI for analysis and knowledge transfer (comparable to ChatGPT).
Important: The goal is not the dark factory. AI takes on repetitive, analytical tasks and prepares decisions — the actual decision remains with the experienced employee. The focus is on integrating domain knowledge from the workforce.
5th ML application: Anomaly detection in production time series
Machine learning algorithms continuously monitor the flow of quantities and recognize unusual patterns before a fault occurs. Typical anomaly types: point anomalies (individual outliers), trend anomalies (long-term deviations), seasonal anomalies (deviation from known periodic patterns) and random fluctuations (increased noise, which indicates unstable plant performance).
Early warning shortens Mean Time to Repair (MTTR). An event is traditionally followed by a latent phase (undetected), followed by disruptive effects, analysis, countermeasures and restart. With AI-based anomaly detection, the latent phase can be significantly shortened — ideally, notifications are made even before the disruptive effect occurs. Notifications are sent directly to the responsible maintenance technician via SMS, in-app message or e-mail and as a note on the shop floor board. Principle: The value of information about a problem deteriorates — the faster you react, the more precise the diagnosis.
6. Generative AI: Suggested causes & measures
Issue: loss of knowledge due to demographic change. Long-standing maintenance specialists with extensive plant knowledge are increasingly leaving (age group 55—60 years). Your experience about causes of faults and solutions is in danger of being lost.
solution: AI-powered knowledge assurance. When employees document the causes and solutions to a malfunction in free text (or via voice input), an LLM aggregates these entries: Similar causes are summarized, suggested solutions are assigned and presented in a structured manner. Free text problems such as spelling mistakes, different languages or varying formulations are compensated by the language comprehension of the model.
Practical example: Entries such as “air filter replaced” and “air filter cleaned with compressed air” are summarized as an aggregated recommendation: “Overheating due to poor ventilation — check ventilation, clean or replace filters.”
At the same time, automatically recorded machine data (e.g. frequency converter signals) can flow directly into the fault location, meaning that employees have to enter less manually.
7. Generative AI: Conversational OEE Analysis
The second LLM application scenario is a conversational AI assistant that has access to real-time plant data. Employees can ask questions in a natural dialog format — without predefined analysis paths, e.g.: “What is the current OEE on line X?” /“How many shutdowns were there yesterday and how long did they last?” /“Which product had the best OEE yesterday?” /“What can I do to reduce set-up time?”
Full PDCA in dialog mode. The assistant supports the entire PDCA cycle: retrieve data, understand it, derive measures, estimate potential. Example: The question of the effect of a 30% set-up time reduction results in an automatic recalculation of OEE — in the example shown, a potential of +8.5% OEE.
Technical implementation & data protection. In the presented system, the LLM (Llama 3.1 from Meta) runs on its own servers in the Frankfurt data center. Production data therefore does not leave the company — no transfer to external cloud services.
conclusion
AI — whether machine learning or large language models — is not a substitute for employee knowledge, but rather an amplifier. The presented approaches help to identify disorders earlier, to preserve knowledge and to accelerate analyses. The result: shorter downtimes, faster response and a sustained increase in OEE.