OEE in practice: a real plant case

In one line

A Swedish battery factory measured the real OEE of its most critical process over 17 production days and the result was 14%, not because of major mechanical failures but because the daily operation (shifts, breaks, quality control, material loading) was not organized to keep the equipment running.

The case: a battery factory in start-up

The case is documented in a master's thesis from Uppsala University (Sweden, 2023), carried out by two researchers who worked full time at the company studied. The company (not identified by name in the study) is a recently founded battery manufacturer, operating since 2020 and still in the product design phase, meaning the specifications of the product it makes kept changing while the study was under way.

The equipment analyzed is the stacking stage: the step in the process where the electrode and separator layers that form the battery cell are stacked. It is one of the most delicate stages in all of battery manufacturing (an alignment error can cause contact between the positive and negative electrodes, with a risk of cell failure or fire), and because of its complexity it is usually the bottleneck in new battery plants.

The Production Department had been noticing problems at that stage for some time, but it had no hard data to confirm whether that equipment was really the bottleneck or not. Only when they cross-checked information from Process, Manufacturing and Quality did the focus converge on that machine. From there, the OEE study was set up.

How they measured: no digital system, everything by hand

The plant had no digital system for capturing stoppages. Everything was recorded by hand, through direct observation on the shop floor and documentation in Excel, during shifts from 08:00 to 17:00, from July 14 to August 1, 2023 (17 days).

Before they had a loss taxonomy that worked, the researchers went through three versions of stoppage categories. The first, taken directly from the literature, proved insufficient as soon as they started recording: stoppages appeared that did not fit any category, or that were ambiguous between two categories at once. They had to run a trial week just to adjust the list of stoppage codes before starting the real measurement. Only the third version, with categories such as "material loading", "process parameter adjustment", "equipment failure", "installation failure", "minor stoppage", "product quality control", "material starvation", "team meeting", "breaks" and "preventive maintenance", was the one used for the final calculation.

This is a practical finding in itself: in a plant that is just starting to measure OEE, the loss taxonomy almost never comes out right the first time. It is worth budgeting a short pilot phase before treating the first numbers as valid.

The result: 14% average OEE

The average OEE over the 17 measured days was 14%. To put that in perspective: the world-class benchmark usually cited for discrete manufacturing is 85%, and the company itself had set a target of reaching 65% by the end of that year.

But 14% should not be compared only against the theoretical ideal. Another OEE thesis done in Sweden, this time at Mälardalen University, using real data from multiple Swedish industrial plants captured with a production reporting system, found an average OEE of 43.5% across those plants (without counting total scheduled time), with values by sector between 43% and 50%.

In other words: 14% is not just low against the theoretical 85% from the textbook. It is low even against the real average of established Swedish plants, which sits around 43-50%. That confirms the problem was not "low OEE because every new plant measures low": there were specific, correctable losses pushing the number down.

An additional finding from the researchers themselves is revealing: analyzing the data, they calculated that the maximum achievable OEE under the current conditions of that line (without changing anything structural, only by optimizing the operation) was just 29%. That is, even eliminating all the avoidable losses identified would not reach the 65% the company had set. The 65% target would also require fundamental changes to the process, not just operating discipline.

The day-to-day variation was also huge: the worst day had a loss of 83% (that is, only 17% efficiency), and the best day a loss of 31% (69% efficiency). That spread, more than the average, is what usually reveals that the problem lies in shift management and not in the equipment. A machine with a chronic physical failure loses time more evenly from day to day.

Which loss dominated: not the equipment, but how the work around it was organized

The central finding of the case is this: availability, not performance or quality, was the factor that hurt OEE the most, and within availability, the four biggest causes were not equipment failures. A Pareto analysis of the lost hours showed that four causes accounted for about 70% of all the availability loss:

1. Operator breaks: 25% of the loss. During the measured period, the company had given part of its staff vacation, so instead of two 8-hour shifts (16 hours/day) there was a single 8-hour shift, covered by one operator trained for that machine. With nobody to rotate with, every time that operator needed a break, the machine simply stopped. On average, 60 minutes per day were lost for this reason alone.
2. Quality control of incoming material: 18%. The quality check of the semi-finished product entering the stacking stage was budgeted at 20-30 minutes, but in practice it took about 60 minutes a day (double), because the quality staff also covered other areas of the plant and the operator had to wait until they were available.
3. Material starvation: 15%. The equipment ran out of input material because the previous process on the line did not deliver on time. While it waited, availability was lost completely: the machine could do nothing.
4. Material loading: 13%. A manual task that is very sensitive to operator skill: loading time ranged from 3 minutes (best case) to 2 hours 10 minutes (worst case), depending on the experience of the person loading and the condition of the material received. It happened 42 times in the 17 days, one out of every five stoppage events recorded over the whole period.

None of the four is a mechanical failure of the stacking equipment. They are, in essence, problems of staffing, coordination between shifts and areas, and supply planning. This is the kind of loss that does not show up if you only look at the equipment and not at how the work around it is organized.

What they recommended for each loss

The study did not stop at diagnosis. For each of the four dominant causes it proposed a concrete correction:

• For breaks: a rotation system with at least two trained operators present at the same time, so one can cover the machine while the other takes a break. The authors themselves point out a real tension: with a single operator, performance drops below 50% of what it is with two, so gaining availability there will probably cost some performance. It is a trade-off to manage, not a free solution.
• For quality control: designate a single point of contact who notifies Quality when the material is ready, standardize the checklist (so it does not vary from person to person) and, above all, move the check up to the previous day whenever possible, so it stops being a step that blocks the line and is done in parallel.
• For starvation: build a buffer of material prepared one day in advance, instead of depending on the previous process delivering just in time.
• For material loading: a master loading plan with a technician assigned per shift specifically for that task, so the variability can be attributed to operator competence or to material complexity, rather than being mixed together as it is today.

As an additional observation, the study noted that there were three product specification changes in the 17 measured days, and that each change destabilized the machine's behavior. Since the equipment has four workstations, they proposed assigning fixed stations per customer or product, so the operator builds up repeated experience with each configuration instead of relearning the equipment's behavior every time the product changes.

How to apply this to your plant

This case does not say "measure OEE and you will find the problem is breaks". The problem will be different in every plant. What is transferable is the method:

• Before investing in fixing major equipment failures, look at the Pareto breakdown of your own availability losses. The biggest loss may not be in the machine but in how the shift around it is organized.
• If your measured OEE is far below even real industry averages (not just the 85% ideal), look for specific, correctable causes before assuming that "this is just how a new plant is".
• Also calculate the maximum achievable OEE with the current operation, without structural changes, as this study did with its 29%, to know whether your target is reachable with operating discipline or needs a fundamental investment in the process.
• If it is the first time your plant measures OEE, give the loss taxonomy a trial week before treating the first numbers as final. Here it took three versions to arrive at categories that really separated the causes well.

Benefits of measuring this way (with your own data, not from a textbook)

• It lets you tell shift-organization losses (staffing, coordination between areas) apart from technical equipment losses. These are problems solved with completely different tools.
• Calculating the "maximum achievable OEE" gives a realistic short-term target, separate from the long-term aspirational one. It keeps a team from becoming demoralized chasing a number that is not yet reachable with the current operation.
• The day-to-day spread (17% on the worst day, 69% on the best) is information in itself: a range that wide points to operational management causes, not to a physical failure of the equipment.

Limitations to keep in mind

• It is a study of a single piece of equipment, in a single plant, over 17 days. It is not a representative sample of the whole battery industry, and the authors themselves treat it as an exploratory case.
• The measurement period coincided with an anomalous situation (vacations, a single shift, a single trained operator), which probably exaggerated the relative weight of the break loss compared with a normal month of two-shift operation.
• The loss taxonomy was built specifically for this plant. It is not directly comparable, cause by cause, with that of another factory that uses different categories.

In summary

In this battery plant, the measured OEE (14%) came out far below not only the world-class ideal (85%) but also the real average of other Swedish plants (43-50%). The dominant cause was not an equipment failure but the lack of a second operator to cover breaks, together with coordination delays with Quality and with the previous process on the line. This is the kind of loss that only appears when you measure with discipline on the shop floor, and that you do not see by looking only at the equipment's technical datasheet.