How to Calculate Reliable OEE on a Heterogeneous Fleet of CNC Machines?

In modern manufacturing, calculating Overall Equipment Effectiveness (OEE) has become the standard for measuring productivity. However, while the theoretical formula is simple, its application on the shop floor is often a real headache.

The major challenge does not lie in the calculation itself, but in the reliability and comparability of the data. How do you compare the performance of a brand-new machine connected via OPC UA with that of an older asset whose status reporting is limited? How do you handle a "feed hold" that skews your cycle times?

This article proposes a pragmatic approach to normalizing your data and obtaining actionable OEE on a heterogeneous fleet, far removed from theoretical approximations.

1. OEE: What the Indicator Actually Measures (and What It Doesn’t)

Before tackling technical complexity, it is essential to recall that OEE is the product of three ratios: Availability, Performance, and Quality.

However, OEE is not an absolute truth; it is the result of a model. If the input data is inconsistent, the result will be mathematically correct but industrially wrong. The main bias occurs when the calculation scope (what is considered "open time" or "production time") varies from one machine to another within the same workshop.

2. Why OEE is difficult to compare across a heterogeneous fleet

The reality of a machining workshop is made up of diversity: different CNC controls (Fanuc, Siemens, Heidenhain, etc.), varied generations, and disparate communication protocols.

The classic mistake is to think that the "IN CYCLE" state means the same thing on all machines. This is not the case. Three major sources create discrepancies:

1. CNC vocabulary:Each manufacturer has its own definition of machine states.

2. Granularity:A modern machine can distinguish a pause during a program, a manual execution, whereas an older machine will only report a generic waiting state.

3. Information availability:Some critical data (such as the position of the feed rate override) is not always natively accessible.

   
   

3. Common biases in the workshop: manual mode, feed hold, and adjustments

This is where the difference lies between theoretical OEE and reliable shop floor OEE. To obtain accurate indicators, you need to address these specific cases:

Manual mode vs Automatic mode

A machine in manual mode can technically machine, but it does not produce at the expected rate. If these times are counted as "production" without distinction, they artificially inflate availability while degrading the performance rate. It is crucial to qualify the mode of operation to isolate actual production.

Feed hold and feed corrections

This is a textbook case: the machine is in cycle, the status reports "production", but the operator has activated the feed hold or reduced the override to 0% to check a dimension. 

Result:The machine is not producing, but the system calculates OEE. To correct this, it is necessary to distinguish between "declared production" (machine status) and "actual production" (machine status + effective activity).

Setup and Maintenance

Should change over times (SMED) be included in the OEE calculation? It depends on your strategy, but the most important thing is consistency. If machine A includes setup in the required time and machine B excludes it, no comparison is possible. A common model must be defined that clearly categorizes production, setup, and maintenance.

4. Define a consistent machine state model: the key to comparability

To manage a heterogeneous fleet, you cannot rely on the raw states from manufacturers. You need to build an abstraction layer: the data normalization.

We recommend a minimal but robust state model to start with:

• Effective production:The machine is actually machining (Active cycle + Feed > 0).

• Ineffective production:The machine is in cycle but is not producing (Waiting, Feed hold, Feed override at 0).

• Adjustment / Tuning:Time spent on preparation.

• Planned stop:Pauses, training.

• Unplanned stop:Breakdowns (machine in alarm and repair time), material shortages.

• Closure: workshop closure.

A common mistake is to want to detail too much from the start. Begin by solidifying these broad categories before refining.

5. Practical method: calculate a reliable OEE in 5 steps

Here is a proven methodology to structure your approach:

1. Define the scope: Which machines, which teams, and which time slots are involved?

2. Define qualification rules: Create a common lexicon for your workshop (e.g., "is a micro stop < 2 min a performance loss or can it be considered a production state?").

3. Distinguish the raw state of the machine from the consolidated status (which includes the reported state): Record the data as it was collected on the machine to validate data consistency.

4. Separate the times: Clearly isolate Adjustment, Maintenance, and Production.

5. Control by sampling: Regularly check the consistency between automatic reporting and the reality on the ground (especially classification discrepancies).

   
   

6. How Atsora facilitates analysis on different machines

At Atsora, we know that value does not lie in the collection of raw data, but in its transformation into decision-making information. Our approach is distinguished by our ability to handle the heterogeneity of machine fleets.

• Multi-protocol connection:We connect to over 20 different protocols and manage more than 180 configurations, making the system independent of manufacturers.

• Standardization and Normalization:Our technology translates the specific language of each machine into a unified data model. An "automatic execution" state on a Fanuc machine is mathematically comparable to an "automatic execution" on a Siemens machine.

• Unbiased control:Thanks to this normalization, our dashboards provide an accurate and comparative view of your performance, essential for making the right investment or continuous improvement decisions.

  
  

Key points

• OEE is useful but incomparable without a common data model.

• The heterogeneity of the fleet amplifies calculation biases (different CNC vocabulary).

• States such as the feed hold, manual mode, or adjustment must be precisely qualified.

• The normalization of machine states is the sine qua non condition for reliable reporting.

• Raw machine data must be translated and contextualized to become usable.

• 
  

Do you want to ensure the reliability of your OEE calculation and gain a clear view of your machine fleet?

At Atsora, we transform your technical data into performance levers. Feel free to contact us to discuss the normalization of your indicators.

FAQ - Frequently Asked Questions