Production Analysis Using Machine Data: Methods and Examples

Production analysis combines machine, process, quality, and product data to systematically examine production performance and behavior. It not only answers the questions what happened, but also helps explain where, when, and under what conditions losses or deviations occurred.

Typical indicators include declining output, longer cycle times, frequent malfunctions, increasing scrap, or unexplained variations between shifts and products.

What is production analysis?

Production analysis is the structured evaluation of data generated during manufacturing. This includes, for example:

• good parts produced, scrap, and rework;
• Start and end times of production steps;
• Machine statuses and downtime;
• Errors, Alarms, and Warnings;
• Process parameters such as temperature, force, pressure, or torque;
• Product, order, batch, and variant information;
• Quality and test results.

The key added value comes from making these connections. An increased scrap rate only becomes actionable once it is clear at which station, for which product variant, and under which process conditions it occurred.

Distinction from Monitoring, Reporting, and Process Analysis

Production Monitoring

Monitoring shows the current status: Which system is running, where is a line located, and which alert is active? It supports operational responses but does not automatically explain longer-term patterns.

Production Reporting

Reporting summarizes key metrics for a defined period. For example, it provides information on how much was produced or what the OEE was. The presentation is often at an aggregated level.

Production Analysis

Production analysis examines key performance indicators in context and allows for a drill-down from the results to the workstation, product, event, or process parameter. It is used to formulate and test hypotheses about the causes of losses.

Process Analysis

Process analysis examines the actual sequence of events across multiple steps: variations, backtracking, wait times, and repetitions. It is particularly relevant when the goal is to explain not only the outcome of a station but also the actual path a product takes through the facility.

What questions can production analysis answer?

A good analysis begins with a specific production question:

Section:	Typical Question:	Appropriate Analysis:
Output	When and at which station does the output decrease?	Output by time, line, station, and product
Cycle Time	Which products or process steps are running slower than expected?	Distribution and trends in cycle times
Availability	Which states cause the greatest time losses?	State duration and state transitions
Malfunctions	Which messages have the greatest impact on output?	Frequency, duration, sequence, and impact
Quality	Under what conditions do scrap and rework occur?	Comparison of process parameters and quality classes
Traceability	What happened to a specific part?	Part history by stations and process steps
Improvement	Has an implemented measure had a lasting effect?	Controlled before-and-after comparison

The Necessary Data Set

Production data is often stored in different systems and structures. Control systems provide signals and status information, inspection systems contain quality metrics, and MES or ERP systems provide additional context regarding products and orders.

For reusable analyses, raw data is converted into domain-specific models:

• The production log describes manufactured parts, process steps, cycle times, and quality.
• The machine state log tracks operating, downtime, setup, and maintenance times.
• The event log standardizes errors and alarms by recording their start time, end time, duration, and system reference.
• KPI models define calculations, filters, target values, and groupings.

Key unifying characteristics include a uniform time reference and consistent identifiers for plants, stations, parts, products, and orders.

Seven Steps from the Symptom to the Cause

1. Identify the production problem

Make your observation measurable. “The line isn’t running well” is too vague. A better way to phrase it is: “The hourly output for product group B has been below the target for the past two weeks during the late shift.”

2. Define the metric and comparison

Define the calculation, target value, time period, and appropriate comparison group. Comparisons are only valid if the product mix, planned timeframe, and other contextual factors are taken into account.

3. Narrow down the discrepancy

Group the results by line, station, product, shift, or time window. The goal is to identify the smallest relevant context in which the deviation becomes reproducibly apparent.

4. Link Related Events

Link notable production cycles to machine statuses, notifications, quality events, and process parameters. A timestamp alone is not always sufficient; part or activity identifiers increase reliability.

5. Identifying Patterns and Formulating Hypotheses

Look for differences between unremarkable and notable cases. Examples include specific product variants, parameter ranges, message sequences, or recurring state changes.

6. Validate the cause from a technical perspective

A statistical correlation may be caused by a common third factor. Production, process managers, quality assurance, and maintenance must verify whether the identified pattern is technically plausible.

7. Review the measure and its impact

Document the change, the timing, and the expected impact. Then compare “before” and “after” periods that are sufficiently long and similar in nature.

Key Analytical Methods

Time Series and Trends

Time series show when output, cycle time, scrap, or process parameters change. They are useful for identifying drift, shift patterns, and changes following interventions.

Distributions Instead of Averages

A mean value can mask unstable processes. Histograms, quantiles, and measures of dispersion reveal whether individual outliers or a broad shift are influencing the result.

Groupings and Comparisons

Comparisons by product, workstation, shift, tool, or recipe help identify the relevant context. Each group should contain a sufficient number of comparable cases.

Pareto Analysis

A Pareto analysis prioritizes defects, downtime causes, or sources of scrap based on their overall impact. In addition to frequency, factors such as duration, loss of quality, and affected output should also be taken into account.

Correlation and Event Context

Process parameters can be compared with quality or cycle time results. Correlations provide clues, but they do not replace a technical root cause analysis.

Process Paths

When sequence, backtracking, or wait times are critical, process analysis supplements the key performance indicator view with real-world workflows involving multiple stations.

Example: Increasing rework at a test station

Suppose the rework rate increases, while output and availability initially appear stable.

A possible analytical approach:

1. Compare quality grades by product variant, inspection station, and time period.
2. Assign the affected parts to the previous process steps based on their identification numbers.
3. Compare the process parameters of good parts and reworked parts as a distribution.
4. Link unusual parameter ranges to the tool, material batch, and machine status.
5. Review a technically plausible hypothesis with the people responsible for the process.
6. Implement the change in a controlled manner and continue to monitor the rework rate and process distribution.

The analysis begins with a key metric but extends through the part history to specific process conditions.

Common Mistakes

• Start with all available data: This requires effort before the analytical question is clear.
• Consider only average values: Variation, outliers, and different product mixes remain hidden.
• Confusing correlation with causation: Abnormal parameters must be technically validated.
• Using inconsistent KPI definitions: Different filters and time models prevent reliable comparisons.
• Failure to document interventions: Without a timeline and expected outcomes, it is virtually impossible to demonstrate effectiveness.
• Building Dashboards Without Drill-Downs: Visibility alone does not lead to the root cause.

Requirements for Production Analysis Software

A suitable solution should:

• Jointly analyze machine, process, quality, and product data;
• navigate from metrics to events and individual parts;
• support flexible groupings, time periods, and comparison groups;
• Keep data models and KPI definitions transparent;
• Take into account on-premises, edge, and existing IT architectures;
• Document findings, actions, and before-and-after comparisons;
• Provide open interfaces for BI, data science, and custom applications as needed.

Bytefabrik provides these analysis paths through Manufacturing Insights. You can find the specific product feature under "Production Analysis." For information on the overarching improvement methodology, see "Production Efficiency: Definition, Key Metrics, and Improvement."