Who this is for: production and IT managers using machine data and indicators to support operational decisions.
Before developing an industrial dashboard, identify the decision it should support. Understanding line stoppages, checking an order’s status and comparing scrap levels require different data and rules. Collecting a large number of signals does not guarantee they answer the question.
Check data quality throughout the journey from machine to report. A value can be correct at its source and become misleading if assigned to the wrong shift, counted twice or displayed without showing that it is out of date. The project therefore needs controls and exception handling as well as charts.
Define the meaning before the number
Write an operational definition for each indicator that production and IT can both understand. Does the piece count represent items started, completed, accepted or packed? Does downtime include changeovers and maintenance? If departments give different answers, the first task is agreeing the vocabulary.
Availability, quality and format are distinct data issues, as NIST’s manufacturing analytics work discusses. A definitions register should accompany the signal list: name, meaning, unit, source and aggregation rule.
References: NIST — Data availability for manufacturing analytics
Prepare a data contract
A data contract describes what one system supplies to another. It need not be a lengthy document: a table agreed with the responsible people may be enough for an initial line. Communication standards and information models help systems exchange data, but the project must still define the meaning of the information being used.
Scroll horizontally to see every column.
| Field | Practical question |
|---|---|
| Source | Which machine or application produces the value? |
| Unit | Is duration measured in seconds or milliseconds? |
| Time | When did the event happen, and when was it received? |
| Quality | How are missing or stale values identified? |
| Identity | How are event, order, batch and machine connected? |
| Aggregation | How does an event contribute to a shift or daily total? |
References: OPC Foundation — OPC UA overview
Test situations that distort the indicators
Compare a data sample with actual observation. Check a shift change, a counter reset, temporary disconnection and repeated delivery of the same event. Define the expected outcome of each case before testing it.
For example, a cumulative counter is not a quantity to add at every reading. After a reset, a negative difference requires an explicit handling rule. Equally, an old value should not appear to be a newly collected measurement. These are illustrative test cases to adapt to the system, not reported client data.
- Distinguish missing events from values that are genuinely zero.
- Identify duplicates and delays while retaining an event trail.
- Agree the production calendar and shift-boundary rules.
Connect every chart to a decision
A useful view includes context, reporting period and data status. Someone investigating a deterioration should be able to distinguish a production issue from interrupted data collection. The interface needs clear labels for these conditions.
Start with a small number of questions and review the report with its intended users. Agree an owner, review frequency and possible action for each indicator. A number with no intended reader or associated decision may simply add noise. Extend the monitoring scope after establishing that the first view accurately represents the process.
Before requesting a dashboard
- List the operational decisions the report should support.
- Agree definitions and units for the principal indicators.
- Identify the source, timing and quality of each required data item.
- Prepare cases covering resets, delays, duplicates and missing data.
- Name the people who validate the data and use the result in production.
Common questions
Can we start from existing spreadsheets?
Yes. They can help reveal definitions and reporting needs. Check provenance, manual edits and calculation rules, however. The new system should not automate a formula without establishing what it represents.
Do we need every data point from every machine immediately?
No. A limited initial scope lets you verify collection, meaning and practical use before expanding. Establish from the outset how additional machines and indicators will be incorporated.
Sources and further reading
Apply this to your project.
Tell us which decisions you want to improve and what data you collect today. We can frame the discussion around monitoring, integration and industrial software using the actual production process.
Discuss your industrial software