Manufacturing Intelligence: From .xls to .ai

Despite all the discussion about digitization and Industry 4.0. Production is rich in data but poor in information. It is not uncommon for a perceived minimum level of productivity data to be recorded by means of imprecise hand writing. The potential of a modern “Manufacturing Intelligence” is not used. It’s not that difficult at all.

Even old controls record a basic amount of data that provides information about the status and potential of productivity. Modern systems naturally offer many more options, but more is not necessarily better. The message is: data is there and the problem of the lack of connectivity of the equipments has been solved. There is now the possibility for almost every machine to get data automatically and with very manageable costs. The time for handwriting is over.

What a connected system then generates is, in IT terms, time series data. The process parameters for the respective time unit are stored in a special database in small time increments. This can be, for example, number of pieces, percent feed or rejects.

The OEE can then be shown in real time with just a few measurement parameters. Once this technical and informational basis has been laid, the infrastructure can be expanded. In order to enrich the data stream with further information, it would be “annotated”. Which production order belongs to this time increment? What was the reason for the downtime? This informational addition can be done by further IT systems, as in the example, by sensors of the machine or by humans.

The resulting “enriched” time series is then in turn the basis for an analysis by algorithms. The type of data storage enables fast access and analysis in real time. Whether it is statistical algorithms or artificial intelligence depends on the question. Both are possible. There is also no limit to whether questions from the past should be answered – for example, finding clusters of certain losses in the data stream – or whether a statement should be made for the future – for example when the next loss is to be expected. The technology allows all application scenarios.

Image: Schematic process of a modern AI data analysis

For some questions, an algorithm must first be trained. In these cases, the question can be clearly assigned to the field of artificial intelligence. Understandably, a time series must first be built up in order to use the algorithms. This data collection time depends on the data processed and can last from a few weeks to many months. An AI algorithm can only provide reliable information when a sufficient data history is available.

The information can be displayed in reports and on Andon boards immediately after installation and configuration. Once the path described has been completed, the step to push notifications in the event of anomalies is only short. For example, scrap can then be communicated to those responsible while they are occurring. The speed of reaction increases, and so does productivity.

This is modern “Manufacturing Intelligence”: Analyzing the plant productivity on the basis of real-time data, making information available to specific employees and increasing the OEE through targeted measures. Please contact us.