Improving quality control and maintenance at manufacturing SMEs using big data analytics

Experiment Description

Zannini SPA produces high precision mechanical components for hydraulic and pneumatic systems for cars and household appliances. The quality control of these components was running only on a statistical basis and there was no real-time feedback for adjustment of the process parameters. This led to undesirable process settings, a high waste rate, and a low-capacity index. The past monitoring process did not take into account all available quality data for improving the components quality.

Now that real-time monitoring and analysis was enabled, it will help the production process in order to identify anomalies and divergences at the moment of yielding the components. The HPC platform will execute and process the huge amount of data which will be produced from Zannini manufacturers, in order to run the simulations needed.

Technical and Economic Impact

It is important to understand that this work has led to a commercial service; D2Twin and you can find in the web page all the information regarding this new Technical and Economic development. Click on the image for access to the D2Twin web page.

CloudiFacturing provided the required infrastructure (through CloudBroker) and support for the integration (through the DIH).
We used a very valuable support for business modelling.
The validation of results helped also in creating a very user-friendly service.

EXPERIMENT OUTCOME

Based on this work, a service was developed in the D2Lab framework (d2lab.nissatech.com). It is commercially offered D2Twin is a data-driven “replica” of an asset (product, system or process). It is based on the analysis of the past behaviour of the asset and it is derived from past data using advanced data analytics techniques. D2Twin extends the concept of the Digital Twin from the point of view of how the asset has been used in the past. Analyzing the data generated by a concrete asset in the past operation, our approach derives the knowledge about the model of the asset. Traditionally, Digital Twin is the result of human expert-driven modeling, independently of its usage. Additionally, D2Twin is a digital replica of a concrete artifact and not the model of a type of assets (e.g. type of machine).
In the nutshell of the analytics approach is the possibility to understand process behaviour (deeply) and represent usual and unusual behaviour in the declarative way that can be used for the real-time monitoring and off-line exploration of the process, including quality control and process improvement.

Key advantages

• Applicability – it can be employed in any domain where the data is available
• Complexity – learning is unsupervised so that no human expert involvement is needed
• Scalability – model learning can be applied on high-dimensional process space
• Up-to-date model – the new model is periodically generated so that the model is always up-to-date, compensating model drift

In general context, the approach provides a significant contribution to the domain of Digital Twins that is one of the increasing, growing markets. According to new research published by Gartner, nearly half of the companies are planning to use digital twin technologies as part of their Internet of Things (IoT) deployments. Gartner found that 48 percent of organisations that are implementing IoT programmes said they are already using, or plan to use, digital twins in 2018. Moreover, every application domain brings unique opportunities for the improvement through Digital Twin development. For example, Gartner predicts that by 2020, at least 50 percent of manufacturers with annual revenues exceeding $5 billion will have at least one digital twin project in play for either products or assets.