Streamlining data acquisition process

With the increased use of sensor technology and connected devices, data analytics has found many applications in the manufacturing industry. Machine data can be used for use cases such as machine utilization, OEE performance, predictive maintenance, or bottleneck analysis. To realize these use cases, the basic necessity is to have relevant data. It is a common belief, that to start with AI use cases you can simply start with collecting data in a data lake.

But stop! Is that the right approach?

Creating such a data lake also requires lots of resources. Only having a data lake does not guarantee that the right data is available for the planned use cases.

The focus should be on gaining the right data and collecting data that is valuable.

Integration of data from shopfloor into IoT platforms is resource intensive. Often at a later stage in the project, the realization occurs that certain additional data is necessary.

‍

The challenges

There is a multitude of challenges when acquiring data. One of the main challenges when it comes to the usability of collected data is that data is often not in the right format. Unstructured data with no clear data models increases the complexity of data management and processing. This data needs to be cleaned, fine-tuned, structured, and labeled before it can be further used. In practice, different departments often use different terminologies for the same entity. Such discrepancies should also be eliminated with the correct labeling of data to maintain consistency.

‍

Another challenge that occurs with available data is that it might not be suitable for the target use case. Consider an example of condition monitoring or predictive maintenance where high-frequency data of machines might be needed. Such data is typically not readily available. Sometimes even if the right data is available, it is not available digitally. For instance, the quality parameters for an inline quality check are recorded on paper due to the lack of in-line sensors. Nowadays machine manufacturers equip machines with sensors and provide packages. These packages must be purchased to access machine data. There are also cases that may risk the infringement of warranty rights when tapping data from existing PLCs provided by the machine builder.

‍

Actual process times are often missing from the available data. So, it is difficult to compare real operational times with standard times and find deviations. Taking an example of an assembly line, reworks, and auxiliary activities may not be recorded in the MES system. This leads to distorted cycle times and complicates data analysis.

‍

With all these challenges, data collection and preparatory data acquisition activities become quite intensive and exhaustive.

‍

The solution - start by answering two questions 

1. Which data is required for the use case?
2. How to collect the required data?  

‍

To answer these questions, here are some pointers:

• Develop use cases concept and focus on use cases with clear business potential. The selected use cases should be aligned with overall business and strategy goals. Also, they should have direct impacts on defined KPIs.
• As a starting point for any data acquisition use case a data gap analysis should be conducted to create transparency on the status quo of available data and data quality. This helps to avoid a time-consuming trial and error proceeding.
• Data gap analysis helps to discover data availability and find out missing data. It also helps to identify the next steps and propose an action plan for getting the right data.
• The potential action for data acquisition can be conducted through sensor retrofit, implementation of digital documentation tools, or by defining and monitoring KPIs.

‍