The chief objective of ‘leanness’ in manufacturing is to eliminate waste, i.e. anything that does not create added value. One of the chief challenges for lean production is to identify these wasted resources. And here big data holds the key. To leverage this we will connect more than 30 of our own factories to one manufacturing data platform via MindSphere and implement a purpose-built Industrial Edge layer. This data architecture also enables a rich portfolio of applications like operational optimization, business analytics, and the implementation of machine learning.
Value creation is the number one priority when implementing such platforms. This entails getting the maximum out of the organization’s data landscape. Traditional IT architecture reaches its limits when it comes to scalability and usability. New architectural patterns need to be developed to harness the ultimate power of data! These are brought forward by Industrial Edge in conjunction with MindSphere data lake solution.
The supply of manufacturing data to the unlimited possibilities of current and future applications must be done without or low effort. This can only be achieved if data interfaces will be reduced or even avoided and all data and interfaces are available for reuse, which will lead to standardization in certain aspects. Basically, to fully capture the value of using big data in manufacturing the factories need to have a flexible data architecture, which enables the different users, internal as well as external, e.g. suppliers, to extract maximum value from the whole data ecosystem.
At the same time, it requires real time performance with low data traffic costs, to get added value use cases down to the shop-floor. Here, the Industrial Edge layer comes into the picture, which processes the data close to the sensors and to the data source (figure 1).
Industrial Edge and the data lake concept together will enable a faster and more powerful solution development than any other data storage and utilization concept. Regardless if you are coming from the shop-floor with a clear operational problem, from the central department to create business reports, from the data scientist point of view to look for new patterns by using analytic sandboxes, or from the data analyst view, introducing new business models like data-as-a-service or even analytics-as-a-service.
Let’s have a look at the conceptual construction of a manufacturing data platform with MindSphere and in this context also Industrial Edge. To start with, here are some general statements:
- The platform will be a colossal storage area for all manufacturing data and will therefore be tremendously powerful for all user levels
- It is a centralized and indexed aggregation of distributed organized data sets
- Big data will be stored independently of its later use, this means as raw data
- With Industrial Edge, the manufacturing data platform is the prerequisite for effective and scalable cloud computing and machine learning
- In this architecture Industrial Edge has multiple purposes like data ingestion, preparation, security-gate, real-time decisions etc.
- Highly integrated, but module- and service-based ecosystem functionalities, e.g. importing/exporting, persisting and analytics (in order to avoid redundancies, it is possible to refer instead of persisting)
Classic Data Warehouse Architecture
To see where the manufacturing industry is coming from and grasp the novelty of the new architectural approach to managing data, let me say a few words about a classic data warehouse architecture.
The classic “Data Warehouse” architecture pattern generally follows the philosophy of understand, transform, load and analyze data. Data will be extracted, transformed and loaded (ELT) from data sources to the data storage.
While this is taking place, some data cleansing and structure creation is performing. Data models are predefined and enable to create department specific reports, thanks to the Online Analytical Processing (OLAP)-cube dimensions. The OLAP-cube allows you to slice, dice, pivot, drill-down, -up and -through and self-service business intelligence (BI).
This drives in two major prerequisites which are also the reason, that this pattern cannot scale the possibilities required for big data analytics. The first prerequisite in this philosophy is, we need to understand the data first with all its consequences like “are there any anomalies”, “what is the source system, cardinally etc.” This is extensive and complex work.
The second, as an implication of the first prerequisite, is that compromises and choices must be made on which data is to be stored and which data is to be discarded. This definitely limits future possibilities for chronological reviews or trend analytics beside the fact, that the effort to decide, which data to bring in and to leave out, is tremendous.
Finally, more time will be spent with data administration than with data analytics to uncover valuable patterns.
Find out how we solves this issue in my next blog: The North Star and its Data.