6 Ansätze zur Datenanalyse in der Industrie

Complex technical systems such as machines, equipment, measurement and control units, or production lines generate substantial data today. To gain insights from this data that lead to improved products and new innovations, engineers rely on finding the right information within the data.

The result of data analysis can be, for example, reports with important key metrics that provide clues about what is happening in a process or on a system. The more efficiently the process data collection → insight generation runs, the more good ideas engineers can develop. And these ideas ultimately determine your company's innovation capacity and competitiveness in the medium term.

The link between data collection and insight generation is data analysis. In this process, raw data is transformed into meaningful information from which engineers can derive new knowledge about technical processes and effects. This article describes five approaches with which such data analyses can be performed.

Approach 1: Do It Yourself

Due to the multitude of heterogeneous data sources in industry, relevant data is often distributed across various databases. Other data may even only be available in Excel files. In the "do it yourself" approach, an engineer or software developer writes a script that reads data from databases via SQL queries or from Excel files via VBA routines, and implements processing routines to convert the flood of data into presumably relevant information.

This approach initially seems practical and cost-effective, but brings some drawbacks in the long term:

• Implementation costs: Implementation and maintenance of the software routines are time-consuming and error-prone because information from many different data sources must be integrated. Each individual database connection must be maintained, and simple changes to database tables lead to errors that are difficult to find.
  
  
• Blind spots: In-house developments often arise out of necessity or spontaneous action and are carried along as isolated solutions without clean interfaces. When the feeling takes hold that not everything is being extracted from the data, helpful support through data analytics and AI technology can only be added with considerable effort.
  
  
• Double burden: If setting up and maintaining data analyses is not part of your core business, there will be no dedicated personnel resources for these activities. In this case, the most valuable specialist staff typically take on these tasks part-time. This works reasonably well during setup, but with increasing questions and wishes from colleagues—naturally always when there's no time anyway—it increasingly becomes a painful double burden.

This approach is not recommended for implementing a long-term data or AI strategy. However, the approach can work well for a lean pilot project to get an initial idea of the information content in the data. The startup costs are low and a data scientist is not necessarily required yet. The biggest challenge is to catch the right moment to switch to a professional solution (approaches 2-5) or to build a dedicated department (approach 6).

Approach 2: Development Project

A development project aims to create software for a clearly defined task. For product development, the result can be a software module that processes product and measurement data and transforms them into a series of visualizations and KPIs that are defined during the requirements analysis.

If you cannot or do not want to carry out all work with internal resources, you can rely on external development service providers. Established service providers specialize in the rapid implementation of complex development projects, increasingly also in the AI field. In addition to software service providers, which are mostly geared toward economic aspects, some research-focused institutions also offer services in the competency area of AI development.

However, if you want to avoid dependency in the event of success, you will have to build up your own personnel resources in parallel.

Approach 3: Business Intelligence (BI) Software

BI software such as Power BI or Qlik Sense excels compared to the "do it yourself" approach with significantly reduced implementation and maintenance costs for data analysis. Data integration is greatly simplified because data sources can be connected via the software's dashboard. And visualizations as well as simple statistical evaluations are standard features of common BI software.

However, due to the broad range of functions, the training period should not be underestimated, and as with the "do it yourself" approach, you need a power user as a contact for colleagues. The further processing of insights gained from the data, i.e., the "action," takes place outside the BI software. It is also important to note that BI software traditionally does not offer the ability to evaluate data using Artificial Intelligence (AI) (Business Intelligence vs. Artificial Intelligence-based Analyses). To that extent, BI software can produce "reports deluxe," but it remains a reporting tool.

Approach 4: Data Science and ML Platforms

On data science and machine learning platforms such as RapidMiner and Azure ML, users can develop their own data and analysis workflows and also have extensive possibilities to adjust the parameters of statistical procedures and machine learning models. This makes it possible to perform data analyses that go far beyond the horizon of simple reports. For widely used applications such as image or speech recognition, there are sometimes even pre-trained AI models and methods available.

These platforms offer many powerful capabilities, but are primarily suitable for data scientists and AI developers. For engineers, the handling is very far removed from the tools of their professional domain and requires a deep dive into the AI and programming world. Such platforms make sense if there is already a dedicated department exclusively dealing with data science. As a rule, the human, specifically the data scientist, remains the central link in the analysis and utilization process, as many platform results require explanation. The transfer of platform results into scalable applications or apps for daily operations is an entirely separate task.

Approach 5: Configurable (Low Code) Software Solution

In the fifth approach, engineers work in a greatly simplified software environment whose functionality and analytical capabilities are specifically tailored to the application context. They can read data from various data sources via a dashboard and then visualize and analyze it, as BI software from approach 3 allows. However, the operation is greatly simplified and unnecessary complexity is avoided by using only the functions relevant to the application and reducing their setup to configuring a few parameters.

If the software solution is properly built, it can be supplemented with custom-tailored AI modules that can also be configured "low code," with minimal effort, for your own purposes. This makes it possible to automate complex tasks such as root cause analysis for unwanted events.

Solutions of this type enable close collaboration between engineers and data scientists in a shared collaborative environment without both sides needing to acquire much expertise from each other. Certainly, a certain dependency is incurred, but that is standard in industrial supply chains anyway. In the long term, costs can be significantly reduced and the focus of your own specialist staff on core business increased.

Approach 6: Dedicated Department

For sustainable in-house development, an internal team is required that must cover the necessary competency areas depending on the implementation strategy. In any case, a development engineer with domain expertise is needed for requirements analysis and interpretation of analysis results.

For a development project, software development and data science or AI expertise are also required. With the approach using BI or ML software, a data scientist is definitely required, and possibly software developers if the analysis results are to be embedded in a software pipeline.

If the goal is to develop your own software solution with which data can be flexibly connected and evaluated, an interdisciplinary team of front-end and back-end developers, data scientists, and AI experts, as well as DevOps engineers, is required. The latter create efficient IT processes for scaling and maintaining the AI application, particularly the regular retraining and fine-tuning of AI model parameters.

Further information can be found in our articles on the necessary competencies for implementing an AI project.

Conclusion

Companies continuously face the challenge of strengthening their competitiveness. In the industrial sector, the innovation capacity of the development department is the decisive factor for this. The better important insights for the development department can be gained from data, the greater the positive impact on the company's innovation capacity. For the critical link of data analysis in the workflow data collection → data analysis → insight generation, we have presented 4 common approaches to you.

At aiXbrain, we pursue the fifth approach with our software solution aiXbrain Dataray.

aiXbrain Dataray is primarily aimed at engineers who can connect and evaluate data sources via a simple user interface. The software provides ready-made, industry-proven analysis modules that need to be configured with only minimal effort depending on the specific requirements. If the solution proves itself during a test phase, it can be scaled to any number of data signals (time series data) and data sources (e.g., machine data databases, equipment systems).