As we have focused on the definition of a smart agent and its benefits for organization in our last article it is now time to get more practical by providing you with some clear recommendations for the first steps in introducing an intelligent agent in your organization.

During manufacturing processes, large quantities of data are being collected via different kinds of sensors. Pressure, temperature, accelerometers to name are a few of the sources which are commonly used. Taking decisions just based on the raw data is a very difficult task and is a time consuming process. Hence data needs to be transformed in a way it is easier to interpret and take suitable decisions to improve efficiency.

Understanding raw data and creating suitable transformations is not a trivial task. One needs to go through a series of steps before deciding on the suitable transformations. We at Tavrit GmbH believe and follow the AI for Manufacturing framework developed by Tvarit GmbH, to tackle any given problem in Industry 4.0 AI transformation.

Tvarit GmbH has built an architecture that utilizes the information collected from the data via different sources, consolidates them, and provides the best outcomes to the stakeholders.

To build a successful Data Science project in Manufacturing Industry, we find it critical to follow this framework which consists of the following components:

  • Understanding the challenge or problems faced by the business.

  • Understanding the process

  • Data Availability

  • Problem Statement Identification

  • ROI of Use case

  • Architecture

  • Data Science Pipeline

Process Understanding

"If you quit on process, you quit on results”, is a famous quote by Idowu Koyenikan.

Understanding an industry domain is an important step and central to all data driven solutions. Without this critical step, developing data driven solutions would be similar to shooting arrows in a dark alley and hoping one of them hits the target, and therefore can be very time and resource consuming. With domain knowledge, one can stream the whole process. E.g. Temperature sensors are quite often used in various manufacturing processes, in steel manufacturing, temperature varies at a slower rate and hence window sizes used to derive analytics will be in the order of 10s of minutes, whereas when used in monitoring drill bit degradation, window sizes are in the order of seconds.

Another important information that one gathers with domain knowledge, is the interdependence of different data sources. This ensures that certain safety nets are being placed around the development of the solution to drive it towards the end aim.

Data Availability

Now that you have the understanding of the involved business/ production processes, it is crucial to answer the data related questions. These are for example: What data points are available for analytics? From which machine and which process is the data collected? And furthermore, it is also important to know what data is missing which can be collected for better results.

3V’s of Big data

We all know, to begin with data analytics we must have the answer to the 3V’s, i.e. variety, velocity and volume. The 3V analysis gives us a clear picture of

  • What different varieties/formats of data do we have in hand?

  • What is the frequency of data collection?

  • What is the size of the data?

This helps in designing the architecture, machine required for computation, how many ETL (Extract Transform Load) scripts must be written etc.

Problem Statement Identification

The purpose of the problem statement is to identify and explain the problem. This includes describing the existing environment, where the problem occurs, and what impacts it has on users, finances, and ancillary activities. machine and which process is the data collected? And furthermore, it is also important to know what data is missing which can be collected for better results.

As part of understanding the domain and the available data it is also crucial to gain knowledge about the challenges faced by the production, e.g. high scrap rates, unplanned machine shutdowns and tool breakages. This can be found out by having conversations with domain experts and people from the specific backgrounds like the quality, maintenance or production department.

Problem Statement Identification

From the defined problem statements, we need to select the use case that has the highest impact and also readily available data. In the end, the aim of our data science activities is to improve the efficiency of the involved manufacturing processes.

If you want to stay up to date on this topic, follow us on LinkedIn and don't miss another part of our blog series "Data Science projects in the Manufacturing Industry". If you have any questions or feedback, don't hesitate to contact us!