With the help of industrial analytics, companies want to gain added value from production data. Intelligent data analysis software alone is not enough; it must also incorporate (human) knowledge about the specific application and the process.
Thanks to the digital transformation, ever larger amounts of data are produced in the production context (keyword: Industrial Big Data). Values such as temperature and pressure measurements or engine performance data have enormous potential if they are sufficiently analyzed and evaluated with the help of data analytics.
Definition of Industrial Analytics: Generic term for the evaluation of all data arising in industry, both machine-generated and from the interaction between man and machine.
The term Predictive Maintenance goes hand in hand with the term Industrial Analytics. Measurement and production data of machines and systems are to be used to derive maintenance information from them and, at best, to predict downtimes before malfunctions even occur. An example: Together with the Smart Data Solution Center Baden-Württemberg, the mechanical engineering company Hermle developed a method for evaluating machine conditions. The SDSC-BW team recorded maintenance data of a machine type for a period of twelve months. With the help of monitored learning methods, the experts derived an approach from this that enables automated evaluation of the machine state.
Smart data analysis does not necessarily require huge data volumes, Andreas Wierse stated, Managing Director of Sicos BW, who founded the Smart Data Solution Center Baden-Württemberg together with the Karlsruhe Institute of Technology (KIT): “Even a lot of small amounts of data can be profitable in combination with other external information. The decisive factor is that data analysis must be able to detect patterns or connections that provide valuable indications for possible process improvements.”
In addition to preventive maintenance, quality assurance is an important application field for industrial data analytics. For example, the mechanical engineering company Grenzebach uses an industrial analytics solution from automation specialist Weidmüller for real-time quality assurance for its friction stir welding systems. “The analytics software compares the forces detected on the sensors during the welding process with an ideal reference data set. As soon as there is a deviation that is outside the defined parameters, the machine operator receives a message and immediately knows that something is not right. This eliminates the need for manual inspection of each weld seam,” Weidmüller's data scientist Dr. Daniel Kress explained.
To determine the reference model, Weidmüller evaluated the data sets of more than 100 weld seams together with Grenzebach and evaluated them using intelligent data analysis methods. The know-how of a mechanical engineer is therefore an essential part of the analyses. Analytics software can predict an error with a certain probability, but the prerequisite for this is always that the error has been classified beforehand.
Deniz Ercan, Head of Nexed Data Analytics at Bosch Connected Industry, also confirmed the importance of application and process know-how of experts on site: "A data analysis specialist cannot simply march into a production process, pull data from different sources and let his software work for itself. Instead, the key lies in the interaction of data and human knowledge.”
In addition, the data must first be prepared. “Depending on the type and age of the machine, the data are often available in a wide variety of structures and formats. Accordingly, collection and standardization are often the most difficult and tedious step,” Mr. Ercan stated Only then can the actual analysis begin. Studies show that up to 80 percent of the effort involved in a data analytics project is invested in data collection, cleansing and preparing.
Bosch expert Deniz Ercan related a data analytics success story: “The quality of a sensor layer fluctuated greatly at a particle sensor manufacturer. Despite intensive research into the cause, however, it was not possible to get to the bottom of the matter.” The data analysis specialists compared all available data – including those that were only indirectly related to the actual problem – and found that a completely different sensor layer was responsible for the quality differences.
“That was surprising enough for our customer. But we were also able to identify a hitherto unknown problem with pseudo-scrap,” Mr. Ercan continued. Due to an error in the machine control, qualitatively flawless parts were classified as rejects. “Once known, employees were able to fix the problem immediately. That alone saved the customer approx. 1000 euros per day, and the cost of data analysis was amortized within a single week.”
The South African machine learning specialist Dataprophet can also report on data analytics successes in quality assurance, for example for a large foundry that manufactures engine blocks for Daimler. Dataprophet Manager Frans Cronje: “The plant was struggling with significant problems due to high scrap and rework rates.” The problem was solved by collecting production data of different formats (from Excel files to Access database data) for 15 months. Subsequently, Dataprophet determined the optimum operating parameters with a forecast model and identified engine blocks on which defects would occur. As a result, the scrap rate was reduced by 50 percent in the first month of operation, and the external scrap rate was reduced to zero percent within the first three months. Mr. Cronje: "Ultimately, the foundry was able to save a total of $1 million per year.”