Case study – Production Anomaly Management in the Chemical Industry

Share on linkedin
Share on facebook
Share on twitter
Industrial scene on Nueces Bay in Corpus Christi, Texas.  Original image from Carol M. Highsmith’s America, Library of Congress collection. Digitally enhanced by rawpixel.

How to manage and avoid quality issues and production anomalies in the chemical industry and beyond.

A Swiss chemical company producing a chemical intermediate product which should be a white powdery substance had the problem that the white powder occasionally had a brown discoloration. This discoloration was not affecting the properties of the substance, however, the batches with the discoloration still had to be discarded as they didn’t fulfill the quality expectations of their customers. 

The problem was that finding the exact reasons for the discoloration was a very difficult task, since the discoloration happened rarely and only a small amount of data points (~20) was available. Additionally, the production process had almost 30 parameters that could be responsible for the discoloration. The parameters spanned from the different suppliers of raw materials, over temperatures and timings at which raw materials were added, to a wide variety of other parameters. This made finding the right parameters responsible for the discoloration almost impossible with standard procedures. 

Project aim

The aim of the project was to determine the cause of the discoloration by using the ~20 data points of the discolored results with additional ~40 data points of results with no discoloration. The first objective was to identify if the production anomaly originated from one of the suppliers or within the Swiss chemical company itself. 

The next step was to identify the parameters with the highest impact on the probability of discoloration. Finally, we compiled a brief report with suggestions for resolving the production consistency issue. 

To make the task even harder, all the steps to reach a conclusion should be automatable, so that the system could be later connected directly to the production management system to analyze and mitigate production anomalies with minimal human intervention.

Results

After a short data preprocessing, we analyzed the given data points and used xT SAAM to generate correlation models. None of the generated models included the supplier as a variable, immediately suggesting that the cause of the discoloration has to be searched within the Swiss chemical company itself, not its supply chain. 

In the next step we identified the 5 most significant variables that affect the discoloration and provided a human-understandable linear model on how these variables interact with each other. We only required 2 days to find these results, and by further automating the process we could reduce the required time to less than an hour. 

We compared our results with the results of an internal investigation of the Swiss chemical company that included a Design of Experiments expert and several of their production specialists over the last months. The conclusions of both investigations were very similar, however, our models had greater depth than the internally created models. 

Conclusion

Within the project we could prove that our solution can significantly reduce time and cost in the analysis and mitigation of production anomalies. Additionally, we could also show that all required steps can be done with limited or even no human interaction, which allows the process to be fully automated and integrated into the production management system.

We are currently in discussion for follow-up projects and the integration of our system into the production management system of the Swiss chemical company.