How to expand analytics capabilities
Increasing adoption of analytics bridges the gap between process experts and data scientists, encouraging manufacturing collaboration
Too often, industrial manufacturers find their process experts and data scientists have different problem-solving approaches. For example, production savvy process experts may be well versed in analytics know-how, but bandwidth constrained when it’s time to conduct the analysis. In contrast, the analytics savvy, mathematically inclined data scientists lack the production knowledge their process expert counterparts possess (see Figure 1).
When an unanticipated event in production takes place, typically process experts try to address the anomaly on their own. Only when they are unable to address it, they ask for the help of data scientists. With multisite manufacturing facilities, the role of the central data science team involves supporting and responding to inquiries and communications from the various plants, which has been known to bottleneck the analytics chain. Unfortunately, this can occur due to the number of requests, swamping the central data science group and causing great delays in feedback or even worse, leaving the process experts out to dry with no responses to their problems.
Increase analytics maturity
The best way to have the process engineer and data scientist adopt a more successful approach is to expand the analytics capabilities. It is important to grow the analytics maturity of process experts using self-service analytics tools because it can enable this team to do more of the data analyses themselves. This would allow them to become self-sufficient with analytics, empowering them to address most production questions without having to ask for support from other sources. For the most complex issues, the central analytics team would help to troubleshoot.
Self-service analytics can help solve about 80% of daily production and process questions that arise. This lends data scientists more time to concentrate on the remaining 20% of complex issues and enables them to be more effective. Taking such a problem-solving approach yields a win-win combination by using both teams’ knowledge and experience.
In recent years, analytics-enabled process experts are evolving beyond this level of analytics maturity. They are eager to increase their analytics knowledge and appreciate the greater levels of flexibility they can attain in regard to their ability to perform all kinds of analysis on their own. The new trend points to a new breed of data scientists now termed as “citizen data scientists.” Comprised of employees whose positions may be outside of the analytics field, citizen data scientists who are now equipped with self-service analytics can perform advanced diagnostic, predictive and prescriptive analytics.
Team optimization with embedded notebooks
Process manufacturing and operational experts can apply self-service analytics software to analyze, monitor and predict operational performance using sensor-generated time-series data. The objective is to empower engineers with analytics software to improve operational excellence, without the need of data scientists.
The next generation version of self-service analytics software in the industrial analytics realm incorporates an embedded notebook (see Figure 2). Having access to more advanced capabilities affords users greater flexibility and a better analytics experience. Process experts can prepare operational data themselves using advanced trend analysis as well as load data views into the notebook. Using data science libraries of their choice, users can create and run the scripts themselves. Any valuable outcomes such as new (predictive) tags and related monitors or notebook visualizations can be presented as dashboard tiles and made accessible throughout the organization.
Derive more meaningful value
The use of embedded notebooks is helping process experts become more proficient at analytics and more confident as they progress closer to the level of data scientists. As process experts become adept at trying out various algorithms and using the notebook technology, they can attain more value from the data in ways they could not before, for example:
- Increase process visualization, adding extra visuals to the dashboard and enhanced reporting for such maps/plots as heatmaps, tree maps or joint plots.
- Conduct statistical analysis that compares sets of production data (good and bad such as t-test/ANOVA).
- Use low pass filtering or exponential smoothing to eliminate data compromised from seasonal effects.
- Develop a nonlinear predictive model to predict quality such as using neural networks.
- Use self-service analytics search results to perform advanced mass balance calculations.
Even with process experts increasing their analytics maturity level, data scientists’ inherent knowledge and abilities in mathematics/statistics/AI/ML/modeling theory, will always create a kind of boundary, marked by skill set, knowledge and expertise among the two teams. Process experts think in terms of their production process and look for trends in their data such as any anomalies that may be stagnating production as well as why it might be happening. On the other side of the coin are the data scientists that tend to be more interested in algorithms, cleaning data and using models. While they are typically not self-service analytics users, they can become users, especially supported by the new notebooks.
A more collaborative environment specifically between the data scientist and process expert can be realized due to these new notebooks integrated with self-service analytics software (see Figure 3). Both data scientists and process experts will be able to more easily work together on joint projects since they can directly continue their work using the same tool in which the process experts prepared the data ahead of time. The result makes for a collaborative, cohesive and interactive team of process experts and data scientists that can more effectively address potential process problems. In addition, this frees up data scientists to concentrate more on projects for the process experts if complex use cases arise.
Analytics use case example
Global specialty chemical company, Clariant Corp. produces care chemicals, natural resources, catalysis and energy, and plastics and coatings. Its Germany plant began to use a self-service analytics software tool and after much success, expanded usage to all of its sites in North America and China, and soon will be extending deployment worldwide.
The self-service analytics software brought collaboration among different people from varied cultures and backgrounds. The software helped enable valuable exchanges around various key process ideas, best practices and potential process improvements. The result was increased collaboration and coordinated global production. Production data and information from multiple continents could be compared across different sites. This eliminated data silos and enabled information to be shared on one screen for all the sites. What ensued are golden production batches, which could only come about using the analytics software and assessing batch run quality using golden fingerprints. Corrective action could be taken sooner, if needed.
Clariant serves as an example of how analytics software brought its process experts and the data scientists together and allowed them to use its strengths and unique skill sets. In fact, Nimet Sterneberg, a Clariant Data Scientist, felt the self-service analytics tool enabled him to “gather the data, getting the right data in the right samples” for his data science efforts (see Figure 4).
Clariant was able to decrease the amount of raw materials needed, and more importantly, decrease the waiting times in terms of the cycle times of its batches, which resulted in a tremendous cost savings. As process experts saw the benefits and improvements the tool helped them to make, the motivation to adopt and deploy the tool was fast and easy.
Increase analytics capacity
Through the advance of self-service analytics software that helps solve 80% of the operational day-to-day problems, self-service analytics tools for use by data scientists seems to be growing due to the flexibility of a notebook environment embedded with the software and delivers a one-click access to prepared contextualized data.
Both the process experts and data scientists are needed for solving operational issues. Luckily, with the increasing analytics maturity and process knowledge of asset engineers and operators, along with the growing collaboration between data scientists and process engineers fueled by the advances of these embedded notebooks, industrial plants and production facilities are positioned for great results (see Figure 5).
Thomas Dhollander, CTO at TrendMiner said, “Classical data science depends on bringing operational know-how to the data scientist, while self-service analytics aims at packaging a subset of the data science modeling capabilities and bringing these to the subject matter expert as a robust set of features (no technical tuning parameters, no data science training needed). Companies that recognize the potential in interweaving these complementary approaches will be the ones that can accelerate their operational efficiency and competitive advantage.”
By investing in this problem-solving approach, organizations can succeed with analytics and secure a competitive edge and successful future in industrial manufacturing. Due to our evolving industrial manufacturing landscape, it’s more critical than ever for both data scientists and process experts to forge new paths together and enable a data and analytics-centric culture.