Predictive Maintenance is by far the most widely implemented use case in the industry, but also may the most frequently failed use case.
The reasons are many, but the most important reasons are robust machines that do not fail frequently and consistently enough to create patterns, and in addition most companies have not got the data necessary to make predictive maintenance a success. The result is in most cases that companies resort to anomaly detection, which as an unsupervised method is much harder to succeed with. So it is time for the industry to move beyond predictive maintenance.
Adding to industrial control systems opens wide opportunities to reduce cost, scrap and greenhouse emissions while improving quality and yield. So the opportunities are many. Current control system architecture is based on technology from the last century with PLCs, SCADA and DCS systems. They are generally not cloud connected or offered as a Service but provide an excellent data source for solutions. For higher value processes and more complicated control challenges many companies have opted for APCs. Again, generally not cloud connected or offered as a service and also from last century.
or is offering a new generation of opportunities to optimize the control systems. As the uses data from the system they generally cost less to develop and is faster to implement than APCs. Another benefit with is that they learn continuously while APCs tend to deteriorate in performance over time as they don’t adapt to changing dynamics beyond their programming. We are also seeing products being developed that target parts of the value chain offering even quicker time to market for the industry as well as cloud-based software as a service solutions.
A practical example is JUMO who implemented a solution in the manufacturing process of platinum thin film temperature sensors to improve quality and yield. The goal was to control a process early in the process chain which determines the resistance of the final product. History showed that the process variations between these points led to problems with the yield of sensors at the final measurement. The interactions at all process steps were too complex and some of them not fully understood, so they could not be taken into account to properly compensate for the variations. Despite having seen several data science challenges such as limited data, time lags in measurements, and some process unknowns the project was a success with 10 percentage points improvement in the yield of sensors with the highest accuracy class.
Control systems are the backbone of industrial production and are everywhere both in the process and manufacturing industry. The examples of where can boost basic control systems stretch across basically all industrial production. Illustrative examples are; reducing energy and environmental footprint in the chemical industry by optimizing the furnace process, using to optimize the cooking process in pulp production where it not only reduces energy but also the use of chemicals, and optimizing the robotic manufacturing in discrete cell-based production.
Looking closer at these we can see that they have all got data science challenges. is in particular interesting where there are unknowns, as in the cooking process for the pulp and paper industry, where the can find relevant patterns. Cooking is a process that is not possible to understand exactly what is going on inside the process. Operational experts together with the control systems can be supported by advanced to optimize the control. Over time many processes can even become autonomous.
About the Authors:
Martin Rugfelt is an entrepreneur that has spent the last 14 years of . Martin is passionate about startups and has been or is involved, as an advisor or board member in several startups working in as diverse fields as predictive healthcare, financial , , natural language and text generation. Martin is the founder and CEO of Industrial company Sentian.ai. He is also a self described foodie with a passion for gilling over open fire and winter pizzas in the cold of Sweden.
Ivan Jursic has a background in condensed matter physics and optics. He has a PhD in experimental physics. During the postdoc time he was a Researcher at the German National Measurement Institute (PTB). The focus of the research was on the ageing effects in high temperature thermometers for the next generation of power plants. After a stint at the University of Applied Science in Lemgo he went to JUMO. At JUMO he is in the R&D Department. The area of work includes development of thin film platinum resistance temperature sensors and the processes involved in the manufacturing of the sensors. The work of Ivan is driven by the desire to understand how the world functions from a sub-atomic level to every day life and beyond.
CognitiveVirtuals are regular worldwide-reaching online events bringing dozens of global leaders and experts together to share their views, experiences and expertise in the development of to the benefit of business and society. These 3 hour-long events are transparently addressing the development of cognitive technologies – including successes and challenges – while reaching and connecting a global online community of over ½ million followers.
All the sessions and formats are strictly content-driven with a non-sales approach, allowing focused and open discussions with content only. These events provide not only a platform to brainstorm and network but also to position experts, leaders, organisation, research developments, the current status and future outlook of .
Check out our upcoming CognitiveVirtual HERE