The manufacturer uses three large factories to produce many of the materials used in modern automobiles, satellites, and weather equipment. The components are expensive to produce and extremely detailed; if manufacturing equipment experiences downtime, needs unexpected maintenance, or produces product defects, the ramifications can be translated into significant losses due to poor production yield.
To avoid potentially costly rejects, ABC began collecting information on their processes, machines, and facilities using data from line sensors, surveys and outcome statistics. Each of their three factories gathered vast quantities of data. To manage and find value in all of this newfound data, the company employed a team of eight data scientists to begin working on better ways to turn these piles of raw data into actionable, proactive information.
ABC faced two major issues with this approach. The first was the size, complexity and disparity of the data collected. In order for the data science team to find value in the data they had to build, validate and implement many different models using a variety of techniques. This process took time, cost money, and yielded little result.
Secondly, the nature of the data scientists’ approach was to use point-in-time historical data as a base for its modelling techniques. As the data scientists got to work, more recent data was pouring in by the minute. Static data modelling is very labour intensive and can take many days, weeks, or even months to produce a working model. Due to the nature of the work, the data used to generate models and predictions could be a month old by the time the data scientists had produced satisfactory results and implemented their model.
ThingWorx Machine Learning
ABC Manufacturing learned of ThingWorx Machine Learning and immediately saw how the AI technology could greatly improve or solve some of the challenges faced by the data science team. The product and implementation teams sat down with ABC to outline a strategic plan centred on ThingWorx Machine Learning, which would be used to automate the modelling, pattern detection and predictive intelligence for the manufacturing processes in an effort to avoid issues related to poor yield.
ThingWorx Machine Learning is designed to automate advanced and predictive analytics. The technology uses proprietary artificial intelligence and machine learning technology to automatically learn from data, discover patterns, build validated predictive models, and send information to virtually any type of application or technology. ThingWorx Machine Learning is built to create intelligent systems by tightly integrating into applications, processes, and technologies already in place.
The team successfully implemented ThingWorx Machine Learning as an end-to-end, automated solution for ABC Manufacturing. Within the first day, the system found that when the facilities experienced drastic swings in humidity levels from day to day, and certain machines operated at a temperature above 90°C, they were more likely to fail.
ThingWorx Machine Learning flagged these machines on days with humidity issues and when the machines ran at a higher temperature, and workers reacted accordingly by either increasing dehumidifying processes at the facility or only running processes during cooler evening temperatures. In a more complex example, ThingWorx Machine Learning was able to detect what was causing unique yield faults within a manufacturing process of 1000+ steps using its Profiling technology. Profiling allows ThingWorx Machine Learning not only to deliver fault detections, but it provided specific conditions that explained failure and yield fault patterns. For instance, ThingWorx Machine Learning was able to analyse billions of points of information to determine that a particular failure pattern occurred when product line 2 was running operation 4467ANX when the ambient temperature was between 25°C and 27°C.
ThingWorx Machine Learning was implemented to convert overwhelming manufacturing data into clear, actionable patterns that are constantly monitored to detect and improve the overall business efficiency and quality. ABC Manufacturing can now better detect, avoid, and manage potential inhibitors to successful production runs within its complex discrete manufacturing processes.
For more information contact Duan Gauche, 1Worx, +27 (0)12 654 0056, duan@1worx.co, www.1worx.co
© Technews Publishing (Pty) Ltd | All Rights Reserved
printer friendly version