MetaCause have submitted a paper in collaboration with our friends at Doncasters Precision Castings (Deritend) in preparation for the ICI conference in Dallas (USA) this October.
For more information, please contact us
Abstract
Process optimisation is a fundamental issue in the future development of the cast metals industry as it continues to develop its people and processes to provide quality solutions for its customers.
The way in which foundries record and utilise data is set to change dramatically as we move away from traditional recording and people development techniques and move towards self-learning software that embraces existing skills, captures abilities and passes these on to the next generation of cast metals engineers.
Process data typically comprises of values for various parameters that are routinely monitored in a process along with discrete information such as suppliers, batch numbers, machines used and operators. Rejection data can be castings, shells or patterns rejected, accepted or marked for rework. Some foundries also scale the severity of defects on the scale of zero to ten.
For an investment casting foundry, the number of variables that need to be assessed for solving a problem generally exceed 15-20 and experts have traditionally used intuition or heuristic methods to focus on three to four key variables as possible causes.
Chi-square statistics, Design of Experiments, Scatter Diagrams, Correlation techniques, Bayesian probability and Decision trees are among the techniques that are more commonly used by Six Sigma experts. Receiver Operating Characteristics (ROC) techniques � predominantly used in the medical domain - are for strange reasons less favoured by Six Sigma experts within the manufacturing community.
This paper will analyse various options available for Six Sigma experts to evolve an evidence based approach for problem solving.
We have undertaken over 10 years of research and our observation is that none of the techniques mentioned above are designed to analyse production data to discover process optimisation opportunities. A process with 5-15% rejection level is also producing 95-85% good quality castings. The challenge is in separating noise from a signal. This is where various techniques tend to disagree. All techniques generally agree on extreme cases.
MetaCause process optimizer software has an ability to suggest optimal process settings, process settings to avoid and process settings with no effect with an appropriate importance weighting value for each setting.
This presentation will explain how to take information collection one step further by analysing historical data and recommending process optimisation opportunities and to develop an environment so that existing production knowledge and expertise is passed from the individual to the company; a must for long-term progression.
This paper will extend our previous work that has been published at various investment casting conferences since the last world investment casting congress in Edinburgh. The originality of our approach over existing statistical techniques will also be presented along with few case studies.
