Tricky product problems are often sticky and chewy like chewing gum: Whatever you do, you can not get rid of them. In practice, they survive because the method of trial and error has repeatedly failed, both in product engineering and in production. The good, highly significant news: Using Statistical Engineering, tricky product problems can be solved quickly and sustainably.
In my professional life, I solved more than 100 tricky quality problems that had survived conventional problem solving of all kinds, mostly trial and error. For example, an automaker had problems with the timing belt noise of an internal combustion engine. Trial and error in product development, ramp-up, and ongoing production did not improve anything. The problem remained very, very expensive, both in the field and in the factory. Using Statistical Engineering, I was able to uncover the root cause and eliminate the failures sustainably within a few weeks.
Already with its first two steps, planning and design, Statistical Engineering enforces the uncovering of the root cause, also from the perspective of third parties. Only the uncovered root cause allows to find the best problem solution in terms of function, speed, quality, costs, and risks.
The limitations of trial and error and the associated waste are not always so obvious. Augustine, CEO of several aerospace companies, noted that the cost of testing low-cost products is immensely high compared to expensive products. Low costs of prototypes seduce to an approach of multiple trials with subsequent errors. This is usually a high and concealed waste.
For the quick and sustainable solution of tricky product problems Statistical Engineering is often more efficient and cheaper than machine learning. Statistical Engineering is applicable even if prior knowledge and data are missing. It simultaneously accesses all principles of human learning and accelerates it. In contrast, machine learning requires existing data, ignores causality, and applies only one of five learning principles at a time, as the computer scientist Pedro Domingos describes in “The Master Algorithm.”
The effort for Statistical Engineering is low. It includes practical learning based on specific problems with the help of textbook, training or advice. On the other hand, machine learning always requires one-time expenses and continuous maintenance costs.
You can acquire the methodology of Statistical Engineering in four ways. For example, I give lectures on Statistical Methods each winter semester at RWTH Aachen University. Or you can find them in the reference book Statistical Engineering, available at Amazon. Or I advise you in solving specific tricky quality problems. Or opt for on-the-job in-house training where selected engineers solve several tricky problems that affect your company’s KPI.