Van Gilder, John F. (1989, ASQC) AC-Rochester Division of General Motors Corp., Flint, MI
Design of experiments is a key tool for adapting current processes to produce improved products. Design of experiments techniques provide an efficient and systematic method for developing the detailed understanding of a process required to identify what changes, if any, in hardware, material inputs, and operating procedure are needed to implement continuous improvement in both product and process performance.
The use of design of experiments in adapting an existing process to manufacture a component which has been modified to improve durability is detailed in this paper. The component is the Exhaust Exygen Sensor and is part of the exhaust emission/engine control system in General Motors passenger vehicles and light duty trucks. For the 1988 model year, the amount of protective ceramic coating applied to the sensing element was doubled to improve sensor life and to provide a more robust design against environmental contamination. This protective coating is applied to the sensor using a flame (thermal) spray process.
The flame spray process is complicated with over a dozen major process controls available to the operator to adjust. Due to the complexity of the process, design of experiments techniques were used to understand in detail how the process functioned, to determine the optimum process settings, and to establish the control limits for the critical process parameters in order to apply the increased protective coating.
This paper also uses one of the four factor full factorial designs conducted on this highly interactive process to demonstrate by example how Taguchi experimental designs (highly fractionalized two level factorials) can yield incorrect and misleading results. This is done by designing an experiment based on the L8 orthogonal array and selecting from the full factorial experiment those eight tests required by the L8 array for the analysis. A comparison is then made between the results that would have been obtained from the Taguchi design with those actually obtained from the full factorial test. This comparison underscores the inappropriateness of using Taguchi methods to investigate interactive processes.