“How Learning by Doing Is Done: Problem Identification in Novel Process Equipment”, 1995 ():
The unit cost of producing manufactured goods has been shown to decline substantially as more are produced. It has been argued that ‘learning by doing’ [experience curve] is at the root of this phenomenon, but the modes of learning actually involved have not been studied in detail.
In this paper we attempt to provide a better understanding of the learning behaviors involved in learning by doing via a study of 27 problems that affected two novel process machines in their first years of use in production. [Computer chip manufacturing] …We elected to focus our study on the early field use of two types of process machine, a solder paste profiler and a component placer. These machines were developed to automate manual procedures previously used to attach surface-mounted integrated circuits to large, complex circuit boards.
First, ‘interference finding’, is described, a form of learning by doing that appears to be central to the discovery of the problems studied. Next, the reasons why the problems identified by templating were not discovered prior to field use—before ‘doing’—are explored. Two causes are identified: an inability to identify existing problem-related information in the midst of complexity, and the introduction of new problem-related information by users and other problem solvers who learn by doing after field introduction of the machine.
We find that problems due to information lost in complexity emerge earlier than do problems due to user learning by doing. Tests of reason are used to show why it would be very difficult to eliminate doing from learning by doing.
Finally, other implications of the study findings are discussed.
…Example: yellow circuit board problem: The component-placing machine uses a small vision system incorporating a TV camera to locate specific metalized patterns on the surface of each circuit board being processed. To function, the system must be able to ‘see’ these metalized patterns clearly against the background color of the board surface itself.
The vision system developed by the machine development group functioned properly in the lab when tested with sample boards from the user plant. However, when it was introduced into the factory, users found that it sometimes failed, and called this to the attention of the machine developers. The development engineers came to the field to investigate, and found that the failures were occurring when boards that were light yellow in color were being processed.
The fact that boards being processed were sometimes light yellow was a surprise to lab personnel. While factory personnel knew that the boards they processed varied in color, they had not volunteered the information to the lab because they did not know that the designers would be interested. Early in the machine development process, factory personnel had simply provided samples of boards used in the factory to the lab. And, as it happened, these samples were green in color. On the basis of the samples, developers had then (implicitly) assumed that all boards processed in the field were green.
…Example [tilted heat sinks]: After the component placing machine was installed in the field, users noticed that it was unable to pick up parts that had ‘tilted’ heat sinks on top. This problem was a surprise to developers. They had not known that such parts existed, and had not designed the machine to handle them.
…in each of the instances in this category, interviewees told us that the information could easily have been provided to the lab, had the developers thought to ask and/or had users thought to volunteer it. But, the relevance of the information was overlooked until it was made clear by templating during use of the machine in the field.
…Example: component slippage problem: Just before the component placing machine places components on a board, little dabs of solder-containing paste are applied to the board, one at each spot where an electrical connection is to be made between a component leg (a wire protruding from the base of the component) and the board. The machine designers knew about this, but chose to use adhesive tape instead of solder in their laboratory simulation of the use environment. (Use of solder would have required setting up the lab to comply with rules regarding the handling of hazardous materials, a costly matter.)
When the component placer was installed in the field, users noticed that components unexpectedly slipped sideways to an unacceptable degree when the robot arm was pressing them onto the board. Investigation showed that the mound-shaped dabs of solder paste were firm enough to push the component sideways if the legs touched down on their sides instead of directly on their tops. This effect did not occur in the lab because the lab had not used solder in its tests.
…Example: location adjustment problem: Each time a new board design was processed by the component placing machine, operators had to tell the machine where to put each of the components to be placed on the new board. They did this by entering the x and y coordinates of each part location in the machine’s computer memory. In case these coordinates required later adjustment, operators and machine designers both assumed that the operators would re-enter new x and y coordinates.
After the machine was installed in the plant, users discovered that they had to adjust x and y coordinates very frequently. They also found that it was very cumbersome to do this by re-entering new coordinates. Instead, they learned to make the needed adjustments via an obscure ’move it over by x amount’ command that was buried several layers down in a software menu on the machine’s control panel. The problem that users then brought to the attention of machine designers was: The ’move it over by x amount command’ is very hard to reach and use. Make a more convenient one!