Vol.3 No.1 2010

85/110

Research paper : Acquisition of skills on the shop-floor (N. Matsuki)−82−Synthesiology - English edition Vol.3 No.1 (2010) 4.4 Derivation by data miningIn manufacturing, responding to various troubles is an important skill. In plating, we built a mechanism for organizing the case studies of defects, to collect and to accumulate the relationships between the candidate causes of defects and the correctional procedures from the past data.The troubles are categorized, and when a trouble actually happens, it is determined to which category a defect belongs. The past data are searched based on the situation where the defect occurred as additional parameter, and countermeasures are taken. This is a type of data mining. This is a method for obtaining the appropriate information from the past storehouse of data, for cases where the theoretical or experimental formulae cannot be estimated. Cluster analysis, neural network, and genetic algorithms may be effective, although these were not conducted in this research. In casting, we use a search method with an interface as shown in Fig. 6 to search the past case studies. This is a method called Eagle Search developed at the Center, and is characterized by the flexibly interchangeable search keys.The fact that the derived model is based on data mining indicates that the derived model under investigation is not sufficiently clarified. If the cause can be theoretically pursued from the troubles that occur and the response can be clarified, a theoretical derived model can be built, rather than using data mining. However, in actual situations, theoretical pursuit may not be possible because the defect is not reproducible or the cost of reproduction is too great. Therefore, data mining where the past case studies and their solutions are accumulated and searched is effective as a skill for responding to trouble.The three derived models were presented above. Comparing them, the method based on theoretical formula seems to be the most effective and reliable. That is because the derivative method can be explained theoretically. The method by experimental formula seems to be reliable following the theoretical formula method since the assumed parameters and their mutual relationships are clear. The method of data mining seems to be the least reliable since the cause-and-effect relationship is unknown. Therefore, the derived models should be reviewed constantly, and a method based on a theoretical formula must be investigated as much as possible.However, the differences among these methods are not necessarily clear. The difference between the derivations by theoretical and experimental formulae is the difference of whether the theory is established or not. For data mining, as the range of the issues become clear and the data are accumulated, the estimation of an experimental formula may become possible. When using this research result, the derived model should be reviewed according to the changes in the situation and the method with higher reliability should be explored further.4.5 Other derivative methodsFor the decision-making skills, we believe the above three methods are typical. However, there are skills other than decision-making that must be transferred. In metal stamping that was studied by RIKEN, the processing methods are different according to the types of parts under investigation. This means that the most important skill is the skill of selecting the processing method according to the characteristic of the parts, and the skill of discerning the points that must be focused on in selecting the processing method. The selection of the processing method design is too complex to be called a simple chain of decisions. In this project, we worked on skill transfer by creating a meta-flow 20100200400520mm 13060N5 TPExperimental variables・ Shape of treated product: Round bar 20φ*100, 200, 400, 520 mm・ Set position: Longitudinal, transverse・ ~ Quenching oil temperature: 130 °C (hot), 60 °C (cold)・ Oil steering speed: High speed, low speed・ ~ N: 5・ General condition used for TP material and carburizing(a)Dependence on TP length of bentness by longitudinal and transverse placement(b)Dependence on TP length of bentness by longitudinal and transverse placementTP length, mmBentness value, mmRed: TransverseBlue: Longitudinal01002003004005006001.00.03.02.05.04.07.08.06.001002003004005006001.00.03.02.05.04.07.08.06.0Oil temperature hot, low speed stirringOil temperature hot, high speed stirringRed: TransverseBlue: LongitudinalTP length, mmBentness value, mmFig. 5 Example of derivation of experimental formula in heat treatment.

※このページを正しく表示するにはFlashPlayer9以上が必要です