Variation management in product development and manufacturing processes.

Belgiu, George ; Albu, Adriana ; Ruset, Vasile 等

1. INTRODUCTION

Design for Manufacture (DFM) has been in general use in consumer products industries for several years. General benefits are: decrease of product development cost and increase in product quality, competitiveness and shortening product development cycles (***, 2009a).

In the manufacturing process there are many sources of variation. In this paper we are concentrate where variation comes from in the machining processes, and how we can reduce this natural influence. Most of the manufacturing processes are automated, but even in this case, there are numerous sources of variation that can conduct to scrap or rework.

[FIGURE 1 OMITTED]

The research and development engineer have to analyze the current product designs or new designs in development and promptly make a decision how to simplify the design for substantial cost savings. Usually, the engineer will explore a large group of manufacturing circumstances: diverse material and process combination, diverse geometric shapes for the 3D model, various scenarios and so on. Figure 1 shows the DFM's position in the product research and development cycle (***, 2009b). In the product development subject, all studies were conducted to several directions: concept development, system-level design, detail design, testing and refinement or production ramp-up (Kamalini et al., 2003). Three directions are used to implement DFM in a firm: (i) cross-functional teams in the company; (ii) design rules according with the company strategy; (iii) specialized software tools for DFM. In reality, DFM strategy is dependent on product and production strategy. However, DFM is not considering the sources of variation.

2. VARIATION MANAGEMENT IN PRODUCT DEVELOPMENT AND MANUFACTURING PROCESSES

If we can control (or predict) the variations before the manufacturing process starts, then the product design and development are appropriate through manufacturing (figure 2). Process variation influence on numerous production processes. The most important are the quality of the parts that are produced, which directly impact on costs. Nevertheless, where products pass the quality control, then other parameters are prejudiced: efficiency, cycle time, utilization, and lead time. For a company, these instruments can be calculated as productivity rate (***, 2007). The manufacturing process output can be controlled by addressing the cause: managing variation in the inputs to the process.

[FIGURE 2 OMITTED]

3. THE RESEARCH COURSE. THE AI SYSTEM

For a specific manufacturing process (i.e. metal cutting process--turning, milling or drilling process) we considered the sources of variations from figure 2. The parameters [S.sub.1], [S.sub.2], [S.sub.6] can be found on figure 3 (***, 2007).

[FIGURE 3 OMITTED]

The variation management module (VMM) is a tool that is able to predict if the final product will be good or not. According to this prediction, the product design can be changed. Again, the sources of variations depend on the manufacturing system type. Generally, all manufacturing's input parameters are (***, 2007):

* [A.sub.1]--lack of training, [A.sub.2]--lack of instructions, [A.sub.3]--lack of procedures, [A.sub.4]--discipline, [A.sub.5]--data entry errors, [A.sub.6] man's setup errors, [A.sub.7]--man's measurement errors for man source of variation;

* [B.sub.1]--accuracy, [B.sub.2]--repeatability, [B.sub.3]--squareness, [B.sub.4] calibration--geometry, [B.sub.5]--rigidity, [B.sub.6]--vibration, [B.sub.7] lack of maintenance, [B.sub.8]--thermal conditions for machine source of variation;

* [C.sub.1]--hardness, [C.sub.2]--stress relief, [C.sub.3]--clamp condition, [C.sub.4] material condition, [C.sub.5]--deflection and distortion, [C.sub.6]--hard spots, [C.sub.7]--material relaxation for material source of variation;

* [D.sub.1]--fixturing, [D.sub.2]--sequence of operations, [D.sub.3]--delays in process, [D.sub.4]--other processes or operations, [D.sub.5]--tooling, [D.sub.6]--thermal compensations, [D.sub.7]--process set up, [D.sub.8]--tool breakage for technology source variation;

* [E.sub.1]--gauge condition, [E.sub.2]--equipment condition, [E.sub.3] measuring error, [E.sub.4]--device error, [E.sub.5]--contamination, [E.sub.6] temperature, [E.sub.7]--soak time, [E.sub.8]--part relaxation for control source of variation;

* [F.sub.1]--ambient temperature, [F.sub.2]--part temperature, [F.sub.3] temperature swings, [F.sub.4]--swarf, [F.sub.5]--coolant, [F.sub.6]--humidity, [F.sub.7]--damage on handling for environment source of variation.

The manufacturing system's output parameters are divided in two categories: (i) standard parameters, [G.sub.1], ..., [G.sub.5]--available for general cases (surface quality, functionality, maintenance, productivity, efficiency); (ii) custom parameters, [G.sub.6], ..., [G.sub.10]--specified by the product designer.

In order to implement this module, artificial neural networks (ANN) were used. These are a branch of artificial intelligence (AI) domain and have been developed to reproduce human reasoning and intelligence (***, 2009d).

Artificial neural networks can be successfully used anywhere there are problems of prediction or classification. This is the reason why the VMM was implemented using ANNs. The products must be classified in good parts and scarps and this classification is made according to some characteristics of the product ([G.sub.1], ..., [G.sub.10]) which are predicted by the neural network.

To create an artificial neural network is necessary to put together a number of neurons. These are processing units with synaptic input connections and a single output (Zurada, 1992) which is calculated with the equation 1.

y = f(x) (1)

In the equation 1 f is a transfer function (step function in this case) and x is determined by the equation 2, where W is the weights vector and P is the inputs vector.

x = [summation.sup.n.sub.i=1] [W.sub.i] [P.sub.i] (2)

The neurons are arranged in layers. A network has to have an input layer (in this application it is made by the parameters which describe the sources of variation, [A.sub.i], [B.sub.i], [C.sub.i], [D.sub.i], [E.sub.i], [F.sub.i]) and an output layer (the predictions, which here are the qualities of the final product, [G.sub.i]). There also can be hidden layer(s) of neurons that play an internal role in the network.

The training of ANN is realized modifying the weights of the connections between neurons and has two phases: an initial phase when the parameters receive their initial values and a second phase, which is iterative, when the parameters are adjusted. The quality of an artificial neural network depends not only by the way of modifying its weights, but also by their initial values. This is the reason why, in order to choose the best neural network, there were created and trained 500 ANNs. The network with the best accuracy is than used to make the predictions regarding the qualities of a product.

This VMM was implemented only for cylindrical parts, manufactured by turning and/or grinding processes. To complete the VSM functionality and to make it powerful software for the product designer, future work will be training the neural network for other categories of parts (mainly prismatic parts and free form surfaces parts).

4. CONCLUSION

Design for Manufacture is a significant component of the productive processes. For a company, developing an optimum DFM is a foundation for a stable operating environment. But this is only a half of the way to control and simplify the manufacturing process.

To achieve the best results in Product Research and Development Process, a company must integrate:

* DFM strategy;

* sources of variation management (VM) strategy.

The best approach to integrate DFM strategy and VM strategy in a design system is to create two software application tools. From practical considerations, as DFM strategy we have used a commercial software tool DFMPro[R] from Geometric[TM], under the CAD platform SolidWorks[R] 2009 (***, 2009c). As a VM strategy, we created a custom expert system--also a plug-in for SolidWorks[R].

5. REFERENCES

Kamalini, R.; Fisher, M. & Ulrich, K.T. (2003). Managing Variety for Assembled Products: Modeling Component Systems Sharing. Manufacturing and Service Operation Management, vol. 5, no. 2, pp. 142-156, ISSN 1523-4614

Zurada, J.M. (1992). Introduction to Neural Systems, West Publishing Company, New York

*** (2007). http://www.renishaw.com/--Pyramid training sources of variation v2.00. Renishaw apply innovation[TM] (PowerPoint presentation)

*** (2009a). http://www.design-iv.com/dfm.htm--Boothroyd Dewhurst's DFM Concurrent Costing Software, Accesed on:2009-05-22

*** (2009b). http://www.ulrich-eppinger.net/--Product Design and Development. A Resource for Students and Professionals in the Field of Product Design and Development, Accesed on:2009-06-10

*** (2009c). http://dfmpro.geometricglobal.com/--Iterative Design Process, Accesed on:2009-03-07

*** (2009d). http://www.statsoft.com/textbook/ Stathome.html--The Statistics Homepage: Neural Networks, Accesed on: 2009-10-07