A Proposal to Use Artificial Neural Networks for Process Control of Punching/Blanking Operations

Transcription

1 556 Flexible Automation and Intelligent Manufacturing, FAIM2004, Toronto, Canada A Proposal to Use Artificial Neural Networks for Process Control of Punching/Blanking Operations Jannes Slomp 1 and Warse Klingenberg Production Systems Design Group, Faculty of Management and Organization University of Groningen P.O. Box 800, 9700 AV Groningen, The Netherlands ABSTRACT Punching/blanking is among the most important sheet metal manufacturing processes in mass production of metal parts and components in a great variety of industries. In recent years, a further understanding of the technological aspects of the punching/blanking process has been gained, offering the opportunity to integrate the new insights in the process control of the manufacturing process. This paper indicates the new insights in the punching/blanking process and shows how these findings can be used in a process control concept. The application of an artificial neural network plays an important role in this concept. 1. INTRODUCTION During the past century, mechanized and automated industries have created wealth by mass-production of industrial and consumer goods at affordable prices. During the past decades, product diversity has drastically increased and product life cycles decreased, while competitive forces have forced companies to carry the lowest possible levels of raw material, work in progress and finished goods stock. This places high demands on manufacturing organizations in terms of flexibility, process quality and cost minimization. In a global competitive environment, where labor rates differ significantly between regions, high-tech manufacturing industries attempt to utilize their technological assets by drastically improving the organization, planning, control and ultimately the productivity of their operations, often through knowledge-intensive solutions. Further research of process planning and in-process control play an increasingly important role in the quest for low cost flexible manufacturing operations. In recent years, the first tangible concepts, which could serve as bases for in-process control solutions for sheet metal operations, started to emerge, combining an understanding of the principles of in-process control of manufacturing processes and the technological aspects of the process in question [1-4]. For manufacturing processes outside of sheet metal forming, for example turning, on-line monitoring of tool wear and in-process control principles were reported by other researchers [5,6]. Punching/blanking is among the most important sheet metal manufacturing processes in mass production of metal parts and components in a great variety of industries. Punching and blanking are probably as old as any sheet metal forming operation, which is said to date back to at least 250 BC. However, the operation is still not fully understood or captured by any comprehensive process model, although significant progress was made, through experimental modeling [2,7,8] analytical modeling [1,2,9-13] and numerical (Finite Element) modeling [1,2,14-17] of the punching/blanking process. These models, however, may still not be sufficiently adequate to accurately predict the process in all circumstances. Therefore, researchers have started to search for methods to monitor and control some key aspects of the punching/blanking process in real time. Some of these research efforts focus on Artificial Neural Networks (ANN), for example by training an ANN with input from the results of a large number of Finite Element simulations [16]. Others propose to use pattern recognition techniques to monitor and control tool wear, for example from measurements of the peak force [18]. The latter can be categorized as a Statistical Process Control (SPC) solution: a procedure that uses statistical methods to analyze the measured results of a process and uses the results of the analysis to control and manage it [19]. Based on statistical information, SPC aims to predict the probability of an 1 Corresponding author: Jannes Slomp, Fax: (+31) ; j.slomp@bdk.rug.nl

2 A Proposal to Use Artificial Neural Networks for Process Control of Punching/Blanking Operations 557 event in the future, thus allowing adjustments to be made to the process that will prevent that event from occurring if it is not desirable. Information from a process is acquired and analyzed using certain limits, trend data or other values. When a parameter reaches a limit, preventive action is advised. This paper presents new insights in technological aspects of the punching process. These insights offer the possibility to develop an on-line process control mechanism for timely punching tool replacement. The new insights in the punching process are gained from analyzing the shape of the force-displacement graph. This is presented in Section 2. Section 3 presents shortly a concept for process control of the punching/blanking process. Section 4 gives the main conclusions of this paper and indicates future research needs. 2. FORCE-DISPLACEMENT GRAPHS In this section, some recent findings of an analysis of the punching process are presented and it is indicated why these findings are important for the establishment of a process control concept. 2.1 OVERVIEW The stages of the punching process can be recognized from the force-displacement graph. At macro-level the process can be divided into five regions as shown in Figure 1. This figure is recognized as a typical forcedisplacement curve [7] for the conventional punching/blanking process and is typical of the graphs found during the present work Plastic Bending F max Shear Deformation 20 Force (kn) Elastic Bending Ductile Fracture 5 Friction Displacement (mm) Figure 1: Typical Force-Displacement graph for punching/blanking At macro-level, the stages in the process are [2]: 1. Elastic deformation. The blank material is bent and drawn into the hole. The graph shows a corresponding straight line. 2. Plastic bending. The yield strength of the blank material is reached, first at the outer fibers and later at all of the fibers in the zone between the tips of the punch and the die. The materials underneath the punch is subject to thinning [2]. 3. Shear deformation causes the contact area between the product and the blank to diminish, causing a gradual decrease in the force. During this stage, or possibly as early as during the plastic deformation stage, damage initiation followed by the nucleation and growth of cracks takes places. 4. Ductile shear fracture. In most of the conventional punching/blanking situations, ductile fracture occurs after shear deformation. The result is a rough, dimpled-rupture morphology on the fractured surface of the product

3 558 Flexible Automation and Intelligent Manufacturing, FAIM2004, Toronto, Canada in a (usually) conical plane, i.e. at an angle from the smooth sheared surface. 5. Friction. The work done due to friction is dissipated when forcing (pushing) the slug through the die hole INTRODUCTION OF A TOOL TIP RADIUS Typical tool wear is shown in Figure 2. The tool wear creating most of the dimensional inaccuracies is due to the breakdown of the corners of the tool tips, which increases the burr height. As mentioned, other researchers [8,15] have demonstrated that the geometry of the tool after tool wear may be approximated by a rounded shape, creating a punch tip radius. Side wear due to burnishing of blank Corner breakdown creating burrs Figure 2: Typical tool wear It seems reasonable to suggest that the introduction of a radius on the tip of the tool causes additional plastic deformation of the blank, since blank material is forced to obey the shape of the tool and flows around the radius of the punch. The radius at the tip of the punch therefore causes a delay in the start of shear cutting, extending the amount of plastic deformation of the blank. The effect of the radius is to reduce the stress concentration in both the tool and the material. Figure 3 shows the Force-Displacement curves for the punching/blanking process using cold rolled steel with the initial blank thickness H 0 = 1.6 mm. The material characteristics of this cold rolled steel are: proof stress σ 0.2 = 209 MPa, ultimate tensile strength σ B = 304 MPa, elongation at break e break = 36%, work hardening factor C = 489 MPa, work hardening exponent n = Radius = 0 mm Radius = 0.1 mm Radius = 0.22 mm Radius = 0.30 mm Radius = 0.39 mm 25 Blanking Force (kn) R = 0.39 mm R = 0 mm R = 0.30 mm R = 0.22 mm R = 0.1 mm Punch Penetration (mm) Figure 3: Force-Displacement curves for increasing punch radii (Cold rolled sheet steel, H 0 = 1.6 mm, clearance = 5%, D p = 9.84 mm)

4 A Proposal to Use Artificial Neural Networks for Process Control of Punching/Blanking Operations 559 The measurements in this section were taken as part of a comprehensive investigation into characteristics of the punching/blanking process [2]. For these measurements, a C-frame press was used with a capacity of 60 tonnes. The press was fitted with a four pillar die set, configured for punching, as well as a load cell and a displacement transducer. Measurements were processed through National Instruments hardware and LabVIEW software. A further detailed description of the experimental procedure is presented in [2]. Figure 3 shows, as expected [18], an increase in the maximum punching force as a result of an increased punch radius. It also shows a consistently increasing delay in the start of shear cutting, represented by the peak force. It is revealed that the final rupture of the blank, indicated by the sudden drop of the punching force, is suppressed by the introduction of a radius. This observation is consistent with an increasing radius. Also the reduction of the punching force during shearing, indicated by the downward slope, is affected by the tool radius. There appear to be several opportunities to monitor the punching/blanking process. The punching force seems to increase with increasing radius. However, the increase is marginal and within 5% of the original force for this particular material. In any measuring environment, it may be difficult to reliably identify this increase as the onset of tool wear. A better way, directly related to the introduction of a tool radius, seems to be to analyze the work done up to the point of maximum force together with the punch penetration at the point of maximum force. It was observed, as presented in Figure 4, that these values increase consistently and significantly with the tool radius, i.e. with tool wear. Figure 4 shows increases in work done to F max of up to 25% with the introduction of a radius of 0.22mm. Figure 5 illustrates that an introduction of a punch radius of 0.22 mm again causes an increase of up to approximately 25% in punch penetration at the point of maximum force. It is noted that the punch penetration prior to the start of shear cutting will also be dependent on the clearance between the tool and the die. These and other phenomena were also investigated and reported in [2] Cold Rolled, Ho = 1.60 mm Work Done up to Fmax (Nm) Punch Tip Radius (mm) Figure 4: Work Done to F max Figure 6 shows measurement of the burr height with increasing punch radius. The trend of the graph, an increase in burr height due to an increase in product radius, is consistent with literature values [7]. It is noted that for this particular material, the increase in the burr height was marginal, probably negligible, with the introduction of a radius of 0.1 mm on the tip of the punch. Only at tool radius values of 0.2 mm and greater does the result become significant in terms of an increased burr height. Figure 4 and Figure 5 show a significant increase in the work done and the punch penetration at the point of maximum force during the stages of elastic and plastic deformation of the blank material in case of tool radius values of 0.2 mm and greater. This increase, derivable from the forcedisplacement graph, can be used as input information to monitor the punching/blanking process. It should be noted that measurements are presented here for a single type of sheet steel. As part of a wider program [2], also measurements were taken using annealed steel with thickness H 0 = 2.95 mm. Note that in this case, if the same punch and die are used, clearance is smaller, since clearance between the punch and the die is usually expressed as a percentage of the sheet thickness. The experiments with the annealed steel produced the same trends. Again, the work done and the punch penetration at the point of maximum force increased significantly with the punch tip radius [2]. Since tests were carried out for only two materials, generalizations of the above findings

5 560 Flexible Automation and Intelligent Manufacturing, FAIM2004, Toronto, Canada 0.50 Cold Rolled, Ho = 1.6 mm Punch Penetration at Fmax/ Ho Punch Tip Radius (mm) Figure 5: Punch penetration at F max Material: Cold Rolled Steel Ho = 1.6 mm Clearance = 5 % 0.60 Burr Height (mm) Punch Tip Radius (mm) Figure 6: Burr Height versus Punch Radius may not be possible from the presented observations. These findings can, however, serve as a justification for further research into the possibilities to monitor tool wear in punching/blanking from ongoing analysis of the shape of the force-displacement graph. The next section presents a process control concept for punching/blanking. 3. A PROCESS CONTROL CONCEPT An automated monitoring and diagnosis system (or process control system) is of ultimate importance in a Computer Integrated Manufacturing (CIM) environment. Such a system enables unmanned production and the manufacturing of high quality products. A well-working process control system can only be developed if useful data can be gathered during the process. This paper has indicated that the Force-Displacement graph offers this useful data. This section will indicate how this data can be applied in an automated process control system. The variety of operations at a punching machine, during the life of a punching tool, complicates the development of an automated monitoring and diagnoses system. A punching machine usually has an integrated tool storage from which sequentially tools are selected for punching/blanking tasks. A tool, during its life, can be used for punching holes in sheets of different materials and thickness. The frequent change of tools and the variety of sheets create a

6 A Proposal to Use Artificial Neural Networks for Process Control of Punching/Blanking Operations 561 particular complexity for the realization of an automated monitoring and diagnoses system. The process control concept proposed here deals with this complexity and consists of two major steps, which have to be performed after each punching/blanking operation. In the first step, the control system has to determine the punching/blanking characteristics (i.e. punch, die and material characteristics) involved in the operation. These characteristics can be gained directly from the NCprogram. Another source of information may be the Force-Displacement graph itself. The maximum punching force can be seen as a good and stable indicator of the punching/blanking characteristics. This maximum force hardly changes in case of a changing punch radius (see Figure 3). Kalpakjian [20] gives a simple formula to estimate the maximum punch force: F=0.7TL(UTS), where T is sheet thickness, L is the total length sheared (perimeter of hole), and UTS is the ultimate tensile strength of the material. The application of this formula among all tools and materials/sheets, which are in use for the machine, may give a first indication whether or not an automated measurement of the maximum punching force enables the identification of the applied punching/blanking characteristics. The sole use of the Force-Displacement graph offers the advantage that no complex connection is needed between the process control system and the NC-program. The second step concerns a further analysis of the Force-Displacement graph, which can be generated after each punching operation. As mentioned in the previous section, a shift in the work done and the punch penetration at the point of maximum force, compared with initial values, is likely to give information about the burr height that results from the punching/blanking process. These two steps automatically follow from the research results presented in Section 2. There are several issues that need to be addressed further in this concept. In case of changing punching/blanking characteristics, a particular punch may be used in various situations. This raises the question how the effect of tool wear, gained on previous punching operations, can be estimated in the Force-Displacement graph in case of new punching/blanking characteristics (i.e. new material and/or thickness of sheets). It is important to estimate this effect in order to control the quality of holes in the new situation. A straightforward method to estimate the effect is to use a simple linear function that translates the percentage of previous change of the shift in work done and the punch penetration at the point of maximum force to the new situation. Research is needed to investigate the empirical presence of such a simple function. Another issue concerns the stability of the Force-Displacement graph for particular punching/blanking characteristics. It is likely that machine and material characteristics are not stable and all kind of error patterns may exist with respect to the measurement of the work done and the punch penetration at the point of maximum force. Back propagation artificial neural networks can be helpful to monitor these error patterns in an automated way [21-24]. In case of a significant shift in the mean of work done and punch penetration at the point of maximum force, the process control system may ask for refurbishment of the punching tool. The application of artificial neural networks for monitoring and diagnoses is suggested in several papers and can also be applied in the process control of punching/blanking processes. In a survey study on the application of neural networks in manufacturing [21], monitoring is seen as one of the major application areas of neural nets. The use of neural nets for control chart pattern recognition is described in several recent papers [22-24]. These papers show that the use of artificial neural networks for pattern recognition tasks has made great strides. In the particular case of monitoring and diagnoses of punching/blanking processes, artificial neural networks can be useful in both steps of the process control framework described in this section. Neural networks can be used to detect the punching/blanking characteristics from the Force-Displacement graph. They can also be used to recognize patterns in the values of certain variables in the Force-Displacement graph. The discovery of these patterns is essential in the control of the punching/blanking process. 4. CONCLUSIONS This paper has presented some new insights in technological aspects of the punching/blanking process. It is shown that an analysis of the Force-Displacement graph can be linked with product quality (i.e. burr height). Important indicators are the work done and punch penetration at the point of maximum force. An on-line measurement of these values supports the possibility of an early and automated detection of quality problems. This paper indicates that artificial neural networks can be helpful to realize an automated monitoring and diagnoses system. A concept for an autonomous process control system for punching/blanking operations is suggested in this paper. This can be seen as a useful step towards the integration of punching/blanking operations in Computer Integrated Manufacturing environments.

7 562 Flexible Automation and Intelligent Manufacturing, FAIM2004, Toronto, Canada REFERENCES [1] W. Klingenberg and U.P. Singh: Finite Element simulation of the punching/blanking process using in-process characterisation of mild steel, Journal of Materials Processing Technology Vol.134, No.3, pp , 2003 [2] W. Klingenberg, Numerical modelling of the punching/blanking process using in-process characterisation of steel, DPhil. dissertation, University of Ulster, [3] K.A. Siribor, An analytical model for pressbrake forming using in-process identification of aerospace material characteristics, DPhil. dissertation, University of Ulster, [4] U.P. Singh, W. Klingenberg, and W. Urquhart: Effect of tool geometry on punching performance, Transactions of the ASME Journal of Engineering for Industry, Vol.116, pp , [5] S.K. Choudhury, V.K. Jain and C.V.V.R. Rao: On-line monitoring of tool wear in turning using a neural network, International Journal of Machine Tools & Manufacture, Vol.39, No3, pp , [6] T. Yandayan and M. Burdekin: In-process dimensional measurements and control of workpiece accuracy, International Journal of Machine Tools & Manufacture, Vol. 37, No.10, pp , [7] K. Lange, Handbook of Metal Forming, McGraw-Hill, New York, [8] C.M. Choy and R. Balendr: Experimental analysis of parameters influencing sheared-edge profiles, Proc. 4 th Int. Conf. on Sheet Metal, University of Twente, The Netherlands, Vol. II, pp , [9] J.A.H. Ramaekers, Härte und Verformung Metallischer Werkstoffe, Ph.D. Dissertation, University of Eindhoven, The Netherlands, [10] J.A.H. Ramaekers and J.A.G. Kals: Strain, stresses and forces in blanking, Proc. IMC Conference Galway, 1986, pp , [11] A. Ghosh, V. Raghu Ram and P.B.Popat: A new approach to the mechanics of the Blanking operation: Theoretical model and Experimental Verification, Journal of Mechanical Working Technology, Vol.11, pp , [12] Q. Zhou and T. Wierzbick: A tension model of blanking and tearing of ductile metal plates, International Journal of Mechanical Sciences, Vol.38, No.3, pp , [13] A.G. Atkins: On cropping and related processes, International Journal of Mechanical Sciences, Vol.22, pp , [14] D.C. Ko and B.M. Kim: Development of an analytical scheme to predict the need for tool regrinding in shearing processes, International Journal of Machine Tools & Manufacture, Vol.40, pp , [15] R. Hambli: Blanking tool wear modelling using the Finite Element method, International Journal of Machine Tools & Manufacture, Vol.41, No.12, pp , [16] R. Hambli: Prediction of burr height formation in blanking processes using neural network, International Journal of Mechanical Sciences, Vol.44, pp , [17] Th. Pyttel, R. John and M. Hoogen: A finite element based model for the description of aluminium sheet blanking, International Journal of Machine Tools & Manufacture, Vol.40, pp , [18] W.B. Lee, C.F. Cheung, W.M. Chiu and L.K. Chan: Automatic supervision of blanking tool wear using pattern recognition analysis, International Journal of Machine Tools & Manufacture, Vol. 37, No.8, pp , [19] R. Burman, Manufacturing Management Principles and Systems, McGraw-Hill, [20] S. Kalpakjian, Manufacturing Engineering and Technology, Addison Wesley, Third Edition, [21] H.-C. Zhang and S.H. Huang: Applications of neural networks in manufacturing: a state-of-the-art survey", International Journal of Production Research, Vol.33, No.3, pp , [22] T-H Hou, W-L Li and L Lin: Intelligent remote monitoring and diagnosis of manufacturing processes using an integrated approach of neural networks and rough sets, Journal of Intelligent Manufacturing, Vol.14, pp , [23] M.B. Perry, J.K. Spoerre and T. Velasco, Control chart pattern recognition using back propagation artificial neural networks, International Journal of Production Research, Vol.39, No.15, pp , [24] C.W. Zobel, D.F. Cook and Q.J. Nottingha: An augmented neural network classification approach to detecting mean shifts in correlated manufacturing process parameters, International Journal of Production Research, Vol.42, No.4, pp , 2004.