|Title of Invention||
A SYSTEM FOR FORMING SHEET PRODUCTS
|Abstract||The invention relates to a system and a process of forming sheets comprising: a sheet deforming means comprising punch, blank holder and die; strain measuring means provided on the said sheet deforming means; data acquisition, means provided with Artificial Neural network (ANN) capabilities, output device configured with the said sheet deforming and strain measuring means; wherein in the process of forming, strain data from the said strain measuring means is fed into the data acquisition and processing means wherein ANN architecture is developed and ANN trained in the said processing means to arrive at a the sheet material specification to produce sheet product/s of desired shapes/geometries, meet user to defined quality parameters, effect of change in tool parameters and processing conditions on quality of the product.|
|Full Text||FORM 2
THE PATENTS ACT, 1970 (39 of 1970)
PROVISIONAL / COMPLETE SPECIFICATION
(See section 10 and rule 13)
1. TITLE OF THE INVENTION:
A PROCESS FOR SELECTION OF MATERIAL SPECIFICATION FOR SHEET METAL
(a) NAME: Indian Institute of Technology, Bombay
(b) NATIONALITY: Indian
ADDRESS: Indian Institute of Technology Bombay, Powai, Mumbai - 400076.
3. PREAMBLE TO THE DESCRITION
The following specification describes invention
4. DESCRIPTION (Description shall start from next page)
5. CLAIMS (not applicable for provisional specification. Claims should start with the preamble - "I/We claim" on separate page)
6. DATE AND SIGNATURE (to be given on the last page of specification)
7. ABSTRACT OF THE INVENTION (to be given along with complete specification on the separate page)
*Repeat boxes in case of more than one entry
*To be signed by the applicant(s)or the authorized registered patent agent
*Name of the applicant should be given in full, family name in the beginning
'Complete address of the applicant should be given stating with postal index no. / code, state and country
*Strike out the column which is/are not applicable
Field of Invention
The invention relates to a system and a process of forming sheets to produce sheet products of desired shapes conforming to user defined acceptance criteria.
Background of the Invention
Large rejections of sheet metal components occur when a sheet is "mis-applied", i.e., when the 'demand' for sheet ductility is far too high compared to what the properties of the raw material can meet. One prescribes a grade of sheet metal suitable for forming at the time of part design and proves the process design for this material. Acceptance criteria defined in terms of certain parameters such as peak strain, springback etc, the permissible variation in the values of which, when restricted within limits prescribed in the product drawing, constitutes quality. Hence they need to be measured and controlled to lie within permissible limits so as to achieve the desired quality of the end product. Such parameter(s) are hereinafter referred to as 'quality parameter(s)'.
Despite having a sheet conforming to standards, one cannot be certain of a good performance of the sheet in the press shop. For a shop engineer therefore, it will be helpful if a tool could tell him as to whether a given coil of material should go into the press shop or not, thereby avoiding repercussions if the sheet does not perform.
Often, merely saying that "a given grade of sheet will not work", might not be enough. One might wonder what specification of the sheet would work with certainty.
The solution to this problem is not addressed and reported explicitly in literature. However, application of Artificial Neural Network (ANN) to metal forming is reported in the literature.
Application of ANN to bending is reported by Inamdar at. e\.[Studies on the prediction of springback in air vee bending of metallic sheets using an artificial neural network, Journal of Materials Processing Technology, Vol. 108, 2000, pp. 45-54.] Springback in air vee bending process is large in the absence of bottoming. Among the various intelligent methods for controlling springback, application of an artificial neural network (ANN) for real time control is disclosed. The reported work describes the development of an ANN based on back propagation (BP) of error.
Manabe et al [Artificial intelligence identification of process parameters and adaptive control system for deep-drawing process, Journal of Materials Processing Technology, Vol. 80-81, 1998, pp. 421-426.] have disclosed an in-process identification method of material properties and lubrication condition in the deep-drawing process of anisotropic sheet metals. The method is based on a combination model of artificial neural network (ANN) and elastoplastic theory.
K. Hans Raj et al [Modeling of manufacturing processes with ANNs for intelligent manufacturing, International Journal of Machine Tools & Manufacture Vol. 40,
2000, pp. 851-868] and there group have disclosed application of ANN to processes such as upsetting and extrusion in metal forming, and machining etc. ANN training for metal-forming (hot upsetting and extrusion) processes and that for metal cutting has been presented. The ANNs were trained with finite element simulation data in the case of metal forming and experimental results in the case of metal cutting. However, the reported work primarily focuses on tools for bulk forming processes, metal cutting and on-line monitoring of tool wear in the manufacturing process. It does not address sheet forming processes, and at any rate, not material specification.
Kim et al. [Application of artificial neural network and Taguchi method to preform design in metal forming considering workability, International Journal of Machine Tools & Manufacture, Vol. 39, 1999, pp. 771-785.] have reported a method of preform design in multi-stage metal forming processes considering workability limited by ductile fracture. The artificial neural network using the Taguchi method has been implemented for minimizing objective functions relevant to the forming process.
D.J. Kim et al [Application of neural network and FEM for metal forming processes, International Journal of Machine Tools & Manufacture, Vol. 40, 2000, pp. 911-925.] have disclosed a technique to apply the artificial neural network in metal forming processes. A three-layer neural network is used and a back propagation algorithm is employed to train the network. The proposed schemes are used to find the initial billet size for axisymmetric rib-web product and to design the die geometry for cylindrical pulley.
R Hambli et al [Application of a neural network for optimum clearance prediction in sheet metal blanking processes, Finite Elements in Analysis and Design, Vol. 39, 2003, pp. 1039-1052.] used ANN for predicting blanking parameters. A methodology to obtain the optimum punch-die clearance for a given sheet material by the simulation of the blanking process is presented.
Ridha Hambli [Ridha Hambli, Prediction of burr height formation in blanking processes using neural network, International Journal of Mechanical Sciences, Vol. 44, 2002, pp. 2089-2102.] has reported combination of predictive finite element approach with neural network modeling of the leading blanking parameters in order to predict the burr height of the parts for a variety of blanking conditions.
Lee et al [A new method of preform design in hot forging by using electric field theory, International Journal of Mechanical Sciences, Vol. 44, 2002, pp. 773-792,] used ANN for obtaining prefom design using ANN.
Ashwini Gokhale [Sheet metal forming in a virtual reality environment using LSDYNA and neural network, 4th European LS-DYNA user conference, 2003, pp.23-33] has reported a virtual reality environment for sheet metal forming process. Generic process models for part families are developed using Artificial Neural Network (ANNs) and FEM. The generic models are used to predict the responses of the manufacturing process to variations in geometric, material and process parameters, in real time.
K. Narasimhan et al [An artificial neural network approach for predicting peak strains developed during sheet metal forming, NUMISHEET 96, 1996, pp. 173-177.] have reported use of ANN to predict the peak strains and maximum punch load developed on a new combination of material and processing conditions.
The work reported by Jun Zhao et al [Parameter identification by neural network for intelligent deep drawing of axisymmetric workpieces, Journal of Materials Processing Technology, Vol. 166, 2005, pp. 387-391] aims at identifying material properties as well as the friction coefficient for axisymmetric workpieces. The work leads to an intelligent control system of the deep drawing process for axisymmetric work pieces. However, the work is restricted to only to axisymmetric work pieces.
However, though, only applications of ANN are reported in the prior art, they do not teach a comprehensive integrated system and process for forming sheets that facilitates a direct selection of sheet material specification with their implications on product quality based on diverse combinations of materials, processing and tool variables without being restricted to any shape. Further, the prior art is restricted to axisymmetric shape of the product and is not capable to handle issues related to a family of non-axisymmetric products.
It has therefore been a long-felt need of industry to provide a system and process for forming sheets that facilitates a direct selection of sheet material specification with their implications on product quality based on small, finite number of diverse combinations of materials, processing and tool variables thereby addressing infinite number of forming conditions without being restricted to any shape and without the use of expensive repetitive simulations to produce sheet metal products of desired shapes.
The present invention satisfies this need and overcomes the limitations of the prior art as the system and process of the present invention provides a direct selection of the materials specification and processing and tool variables based on an ANN developed using finite, small number of training samples and predicts the product quality under infinite number of combinations of materials without any limitation on the shape of the product.
Summary of the Invention
The main object of the present invention is to provide a system and process to operate the same so as to arrive at the sheet material specification to produce formed sheet metal products of desired shapes conforming to user defined acceptance criteria.
Another object of the present invention is to anticipate consequence of changing certain dimensions of the product, the tool variables and the processing conditions on the quality parameter like maximum thickness strain
Thus according to the present invention, the system comprises of:
sheet deforming means such as press comprising, either or both of the die and punch made from solid, liquid or gaseous material; peak strain (or any measurable parameter(s) defined as acceptance criteria in the drawing, i.e., quality parameter(s)) measuring means; data acquisition, processing means and output device configured with the press and strain/ quality parameter measuring means;
wherein in the process of forming, strain data from the said press & die system is fed into the data acquisition and processing means to develop ANN architecture, train artificial neural networks (ANN) and arrive at the sheet material specification to produce sheet products of desired shapes meeting the prescribed acceptance criteria and also enabling at once an assessment of the effect of change in tool parameters and processing conditions on the quality of the product irrespective of material and the shape of the product.
Description of the Invention
Features and advantages of this invention will become apparent in the following detailed description and the preferred embodiments with reference to the accompanying drawings.
Figure 1: Schematic of the system (Sheet 1)
In one of the embodiments, the schematic of the system is shown in Figure 1. The system 1 comprises of sheet deforming means 2 including die and punch, strain measuring means 3, data acquisition means 4, processing means 5 and output device 6. The strain of the deformed sheet in 2 is measured using 3. Further, this data is fed to the data acquisition means 4 through 7 and further supplied to the processing means 5 wherein the ANN architecture as well as the weights are established to obtain a trained ANN which is used to arrive at material specification in steps of:
• mounting in a sheet deforming means 2 a punch, a blank holder and a die of dimensions corresponding to those of the smallest product (based on product features) in the family of products;
• placing sheet between the die and the punch a;
• deforming the said sheet to the desired shape;
• feeding user defined acceptance criteria (eg. failure strain) to the data processing means 5;
• measuring quantities constituting the acceptance criteria (like measuring maximum thickness strain using strain measuring means) and /or user defined process parameters and feeding the same to the data acquisition means 4, processing means 5;
• comparing the data with user defined quantities constituting the acceptance criteria;
• mounting in a sheet deforming means 2 a punch, a blank holder and a die of dimensions corresponding to those of the largest product (based on product features) in the family of products;
• placing between the die and the punch a sheet;
• deforming the said sheet to the desired shape;
• measuring maximum thickness strain (using strain measuring means 3) and /or user defined process parameters and feeding the same to the data acquisition, processing means;
• comparing the data with user defined quantities constituting the acceptance criteria;
• deforming at least two sheets each having different set of material properties for the smallest and the largest product using above mentioned process;
• measuring maximum thickness strain and feeding the same to the data acquisition means 4, processing means 5;
• comparing it with the acceptance criteria and denoting the product as safe or failed;
• Using the above established data stored in the data acquisition means 4 to develop the ANN architecture to the desired accuracy of prediction using the quantities corresponding to the product geometry, material properties, tool design and processing conditions;
• training the ANN using the said quantities in the said data processing means 5;
• in the data processing means 5, using the said trained ANN predict acceptability of product for a given set of material properties;
• if product is not acceptable; varying the material properties and/or the processing conditions in the data processing means using the said trained ANN so as to arrive at a set of material properties and processing conditions that will lead to an acceptable product.
• displaying the said set of material properties on display means 6;
• mounting in a sheet deforming means 2 a punch, a blank holder and a die of dimensions corresponding to those of the desired product (based on product features) belonging to the given family of products;
• placing between the die and the punch the sheet with the said material properties arrived at using the ANN;
• deforming the said sheet to the desired shape.
In one of the embodiments sheet is of any material such as polymer based, metal based which can deform
In other embodiment data acquisition, processing means and output device are integrated with the sheet deforming means.
In yet another embodiment press comprises of either or both of the die and punch made from solid, liquid or gaseous material;
The aim of the experiment is to illustrates manufacturing of a cylindrical cup (from a family of deep drawn cups as sheet metal products) using a sheet metal selection arrived at using the process of the present invention and further forming of the metal sheet in the system used in the present invention. Further, validation of the results of the present invention is presented in the tabular form. Here the minimum thickness of the cup after drawing is taken as the quality parameter.
• The following parameters including the tool design parameters, are kept
1. Die radius = 7 mm
2. Punch radius = 9.5 mm
3. Clearance = 1.25 mm
4. Punch Travel = 0-90 mm
• a sheet metal of mild steel is mounted in deforming means comprising a punch, a blank holder and a die of dimensions corresponding to those of the smallest deep drawn cup in the deep drawn family of products;
• the said sheet is placed between the die and the punch;
• the said sheet metal is deformed to the desired shape;
• the said sheet is deformed to the desired shape;
• user defined failure strain criteria is fed to the data processing means;
• maximum thickness strain, blank holding force and other tool design parameters are fed to the data acquisition, processing means;
• they are compared with user defined quantities constituting the acceptance criteria in the data processing means;
• similar procedure was carried out and data was collected for the largest deep drawn cup;
• sheets of materials such as galvanized iron and aluminum were deformed for the smallest and the largest product using above mentioned process;
• maximum thickness strain, blank holding force and other tool design parameters were measured and fed to the data acquisition, processing means;
• the above established data stored in the data acquisition means was used to develop ANN architecture in the data processing means;
• the weights in the ANN were established to obtain a trained ANN using the said data in the said data processing means;
• parameter such as thickness (strain distribution) was predicted / arrived at using data processing means and said trained ANN for a particular set of material properties;
• the said thickness was compared in the data processing means with the already fed acceptability criteria;
• the set of material properties was displayed on the display means;
• a deep drawn cup of the material properties was formed using deforming means comprising a punch, a blank holder and a die
The Input Parameters
1. Yield strength (MPa)
2. Strain hardening exponent
3. Strength coefficient (MPa)
4. Plastic strain ratio (R value)
5. Limiting drawing ratio'
6. Die radius (mm)
7. Punch nose radius (mm)
8. Clearance between die and punch (mm)
9. Blank holding force (KN)
10. Coefficient of friction
11. Thickness (mm)
12. Punch travel (mm)
Out of these parameters, few parameters are material properties, few are tool design parameters and remaining are process parameters.
The Output Parameters
The developed neural network has 3 outputs which are as follows,
1. Major Strain (^^)
2. Minor strain (e2 )
3. Check that whether this part can be deep drawn in a single stage or not (1 or 0)
If it can be drawn, output should be 0 or close to zero, as the non-zero value is interpreted as the probability of rejection.
Generation of the Data
To generate the data, a number of simulations for the cylindrical deep drawn cup with commercial sheet metal forming software 'PAM STAMP 2G' were carried out considering the various combinations of material, tool and processing variables emerging from the Design of Experiments methodology. The materials taken into consideration were mild steel, copper (a brass) & aluminum alloy. Their properties were found from literature available.
Characteristics of Materials Used for Simulations
Table 1: Material Specifications
Material E (GPa) K(MPa) n YS (MPa) R «
Mild Steel 206 350-650 0.16-0.24 60-160 0.6-2.0 0.33
Aluminum alloy 65 60I 0.32? 126 0.78 0.33
Copper 130 950 0.5 190 1.44 0.2S5
In this project the following parameters including the tool design parameters, are kept constant during all simulations,
1. Die radius = 7 mm
2. Punch radius = 9.5 mm
3. Clearance = 1.25 mm
4. Punch Travel = 0-90 mm
The following parameters including the process parameters are varied during simulations so that the neural network gets trained within these two extreme limits of these parameters.
Table 2: Range of Parameters
Sr. Mo. Parameter.!; Low Value High Value
I t (mm) 0.8 1.5
2 BHf (KN) 70 100
3 (D/d) ratio 2.1 2.3
4 « 0.05 0.12
Training of ANN
The ANN is trained using the data given in Table 3. Table 4 give the sample data used for training ANN.
Table 3 Input values for cylindrical cup simulations to train ANN
Simul ation t (mm) n k (MPa) R YS (MPa)
1 0.8 0.16 350 0.6 160
2 1.1 0.16 350 0.6 60
3 0.8 0.24 350 0.6 60
4 1.1 0.24 350 0.6 160
5 0.8 0.16 650 0.6 60
6 1.1 0.16 650 0.6 160
7 0.8 0.24 650 0.6 160
8 1.1 0.24 650 0.6 60
9 0.8 0.16 350 1.8 60
10 1.1 0.16 350 1.8 160
11 0.8 0.24 350 1.8 160
12 1.1 0.24 350 1.8 60
13 0.8 0.16 650 1.8 160
14 1.1 0.16 650 1.8 60
15 0.8 0.24 650 1.8 60
16 1.1 0.24 650 1.8 160
Table 4 Sample training data for ANN
Dat Major Minor Draw
a YS n K RDRRdRpC H BHF n t strain strain Depth
1 160 0.16 350 0.6 2.1 7 9.5 1.25 0 100 0.12 0.8 0 0 0
2 160 0.16 350 0.6 2.1 7 9.5 1.25 1 100 0.12 0.8 0.00041 0.00004 1
3 160 0.16 350 0.6 2.1 7 9.5 1.25 2 100 0.12 0.8 0.00133 0.00046 2
4 160 0.16 350 0.6 2.1 7 9.5 1.25 3 100 0.12 0.8 0.00325 0.00175 3
5 160 0.16 350 0.6 2.1 7 9.5 1.25 4 100 0.12 0.8 0.00638 0.00347 4
6 160 0.16 350 0.6 2.1 7 9.5 1.25 5 100 0.12 0.8 0.01105 0.00567 5
7 160 0.16 350 0.6 2.1 7 9.5 1.25 6 100 0.12 0.8 0.01547 0.00816 6
8 160 0.16 350 0.6 2.1 7 9.5 1.25 7 100 0.12 0.8 0.02195 0.01066 7
9 160 0.16 350 0.6 2.1 7 9.5 1.25 8 100 0.12 0.8 0.02878 0.01289 8
10 160 0.16 350 0.6 2.1 7 9.5 1.25 9 100 0.12 0.8 0.03781 0.01576 9
11 160 0.16 350 0.6 2.1 7 9.5 1.25 10 100 0.12 0.8 0.04916 0.01776 10
12 160 0.16 350 0.6 2.1 7 9.5 1.25 11 100 0.12 0.8 0.06261 0.02124 11
13 160 0.16 350 0.6 2.1 7 9.5 1.25 12 100 0.12 0.8 0.07649 0.02571 12
14 160 0.16 350 0.6 2.1 7 9.5 1.25 13 100 0.12 0.8 0.09439 0.03116 13
15 160 0.16 350 0.6 2.1 7 9.5 1.25 14 100 0.12 0.8 0.1164 0.03732 14
16 160 0.16 350 0.6 2.1 7 9.5 1.25 15 100 0.12 0.8 0.14612 0.04432 15
17 160 0.16 350 0.6 2.1 7 9.5 1.25 16 100 0.12 0.8 0.20005 0.04824 16
18 160 0.16 350 0.6 2.1 7 9.5 1.25 17 100 0.12 0.8 0.29415 0.05257 17
19 160 0.16 350 0.6 2.1 7 9.5 1.25 18 100 0.12 0.8 0.45474 0.05380 18
20 160 0.16 350 0.6 2.1 7 9.5 1.25 19 100 0.12 0.8 0.81201 0.05647 19
The trained ANN is used to obtain the specification for the cylindrical cup. Table 5 gives the user entered specifications in column 1 and ANN suggested specifications for the cylindrical cup in the column 2. The experiments were carried out and it was found that the ANN predicted specifications found to be correct. Table 6 gives the comparison of ANN output with experiment and FEM (finite element analysis). Figure 2 gives thickness vs. distance from centre of the cup obtained experimentally for cylindrical cup and figure 3 gives thickness vs. punch travel obtained using finite element simulation for cylindrical cup (Sheet 2). Figure 4 shows photograph of successfully formed cylindrical cup with ANN suggested material specification (Sheet 3).
Table 5 Specification obtained for cylindrical cup using ANN
ANN input (1) ANN suggested specifications (2)
Initial material specifications are : Yield strength = 200 Strain hardening index = 0.22 Strength coefficient = 350 Anisotropy index = 1 Sheet thickness = 0.8 Cup diameter = 65 Die radius = 7 punch radius = 5 Clearance = 1.25 blank holder force = 25 friction coefficient =0.12 Cup height = 35 FEM predicted thickness strain for Initial material specification = -0.12584 User defined criteria for thickness strain = -0.1 Final material specifications are : Yield strength > 170 Strain hardening index > 0.24 Strength coefficient > 550 Anisotropy index > 1.6 Sheet thickness > 0.8 Cup diameter = 65 Die radius = 7 punch radius = 5 Clearance = 1.25 blank holder force = 15 friction coefficient = 0.12 Blank diameter = 115 ANN predicted thickness strain = -0.0645
Table 6 Comparison of ANN output with experiment and FEM
ANN o/pMin. Thicknessin mm Expt. o/pMin. Thicknessin mm FEM o/pMin. Thicknessin mm
0.75 0.74 0.76
It was observed from the experiments that:
a. It was possible to form without failure (excessive thinning in this case),
cups from material and process variables suggested by the ANN as above.
A new material specification was suggested by the ANN when the current
one was found to be inadequate
b. The minimum thickness experimentally measured agrees with the
prediction of the trained ANN
c. The prediction of the maximum draw depth by the ANN agrees with that of
the PAMSTAMP simulations.
d. The prediction of the ANN on the status of the cup (failed or otherwise)
agrees with that from the simulations.
e. PAMSTAMP Simulations to verify the product formability using the new
material specification suggested by the ANN show that the material and
processing variables are acceptable. A new material specification is suggested by the ANN when the current one is found to be inadequate. The simulations also confirm the prediction of the inadequacy of the current specification..
Thus it is evident from the above experiment that the system and process of the present invention which effectively uses trained ANN results in catering to infinite number of product shapes (geometries) in the same family resulting in substantial reduction in the wastage of material, machine, labour and other resources, enabling quick introduction of a new product geometry belonging to the same family, enabling assessment of the effect of change in tool parameters and processing conditions on product quality, irrespective of material and the shape of the drawn product belonging to the family for which the ANN was trained.
INDIAN INSTITUTE OF TECHNOLOGY, BOMBAY Associate Dean, Research and Development
^jfr PtisfK* 3TPT 3TPT -gt Y*f fet Oirector. HT Bombay
|Indian Patent Application Number||808/MUM/2006|
|PG Journal Number||31/2010|
|Date of Filing||26-May-2006|
|Name of Patentee||INDIAN INSTITUTE OF TECHNOLOGY, BOMBAY|
|Applicant Address||INDIAN INSTITUTE OF TECHNOLOGY, POWAI, MUMBAI 400076.|
|PCT International Classification Number||G01N3/00|
|PCT International Application Number||N/A|
|PCT International Filing date|