Predictions of the life for micro multi-punch die using neural network

Abstract

The design of packaging bags for passive components is becoming increasingly difficult because these components are being miniaturized. In this study, packaging for passive components was manufactured by a punching process. A single 12-punch die set was utilized experimentally with the aim of improving production efficiency. The neural network was applied to establish a model based on the relationship between a variety of factors: position number of micro multi-punch die set, total punched lengths, clearance, and wear. A simulated annealing (SA) optimization algorithm with a performance index was applied to the neural network. This determined the optimal punching parameters and resulted in a satisfactory outcome. Engineers can utilize this technique in practice to find optimal processing parameters, thus avoiding poor design of punching die and the need for lengthy repairs. The neural network can predict individual wear of each punch (or die cavity) under any clearance at a specific punch-length. Accurate prediction of punch/die wear means that worn components may be replaced in a timely manner, avoiding the fabrication of poor-quality products. Engineers can utilize the results on the factory floor to find optimal clearance conditions. This avoids poor design of punching die and the need for lengthy repairs.

Appendix: The abductive network of punch AO-side

(N normalizer node, W white Node, S single node, T triple node, PSE = 4.2958 × 10⁻⁷)

$$ N_{1} = - 2.21 + 2.77*10^{ - 5} *\left( {{\text{punched}}\;{\text{length}},\;32,000\sim 128,000} \right); $$

$$ N_{2} = - 1.86 + 0.287*({\text{punch}}\;{\text{number}},1\sim 12); $$

$$ N_{3} = - 3.16 + 322*\left( {{\text{clearance}},0.06\sim 0.016} \right); $$

$$ \begin{aligned} D_{7} & = - 0.0226 + 0.513*N_{1} - 0.179*N_{2} + 0.0621*N_{1}^{2} - 0.039*N_{2}^{2} \\ & \quad + 0.0282*N_{1} *N_{2} + 0.257*N_{1}^{3} - 0.0116*N_{2}^{3} ; \\ \end{aligned} $$

$$ \begin{aligned} D_{8} & = - 0.326 - 0.111*N_{2} - 0.457*N_{3} - 0.116*N_{2}^{2} - 0.243*N_{3}^{2} \\ & \quad + 0.0 9 4 8*N_{2} *N_{3} + 0.0304*N_{2}^{3} + 0.212*N_{3}^{ 3} ; \\ \end{aligned} $$

$$ S_{9} = 0.304 - 0.483*N_{3} - 0.328*N_{3}^{2} + 0.255*N_{3}^{3} ; $$

$$ \begin{aligned} T_{6} & = 0.204 + 1.09*D_{7} + 0.664*D_{8} + 0.57*S_{9} - 0.00367*D_{7}^{2} \\ & \quad - 3.42*D_{8}^{2} - 12.6*S_{9}^{2} - 0.00488*D_{7} *D_{8} + 0.129*D_{7} *S_{9} \\ & \quad + 13.5*D_{8} *S_{9} + 0.42*D_{7} *D_{8} *S_{9} - 0.0669*D_{7}^{3} \\ & \quad - 3.41*D_{8}^{3} + 2.5*S_{9}^{3} ; \\ \end{aligned} $$

$$ S_{10} = 0.0382 - 0.179*N_{2} - 0.039*N_{2}^{2} - 0.0116*N_{2}^{3} ; $$

$$ \begin{aligned} T_{5} & = - 3.87 + 0.817*T_{6} + 132*S_{10} + 22.6*N_{2} + 0.00712*T_{6}^{2} \\ & \quad - 887*S_{10}^{2} - 18.8*N_{2}^{2} + 6.12*T_{6} *S_{10} + 1.21*T_{6} *N_{2} \\ & \quad - 289*S_{10} *N_{2} - 1.1*T_{6} *S_{10} *N_{2} - 0.0119*T_{6}^{3} - 393*S_{10}^{3} ; \\ \end{aligned} $$

$$ U_{4} = {\text{OUTPUT}} = 0.0168 + 0.00685*T_{5} $$