#TITLE_ALTERNATIVE#

• Main Author: MARTYN (NIM : 23408060), ERVINA
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/21986
• Institution: Institut Teknologi Bandung

There are 3 factors of production process which contribute to the high cracking proportion within decorative pottery, which are : sun-dried process, glazing process, and burning process. The weakness of Taguchi Method which only gained the optimum setting level on the designated level will be overcome by using Response Surface Methodology (RSM) and Back Propagation Network (BPN). The using of both methods also made the study of cracking proportion within decorative pottery focus on the development of Taguchi-RSM and Taguchi-BPN method. Main objective from the study of cracking proportion on pottery were : (i) found the most contributed factors to high proportion of cracking on pottery; (ii) decided factor level setting which produced minimum cracking proportion, by using Taguchi-RSM Method and Taguchi- BPN Method; and also developed updating process of weighted-BPN by using Taguchi-BPN Method which implemented on a daily routine data for a continuous improvement of production process and products. Methodology using for this study covered the examination of Taguchi Method, examination of RSM with <br /> <br /> <br /> <br /> <br /> <br /> <br /> software SAS system for Windows ver. 9.0, and examination of BPN using software Matlab 7.5.0-R-2007b. From the study result, development of Taguchi Method-RSM and Taguchi-BPN Method were also performed. Taguchi-RSM and Taguchi-BPN Method compared based on error value of output value differentiation and the least predicted value. Taguchi-BPN Method given the smallest error value. There are 2 development of BPN architectural, which are BPN 1 and BPN 2. BPN 1 using the production process factor level for data input (p) and cracking proportion for target input (t), which guide the searching of optimum factor level by setting the value for each level of controlled factor. BPN 2 using the cracking proportion for input data (p) dan production process factor level for target input (t), which guide the searching of optimum factor level by setting the cracking proportion value as per user’s request. Both of <br /> <br /> <br /> <br /> <br /> <br /> <br /> BPN 1 and BPN 2 architecture were design for the continuous improvement of production quality process and decorative pottery products. Study results showed the optimum production process factor level from BPN 1 are : Factor A (sun-dried period) of 4 days, Factor B (burning temperature) of 1125°C, and Factor C (burning period) of 8,5 hours, and SNR value -14,9788. Otherwise, optimum production process factor level from BPN 2 are : Factor A (sun-dried period) of 4.5 days, Factor B (burning temperature) of 1150°C, and Factor C (burning period) of 8,5 hours, and SNR value -5.86. Aferward, the optimum production process factor level from New BPN 1 are : Factor A (sun-dried period) of 4 days, Factor B (burning <br /> <br /> <br /> <br /> <br /> <br /> <br /> temperature) of 1120°C, and Factor C (burning period) of 8,75 hours, and SNR value -7.79. Otherwise, optimum production process factor level from New BPN 2 are : Factor A (sun-dried period) of 4.9 days, Factor B (burning temperature) of 1140.1°C, and Factor C (burning period) of 8,38 hours, and SNR value 0. The generation of BPN 1 and BPN 2 architecture also defined the general rule BPN 1 and BPN 2. General rule of BPN 1 generation are the amount of neuron within the layer input are the sum of production process factor which included on the experiment, the amount of neuron output is the total of replication within quality characteristic respons which studied, the amount of neuron on hidden layer are bigger than total replication of quality characteristic which performed, learning rate are 0.05, logsig transfer function for hidden layer and purelin transfer function for output layer,trained logarithm using was train LM. General rule of BPN 2 generation are the amount of neuron within the layer input are the total replication of quality characteristic respons which need to define on the experiment, the amount of neuron output is the total of production process factor which involved on the experiment, the amount of neuron on hidden layer are bigger than total replication of quality characteristic which performed, learning rate are 0.05, logsig transfer function for hidden layer and purelin transfer function for output layer, trained logarithm using was train LM. Contribution of this study is the ability to find the otimum condition for every level factor which involved on the experiment. The conclution of this study is the development of Taguchi- BPN method can be use for the improvement of factor level from a production process to gain the continuous quality product improvement.