Integrating fractal dimension and neural network for detecting unnatural patterns on process mean | |
| Author | Ussanee Purintrapiban |
| Call Number | AIT DISS. no. ISE-03-03 |
| Subject(s) | Neural networks (Computer science) Fractals Manufacturing processes |
| Note | A dissertation submitted in pa1tial fulfillment of the requirements for the degree of Doctor of Engineering, School of Advanced Technologies |
| Publisher | Asian Institute of Technology |
| Series Statement | Dissertation ; no. ISE-03-03 |
| Abstract | This study concentrated on the development of tools to identify unnatural patterns in process data. The first tool is based on fractal dimension. The second is developed by integrating fractal dimension with neural network for detection of cyclical pattern. The last one is an integrated model developed specifically for detecting unnatural patterns. Fractal dimension is an index used to measure the complexity of an object. The first proposed tool exploits this feature in that fractal dimension is used as a supplement to traditional control cha1ts to detect patterns in process data. The results obtained via Monte Carlo simulation demonstrated that the fractal dimension based method is effective to detect non-periodic patterns when the maximum magnitude of random noise r in terms of standard deviationa is not greater than 0.20. Fmthermore, it can also be used as a tool to estimate the period value of cyclical data especially when the signal-tonoise ratio is greater than 2.50. The second tool is proposed specifically for detecting cycle pattern on process mean. The main idea is to calculate the fractal dimension of the data before feeding it as the input to the neural network system. The output of the system is a particular period of cycle or the natural pattern. The period value of cycle at the various amplitudes can be determined quickly and accurately when the signal-to-noise ratio is greater than 3.00. This approach is suitable for automated manufacturing environment as a supplementary tool of traditional control charts. To improve on the flexibility of detection, a second neural network model is developed for detecting the unnatural patterns on process mean. This integrating model has been designed to be quick and accurate for the monitoring of automated manufacturing process with little or no human intervention. The model consists of two main pa1ts. The first part is the neural network with fractal dimension as its filter. This part of the system is trained to detect three patterns, i.e., systematic, stratification, and cyclic patterns. If the result from the first part of the system is not one of the three targeted patterns, the original process data is then pre-processed by the multi-median method before feeding into a second neural network for detecting the upward and downward linear trends, upward and downward shifts, and mixture. The simulation results show that this model can detect the patterns of interest quite well within the first window size of process data especially when the ratio of signal-to-noise is greater than 0.25 for linear trends and is greater than 2.50 for other patterns. In addition, this approach can provide the value of the period of cycle when the cyclical pattern existed in the data. Compared with the fractal dimension approach, the integration-based model is very effective and robust for detecting the unnatural patterns and for estimating the value of period of data with cyclical behavior. However, the fractal-based approach is simpler to use. All three approaches proposed in this study can provide type of patterns commonly occurred in manufacturing processes. This useful information would assist users in identification and removal of assignable causes in manufacturing processes. Due to this feature of the proposed approaches, they outperform the traditional control chaits that can only signal either in control or out-of-control process. The systematic use of these tools is an excellent way to reduce variability in the manufacturing process that may lead to opportunities for process improvement. |
| Year | 2003 |
| Corresponding Series Added Entry | Asian Institute of Technology. Dissertation ; no. ISE-03-03 |
| Type | Dissertation |
| School | School of Advanced Technologies (SAT) |
| Department | Department of Industrial Systems Engineering (DISE) |
| Academic Program/FoS | Industrial Systems Engineering (ISE) |
| Chairperson(s) | Voratas Kachitvichyanukul; |
| Examination Committee(s) | Huynh Ngoc Phien;Huynh Trung Luong; |
| Scholarship Donor(s) | King Mongkut's University of Technology Thonburi; |
| Degree | Thesis (Ph.D.) - Asian Institute of Technology, 2003 |