Control chart patterns recognition using run rules and fuzzy classifiers considering limited data

• Main Author: Zaman, Munawar
• Format: Thesis
• Language: English
• Published: 2017
• Subjects: TJ Mechanical engineering and machinery
• Online Access: http://eprints.utm.my/id/eprint/78884/1/MunawarZamanMFKM2017.pdf; http://eprints.utm.my/id/eprint/78884/; http://dms.library.utm.my:8080/vital/access/manager/Repository/vital:108996

Statistical process control chart is a common tool used for monitoring and detecting process variations. The process data streams, when graphically plotted on control chart reveal useful patterns. These patterns can be associated with possible assignable causes if properly recognized. These patterns detections are useful for process diagnostic. Different types of control chart pattern recognition methods are reported in literature. Most of the existing data-driven methods require a large amount for training data before putting into practice. Short production run and short product life cycle processes are usually constrained with limited data availability. Thus there is a need to investigate and develop an effective control chart pattern recogniser (CCPR) methods for process monitoring with limited data. Two methods were investigated in this study to recognize fully developed control chart patterns for process with limited data on X-bar chart. The first method was combination of selected run rules, as run rules do not require training data. Classifiers based on fuzzy set theory were the second method. The performance of these methods was evaluated based on percent correct recognition. The methods proposed in this study significantly reduced the requirements of training data. Different combination of Nelson’s run rules; R2,R5,R6 for shift and trend, R3,R5,R6 for cyclic, R4,R5,R8 for systematic and R7 for stratification patterns were found effective for recognizing. Differentiating between the shift and trend patterns remains challenging task for the run rules. Heuristic based Mamdani fuzzy classifier with fuzzy set simplification operations using statistical features gave more than ninety percent correct patterns recognition results. Adaptive neuro fuzzy inference system (ANFIS) fuzzy classifier with fuzzy c-mean using statistical features gave more prominent results. The findings suggest that the proposed methods can be used in short production run and the process with limited data. The fuzzy classifiers can be further studied for different input representation.