Abstract
In today’s highly competitive global market, winning requires near-perfect quality. Although most mature organizations operate their processes at very low defects per million opportunities, customers expect completely defect-free products. Therefore, the prompt detection of rare quality events has become an issue of paramount importance and an opportunity for manufacturing companies to move quality standards forward. This article presents the learning process and pattern recognition strategy for a knowledge-based intelligent supervisory system, in which the main goal is the detection of rare quality events. Defect detection is formulated as a binary classification problem. The l1-regularized logistic regression is used as the learning algorithm for the classification task and to select the features that contain the most relevant information about the quality of the process. The proposed strategy is supported by the novelty of a hybrid feature elimination algorithm and optimal classification threshold search algorithm. According to experimental results, 100% of defects can be detected effectively.
Highlights
In today’s highly competitive global market, winning requires near-perfect quality, since intense competition has led organizations to low profit margins
For a data-rich analysis, the hold-out validation method is recommended, an approach in which a dataset is randomly divided into three subsets: training, validation, and testing
The l1-regularized logistic regression (LR) algorithm enjoys the following desirable properties: (1) It induces parsimony while maintaining convexity;[41] (2) it is founded on the likelihood principle, maximum likelihood provides a consistent approach to parameter estimation problems and has desirable mathematical and optimality properties;[42] (3) according to large sample theory, as the sample size tends to infinity, the sampling distribution of the maximum likelihood estimate (MLE) becomes Gaussian;[29] and (4) since many candidate models are created to approximate the true model, well-known likelihood-based model selection criterion (Akaike information criterion (AIC) or Bayesian information criterion (BIC)) can be applied to solve the challenge posed by over-fitting due to model complexity
Summary
In today’s highly competitive global market, winning requires near-perfect quality, since intense competition has led organizations to low profit margins. Feature interpretation is out of the scope of this approach, analyzing the data-derived predictive model from a physics perspective may support engineers in systematically discovering hidden patterns and unknown correlations that may guide them to identify root causes and solve quality problems. The 0.5 threshold may not be the best classification threshold, and accuracy[25] may be a misleading indicator of classification performance To address this scenario, the concept of maximum probability of correct decision (MPCD) is used as a measure of generalization performance.[38,39] A model selection criterion tends to be very sensitive to FNs—failure to detect a quality event—in highly unbalanced data. This stage ensures that the model satisfies the learning target for the project at hand
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
