Screening Out Potentially Defective Products in Micro-Transformer Production by Intelligently Integrating Mechanical and Electronic Signals

Along with unstable mechanical status, the critical uncertainty of micro-transformer at electronic specification boundary leads to potentially defective products (PDPs) in qualified products. Hence, an intelligent prediction of PDPs is proposed by integrating mechanical and electronic signals and tracing defective product information. The objective is to screen out the PDPs by the relationship between critical state of normal test equipment and the defective ex-factory products in actual production. On the base of time and frequency domain eigenvalues, Artificial Neural Network (ANN) and Naïve Bayes (NB) were firstly employed to process the vibration signal of equipment testing probe for network training and prior probability, respectively. Then, the various electronic signals related to quality performance parameters were categorized to screen out the PDPs by K-Nearest Neighbor (KNN) and Support Vector Machine (SVM) along with spatial distance and hyperplane databases, respectively. Finally, the screen-out threshold was introduced to evaluate the reasonableness with reference to the phase-out divergence. It is shown that the NB more precisely recognizes the equipment status by timely updating small and incomplete samples than the ANN. Compared to the SVM, the KNN validly screens out the PDPs due to its superior processing capability of electronic signals. As a result, the screen-out threshold is set as 5–6% to balance the screen-out number and precision.