Abstract
In electronics manufacturing, the necessary quality of electronic components and parts is ensured through qualification testing using standards and user requirements. The challenge is that product qualification testing is time-consuming and comes at a substantial cost. The work contributes to develop a novel prognostics framework for predicting qualification test outcomes of electronic components enabling the reduction of qualification test time and cost. The research focuses on the development of a new, prognostics-based approach to qualification of electronics parts that can enable “smart testing” using data-driven modelling techniques in order to ensure product robustness and reliability in operation. This work is both novel and original because at present such approach to qualification testing and the associated capability for test time reduction (respectively cost reduction) it offers are non-existent in the electronics industry. An effective way of using three different methods for development of prognostics models are identified and applied. Predictive models are constructed from historical qualification test data in the form of electrical parameter measurements using Machine Learning (ML) techniques. ML models can be imbedded within the sequential electrical tests qualification procedure and enable the forecasting of the pass/fail qualification outcome using only partial information from already completed electrical tests. Data-driven prognostics models are developed using the following machine learning techniques: (1) Support Vector Machine (SVM), (2) Neural Network (NN) and (3) K-Nearest Neighbor (KNN). The results show that with just over half of the individual tests completed, the models are capable of forecasting the final qualification outcome, pass or fail, with accuracy as high as 92.5%. The predictive power and overall performance of the researched models in predicting qualification test binary outcomes with varying ratios of Pass and Fail data in the processed datasets are analysed.
Highlights
The global market for electronic products is anticipated to reach US$2.4 trillion per year by 2020 [1]
FRAMEWORK FOR PREDICTING QUALIFICATION TEST OUTCOME The type of qualification testing considered in this research work is the most common electrical parameter testing where the qualification involves undertaking a sequence of discrete tests associated parameter measurements of the electronic part/product
COMPARATIVE ANALYSIS AND DISCUSSIONS When all the results are plotted for each training set and algorithms (Fig. 11), it is very clear that overall, Support Vector Machine (SVM) provides the best prediction accuracy (92.50%) and K-Nearest Neighbor (KNN) provides the second best prediction accuracy (80%) for the smallest size (1164) of the training set where the ratio of fail to pass data is 1.0
Summary
The global market for electronic products is anticipated to reach US$2.4 trillion per year by 2020 [1]. While most of the research and applications of machine learning and computational intelligence techniques relate to the process monitoring and control of electronics and diagnostics/prognostics under failure test or in-field operational loads, the use of such technologies to improve or optimise qualification testing of electronics parts is yet to be realised and demonstrated. The aim of this study is to develop a machine learning based novel computational approach for predicting qualification test outcomes of an electronic module using historical test data. This approach would help to reduce test time and huge cost associated with the qualification testing. Prediction accuracies from each algorithm are assessed and a performance comparison of the algorithms is presented in this paper
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
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 call
