Wireless Power Transfer System Testing: Optimization & Quality Assessment Using ML

Sammanfattning

Wireless Power Transfer (WPT) is increasingly utilized across various practical domains, requiring continual development and optimization of supporting technologies. An essential aspect of this progress is the efficient testing of WPT Systems. This thesis explores optimization of the testing process using Machine Learning(ML) and data analysis, applied on historical data represented by test outcomes of different products, known as test reports. The research focuses on three primary objectives: product fault categorization, test-quality analysis and test failure prioritization. Product fault categorization employs clustering methods to find common groups of failed tests, in the hope it captures underlying product fault. Test-quality analysis involves examining measured value trends to develop a more nuanced performance metrics and the conversion of the values to useful features for the test failure prioritization. Test failure prioritization employs predictive models and optimal static orders to expedite the running of tests which will fail during product testing. Critical components in implementing the techniques involve data preparation, feature extraction and innovative new adaptions of existing algorithms to accommodate limitation in data quantity. The results demonstrate significant improvements in testing efficiency, with theaverage number of tests conducted before detecting a failure decreasing from 51.6 to 8 through the application of ML on historical data. Additionally, the locality of failures across a 0-100 scoring system improved from a average 37.8 to 79.5 when using optimal static ordering. Although clustering and test-quality analysis yielded promising results with discernible value trends, additional validation is necessary to establish their full utility.