Analyzing Board Health with Machine Learning

Lovering, Joshua; Oshiro-Leavitt, Clay N.; Tang, Bryson M.; Frith, Suverino L.

Student Work

Analyzing Board Health with Machine Learning

Public

The goal of this project was to determine the reason for unexpected test failures through analysis of Dell EMC motherboard testing log data. We began with an extensive exploration of our extracted features, through a series of statistical analysis and plotting. Our solution consisted of three components: a warning system to detect trend changes in feature values, a machine learning model to predict failure tests, and the standardization of log data to increase efficiency of future analysis. The warning system analyzes a group of test logs using time series and regression analysis and outlier detection to calculate the date at which future tests would reach a mean value equal to an outlier value of the current tests. The machine learning model was trained on 548 features from a set of 826 passing and 511 failing boards to classify unlabeled data, and received a testing accuracy of 63 percent. Lastly, we contributed a standardized log format proposal based upon JSON which would increase the value of their log data. This was determined after the extensive time required for parsing the different formats across test steps during feature extraction.

This report represents the work of one or more WPI undergraduate students submitted to the faculty as evidence of completion of a degree requirement. WPI routinely publishes these reports on its website without editorial or peer review.

Creator

Subject

Publisher