Machine Learning in Failure Analysis of Optical Transceiver Manufacturing Process

Abstract

Failure Analysis is a systematic process of collect and analyze data to determine the cause of failure, and identify effective corrective actions. It is an important discipline in many manufacturing industry, and when apply correctly can help to save money, resources and prevent further damages. Common failure analysis techniques in manufacturing include Ishikawa Cause-and Effect analysis a.k.a. Fishbone Analysis and Failure Mode and Effects Analysis (FMEA) are also used in optical transceiver manufacturing. While both methods are effective in providing high level assessment of failure causes, they may be visually cluttering when more complex defects are involved and the interrelationships between causes are not easily identifiable. This paper examines the application of Supervised Machine Learning in defect detection, quality assurance and throughput improvement. Machine learning helps manufacturers visualize previously impenetrable problems and reveal those that they never knew existed, including hidden bottlenecks or unprofitable production lines [1], further enhancing Failure Analysis methods in manufacturing.