Failure Prediction Method for Long Life Photoconductor Based on Statistical Machine Learning

Abstract

In the production printing market, the electrophotography system is asked for high durability, and long-life photoconductors are usually implemented in it. Ordinary method of periodic maintenance uses print volume counter and preset limitation value to alert an operator. But damage processes depend on various phenomena and there are difficulties to predict the exact lifespan of each photoconductor. We examined the methodology for developing new risk decision rule by using the logged data of the field machines. In this method, we focused on photoconductor's three basic parameters and used one of the machine learning method “AdaBoost” to find failure signal pattern. These basic parameters were pre-calculated to the physical and statistical characteristic value to estimate abnormality of these value. We selected learning data from the typical fatigue log data and “AdaBoost” algorithm generated the risk decision rule that consist with the weak learners. The field tests showed enough result for practical use. Additionally it was confirmed that this risk decision rule was almost coincident with the photoconductor deterioration model knowledge by “Score plots”. We obtained the calculation method to determine the risk of image defect from monitoring signal data log and it can be predicted whether photoconductor should be replaced or not.This failure prediction method can reduce urgent imaging trouble and the loss of photoconductor's lifespan that occurred in the ordinary method.