Failure prediction system and failure prediction method

Abstract

The present invention performs machine learning wherein: the period between a reference time point, which corresponds to a failure prediction time point, and an earlier time point, which is a predetermined first length of time earlier than the reference time point, is set as a warning sign detection period; if no failure occurred during the period between the reference time point and a later timepoint, which is a predetermined second length of time later than the reference time point, then that period is set as a prediction period, and otherwise if a failure occurred during the period betweenthe reference time point and the later time point, then the period between the reference time point and the time point at which the failure occurred is set as the prediction period; the reference time point is then sequentially shifted; a feature quantity based on operation log information obtained during the warning sign detection period is set as an explanatory variable; and a failure index value is set as a response variable, the failure index value being based on whether or not occurrence of a specific event is indicated by operation log information associated with failure occurrence indicated by failure record information obtained during the prediction period, and also based on the length of time between the reference time point and the time point at which the failure occurred.

Description

technical field [0001] The present invention relates to a system for diagnosing whether there is a symptom of a failure in a device and predicting the failure of the device. Background technique [0002] When installations and equipment stop due to breakdowns, maintenance costs increase and customer satisfaction suffers. Therefore, preventive maintenance of devices and equipment is required. Preventive maintenance is to record the operation history and operating status of equipment and equipment, and perform maintenance such as parts replacement based on this information before failure occurs, so as to prevent the entire equipment and equipment from stopping in advance. As a system for predicting future failures based on the operation history of the device, it is proposed to extract a certain period of time before the failure from the operation log and associate it with the failure, and use the associated operation log and failure information as input to perform machine A ...

AI Technical Summary

Problems solved by technology

When learning and fault prediction are performed using operation logs extracted at a single time difference before a fault occurs, the operation log does not necessarily include symptoms of faults, which will deteriorate prediction accuracy

Images