Room: Karl Dean Ballroom A1
Purpose: From 2015 to 2017, four TrueBeams (TBs) at our main campus had 6.4 hours/year/TB of unexpected MLC downtime, or 15.7 failures/year/TB. We propose a proactive maintenance model using historical MLC performance data to predict MLC dysfunctions and promote preemptive maintenance and thereby reduce downtime.
Methods: MLC failures (primary/secondary, trajectory and standstill deviations) are assumed to correlate with MLC performance quantified from trajectory logs. A cohort of data from service reports and trajectory logs is used to establish a correlation model for predicting MLC dysfunctions. Specifically, service reports logged by our on-site engineers record MLC status of each failure, including service date, reason (types of deviation on a leaf), and actions taken (MLC initialization or motor/T-nut replacement), while trajectory logs record the ordered/actual leaf positions in 50Hz. In quantifying leaf performance, an event is defined as detecting a leafâ€™s position deviation â‰¥ 0.1mm from a trajectory log. The distributions of the accumulative events as functions of sorted days and individual leaves are associated with logged failures to find a correlation for predicting dysfunctions. In evaluation, a model containing TB1â€™s 124-week data, including 48 failures and 29,806 trajectory logs, is built and applied to TB2 for validation.
Results: According to TB1â€™s correlation model, if a leaf scores â‰¥ 20 events/day in any four days during a 10-day moving window, the leaf should mark as â€œpotential failureâ€?. Using this criterion, our program predicts 13 failures in validating TB2â€™s 2016 records: 7 are true positives, 6 are false alarms, but missing 6 actual failures. Proactively addressing the predicted failures would reduce TB2â€™s 2016 downtime from 6 to 2.8 hours with a cost of 2.8-hour service time for the false alarms.
Conclusion: The preemptive maintenance model based on trajectory logs can capture half of the actual failures, significantly reduce downtime and smooth clinical workflow.