A model for preemptive maintenance of medical linear accelerators-predictive maintenance

Area of Science:

• Medical Physics
• Engineering
• Data Science

Background:

• Unscheduled downtime in cancer treatment accelerators significantly impacts patient quality of life.
• Current accelerator data is underutilized for proactive service engineering.
• Predictive maintenance can optimize resource allocation and minimize disruptions.

Purpose of the Study:

• To develop an effective process for detecting unexpected deviations in accelerator systems.
• To predict component failure or system dysfunction before interlocks are triggered.
• To enable proactive maintenance and reduce unscheduled downtime.

Main Methods:

• A predictive maintenance (PdM) model utilizing daily quality assurance (QA) treatments and log file analysis.
• Statistical Process Control (SPC) analysis of operating and performance values against established baselines.
• A hybrid chart limit technique combining standard SPC 3σ limits with empirical factors.

Main Results:

• The PdM model monitors 7 accelerators, analyzing 45 parameters and 120 MLC leaves using I/MR charts.
• A defined warning and alarm rule based on consecutive values relative to hybrid 3σ limits.
• 95.6% of introduced synthetic errors were successfully detected by the SPC charts.

Conclusions:

• The developed PdM model demonstrates potential for reducing unscheduled accelerator downtime.
• Long-term monitoring is necessary to fully establish the model's effectiveness.
• Proactive maintenance strategies can improve patient care and operational efficiency.