📄 中文摘要
准确的仪器校准对于确保测量的可追溯性、可靠性和合规性至关重要。固定间隔的校准程序虽然易于管理，但忽视了仪器在不同条件下以不同速率漂移的事实。研究将校准调度视为预测性维护问题：根据最近的传感器历史数据，估计漂移时间（TTD），并在违规发生之前进行干预。通过选择对漂移敏感的传感器、定义虚拟校准阈值以及插入模拟重复校准的合成重置事件，将NASA C-MAPSS基准适配到校准设置中。随后，比较了经典回归模型、递归和卷积序列模型的性能，以评估不同方法在校准调度中的有效性。
基于变换器的风险感知仪器校准预测性维护
📄 English Summary
Transformer-Based Predictive Maintenance for Risk-Aware Instrument Calibration
Accurate calibration is essential for instruments that require traceable, reliable, and compliant measurements over extended operating periods. Fixed-interval calibration programs are easy to administer but overlook the fact that instruments drift at varying rates under different conditions. This research frames calibration scheduling as a predictive maintenance problem: estimating time-to-drift (TTD) based on recent sensor histories and intervening before a violation occurs. The NASA C-MAPSS benchmark is adapted into a calibration context by selecting drift-sensitive sensors, defining virtual calibration thresholds, and inserting synthetic reset events that simulate repeated recalibration. The performance of classical regressors, recurrent, and convolutional sequence models is then compared to evaluate the effectiveness of different approaches in calibration scheduling.