A recent study titled "Variable-length multiobjective optimization of inspection schedules for condition-based maintenance of vertical-cavity surface-emitting lasers under imperfect detection" has been published in Reliability Engineering and System Safety. The study was conducted by researchers from the Changchun Institute of Optics, Fine Mechanics and Physics of the Chinese Academy of Sciences. The framework jointly selects inspection count and timing and the degradation thresholds for servicing or replacement. 

Laboratory tests showed that allowing schedule length to vary improved the hypervolume metric by 22.8 percent over the best fixed-length schedule, indicating a broader set of useful trade-offs between service life, cost, and inspection burden. The optimized policies also remained distinct under parameter uncertainty, giving operators choices that favor simpler schedules, longer average life, or greater protection against early failures.

Vertical-cavity surface-emitting lasers are widely used in short-reach optical interconnects. Their output power declines gradually under sustained operation, and missed deterioration can allow a device to cross a soft-failure threshold before maintenance. Routine screening is relatively inexpensive but can miss degradation or produce false alarms. A confirmatory diagnostic provides more detailed condition information but adds time and cost. Most scheduling methods fix the number of inspections before optimization and treat degradation, detection quality, and parameter uncertainty separately. That setup can add visits where they yield little information or miss the periods when many devices are approaching an action threshold.

To connect these decisions, the researchers first converted optical-power measurements into normalized power loss. They fitted power-law degradation models to data from accelerated aging tests and also checked Gamma and Wiener process alternatives. A staged inspection model then linked a low-cost screening result to a confirmatory diagnostic through a specified detection-probability curve. Monte Carlo simulation propagated device-to-device degradation differences, missed detections, false alarms, and maintenance effects. The variable-length evolutionary optimizer searched the inspection count, inspection times, and condition thresholds simultaneously. It evaluated each candidate against expected cumulative cost, expected service life, and inspection burden instead of forcing every policy to use the same schedule length.

In the reference laboratory case, variable-length optimization produced a wider and better-performing set of solutions than the fixed-length alternatives. The improvement did not come from maximizing inspection frequency. The optimizer retained simple schedules when lower burden was preferred and added visits only when the projected extension in service life justified their cost. This created a continuous set of choices between lean inspection plans and more intensive schedules that sought longer operation. The result is useful for maintenance planning because it allows decision makers to select a policy according to cost tolerance and reliability priorities rather than accept a single preset inspection count.

Parameter uncertainty changed which schedules were preferred. One policy minimized inspection complexity, another emphasized average service life, and a third prioritized lower-tail performance to protect weaker devices. The team also compared its logistic detection model with public hit-or-miss records, providing an external check on calibration. These tests show that the framework can support different maintenance priorities without treating one schedule as universally best.

Professor Loon Ching Tang, a Fellow of the Academy of Engineering Singapore and a professor at the National University of Singapore, said the study addresses practical challenges in detecting degradation and planning maintenance for vertical-cavity surface-emitting lasers. By incorporating device-specific degradation and detection behavior into reliability research, the work offers a useful direction for integrating reliability engineering with optoelectronics.

The study used a small accelerated-aging sample, and the laser-specific screening curve was a reference assumption rather than a field calibration. The reported policies therefore remain conditional on the degradation and maintenance-effect models. Future work will add field-calibrated detection data, measured maintenance effects, and online updates so schedules can adapt as operational evidence accumulates.